{"id":227133,"date":"2026-07-25T17:16:19","date_gmt":"2026-07-25T17:16:19","guid":{"rendered":"https:\/\/www.9senses.ai\/?page_id=227133"},"modified":"2026-08-27T12:24:07","modified_gmt":"2026-08-27T12:24:07","slug":"what-is-ai","status":"publish","type":"page","link":"https:\/\/www.9senses.ai\/fr\/what-is-ai\/","title":{"rendered":"What AI really is&#8230;"},"content":{"rendered":"<div class=\"et_pb_section_0 et_pb_section et_section_regular et_block_section\">\n<div class=\"et_pb_row_0 et_pb_row et_block_row ns-hdr\">\n<div class=\"et_pb_column_0 et_pb_column et_pb_column_4_4 et-last-child et_block_column et_pb_css_mix_blend_mode_passthrough\">\n<div class=\"et_pb_text_0 et_pb_text et_pb_bg_layout_light et_pb_module et_flex_module ns-eyebrow\"><div class=\"et_pb_text_inner\"><p>9senses on artificial intelligence<\/p>\n<\/div><\/div>\n<\/div>\n<\/div>\n\n<div class=\"et_pb_row_1 et_pb_row et_flex_row ns-hdr\">\n<div class=\"et_pb_column_1 et_pb_column et_block_column et_pb_css_mix_blend_mode_passthrough et_flex_column_24_24 et_flex_column_24_24_tabletWide\">\n<div class=\"et_pb_post_title_0 et_pb_post_title et_pb_bg_layout_light et_pb_module et_block_module\"><div class=\"et_pb_title_container\"><h1 class=\"entry-title\">What AI really is\u2026<\/h1><\/div><\/div>\n\n<div class=\"et_pb_text_1 et_pb_text et_pb_bg_layout_light et_pb_module et_block_module\"><div class=\"et_pb_text_inner\"><p>When most of us at 9senses began working with what is now labeled <strong>Artificial Intelligence,<\/strong> we didn't use that term. Back then, we were talking about non-linear computing, fuzzy logic, heuristics, machine learning, among others.<\/p>\n<p>Today, many people think that AI makes computers as smart as humans. In reality, computer software is still far away from reaching that level, but today it is able to <a href=\"\/why-9senses\">emulate and even surpass human capabilities in specific fields<\/a>, particularly those that require the processing of large amounts of information or the generation of output from a large data pool. We would like to instill a bit of clarity here, at the cost of taking some of the magic of AI away, as did Joseph Weizenbaum, the legendary creator of <a href=\"#eliza\">Eliza:<\/a><\/p>\n<\/div><\/div>\n<\/div>\n\n<div class=\"et_pb_column_2 et_pb_column et-last-child et_block_column et_pb_css_mix_blend_mode_passthrough et_flex_column_24_24 et_flex_column_24_24_tabletWide\">\n<div class=\"et_pb_code_0 et_pb_code et_pb_module\"><div class=\"et_pb_code_inner\"><div class=\"ns-vmap-editor-note\" style=\"position:relative;width:100%;overflow:hidden;aspect-ratio:5\/4;max-height:230px;\"><div style=\"position:absolute;inset:0;display:flex;align-items:center;justify-content:center;pointer-events:none\"><div style=\"padding:10px 12px;border:1px dashed rgba(170,180,205,.55);border-radius:6px;color:#8a93a6;font:600 12px\/1.4 sans-serif;background:rgba(8,13,25,.46)\"><div style=\"margin-bottom:6px\">Vector map \u2014 Main<\/div><code style=\"display:inline-block;padding:3px 6px;border-radius:4px;background:rgba(127,140,170,.12);color:inherit\"><div class=\"ns-vmap-editor-note\" style=\"position:relative;width:100%;overflow:hidden;aspect-ratio:5\/4;max-height:230px;\"><div style=\"position:absolute;inset:0;display:flex;align-items:center;justify-content:center;pointer-events:none\"><div style=\"padding:10px 12px;border:1px dashed rgba(170,180,205,.55);border-radius:6px;color:#8a93a6;font:600 12px\/1.4 sans-serif;background:rgba(8,13,25,.46)\"><div style=\"margin-bottom:6px\">Vector map \u2014 Main<\/div><code style=\"display:inline-block;padding:3px 6px;border-radius:4px;background:rgba(127,140,170,.12);color:inherit\">[ninesenses_vectormap #1]<\/code><div style=\"margin-top:6px;font-weight:400\">Max height: 230px. Preview on the live page.<\/div><\/div><\/div><\/div><\/code><div style=\"margin-top:6px;font-weight:400\">Max height: 230px. Preview on the live page.<\/div><\/div><\/div><\/div><\/div><\/div>\n<\/div>\n<\/div>\n<\/div>\n\n<div class=\"et_pb_section_1 et_pb_section et_section_regular et_block_section\">\n<div class=\"et_pb_row_2 et_pb_row et_block_row ns-quote\">\n<div class=\"et_pb_column_3 et_pb_column et_pb_column_4_4 et-last-child et_block_column et_pb_css_mix_blend_mode_passthrough\">\n<div class=\"et_pb_text_2 et_pb_text et_pb_bg_layout_light et_pb_module et_block_module\"><div class=\"et_pb_text_inner\"><blockquote style=\"border-left:0;padding-left:0;margin:30px 0 8px 0\"><p>\u201cWhat I had not realized is that extremely short exposures to a relatively simple computer program could induce powerful delusional thinking in quite normal people.\u201d<\/p><\/blockquote>\n<p><span>Joseph Weizenbaum (1923-2008), Inventor of Eliza<\/span><\/p>\n<\/div><\/div>\n<\/div>\n<\/div>\n<\/div>\n\n<div class=\"et_pb_section_2 et_pb_section et_section_regular et_block_section ns-block\" id=\"History_of_AI\" style=\"max-width:1080px;margin-left:auto;margin-right:auto;border-radius:0;hyphens:auto;-webkit-hyphens:auto;-ms-hyphens:auto\" lang=\"en\">\n<div class=\"et_pb_row_3 et_pb_row et_block_row ns-prose\">\n<div class=\"et_pb_column_4 et_pb_column et_pb_column_4_4 et-last-child et_block_column et_pb_css_mix_blend_mode_passthrough\">\n<div class=\"et_pb_text_3 et_pb_text et_pb_bg_layout_light et_pb_module et_block_module\"><div class=\"et_pb_text_inner\"><h2 style=\"display: inline; margin: 0 0.6em 0 0; padding: 0;\">History of AI<\/h2>\n<p><span>The idea of a machine-driven intelligence is not new. Literature has come up with speaking automatons way before the steam engine was invented, and since the arrival of computers, we have hoped for and <a href=\"\/can-ai-end-humanity\">feared AI smarter than humans<\/a>.<\/span><\/p>\n<\/div><\/div>\n<\/div>\n<\/div>\n\n<div class=\"et_pb_row_4 et_pb_row et_block_row ns-widget\">\n<div class=\"et_pb_column_5 et_pb_column et_pb_column_4_4 et-last-child et_block_column et_pb_css_mix_blend_mode_passthrough\">\n<div class=\"et_pb_code_1 et_pb_code et_pb_module\"><div class=\"et_pb_code_inner\">\t<div id=\"ns-aih-OmCoiZk\" class=\"ns-aih ns-aih-theme-dark\"\n\t\tdata-autoplay=\"yes\"\n\t\tdata-interval=\"8000\"\n\t\tdata-start=\"0\"\n\t\tstyle=\"--ns-aih-accent:#58a7f9;--ns-aih-secondary:#0c71c3;--ns-aih-pane-min:240px;\">\n\n\t\t<div class=\"ns-aih-strip-wrap\">\n\t\t\t<div class=\"ns-aih-rail\" role=\"tablist\" aria-label=\"AI history milestones\">\n\t\t\t\t<div class=\"ns-aih-rail-track\">\n\t\t\t\t\t<div class=\"ns-aih-rail-line\" aria-hidden=\"true\"><\/div>\n\t\t\t\t\t<canvas class=\"ns-aih-rail-canvas\" aria-hidden=\"true\"><\/canvas>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\"\n\t\t\t\t\t\t\tclass=\"ns-aih-dot is-active\"\n\t\t\t\t\t\t\tdata-index=\"0\"\n\t\t\t\t\t\t\tstyle=\"left:0%\"\n\t\t\t\t\t\t\trole=\"tab\"\n\t\t\t\t\t\t\taria-selected=\"true\"\n\t\t\t\t\t\t\taria-label=\"1816 - Fiction\">\n\t\t\t\t\t\t\t<span class=\"ns-aih-dot-core\" aria-hidden=\"true\"><\/span>\n\t\t\t\t\t\t\t<span class=\"ns-aih-dot-year\">1816<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\"\n\t\t\t\t\t\t\tclass=\"ns-aih-dot\"\n\t\t\t\t\t\t\tdata-index=\"1\"\n\t\t\t\t\t\t\tstyle=\"left:49.556%\"\n\t\t\t\t\t\t\trole=\"tab\"\n\t\t\t\t\t\t\taria-selected=\"false\"\n\t\t\t\t\t\t\taria-label=\"1950 - The Turing Test\">\n\t\t\t\t\t\t\t<span class=\"ns-aih-dot-core\" aria-hidden=\"true\"><\/span>\n\t\t\t\t\t\t\t<span class=\"ns-aih-dot-year\">1950<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\"\n\t\t\t\t\t\t\tclass=\"ns-aih-dot\"\n\t\t\t\t\t\t\tdata-index=\"2\"\n\t\t\t\t\t\t\tstyle=\"left:60.317%\"\n\t\t\t\t\t\t\trole=\"tab\"\n\t\t\t\t\t\t\taria-selected=\"false\"\n\t\t\t\t\t\t\taria-label=\"1966 - Eliza\">\n\t\t\t\t\t\t\t<span class=\"ns-aih-dot-core\" aria-hidden=\"true\"><\/span>\n\t\t\t\t\t\t\t<span class=\"ns-aih-dot-year\">1966<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\"\n\t\t\t\t\t\t\tclass=\"ns-aih-dot\"\n\t\t\t\t\t\t\tdata-index=\"3\"\n\t\t\t\t\t\t\tstyle=\"left:73.097%\"\n\t\t\t\t\t\t\trole=\"tab\"\n\t\t\t\t\t\t\taria-selected=\"false\"\n\t\t\t\t\t\t\taria-label=\"1980s - Machine Learning\">\n\t\t\t\t\t\t\t<span class=\"ns-aih-dot-core\" aria-hidden=\"true\"><\/span>\n\t\t\t\t\t\t\t<span class=\"ns-aih-dot-year\">1980s<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\"\n\t\t\t\t\t\t\tclass=\"ns-aih-dot\"\n\t\t\t\t\t\t\tdata-index=\"4\"\n\t\t\t\t\t\t\tstyle=\"left:79.822%\"\n\t\t\t\t\t\t\trole=\"tab\"\n\t\t\t\t\t\t\taria-selected=\"false\"\n\t\t\t\t\t\t\taria-label=\"1990s - Playing Chess (and Winning)\">\n\t\t\t\t\t\t\t<span class=\"ns-aih-dot-core\" aria-hidden=\"true\"><\/span>\n\t\t\t\t\t\t\t<span class=\"ns-aih-dot-year\">1990s<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\"\n\t\t\t\t\t\t\tclass=\"ns-aih-dot\"\n\t\t\t\t\t\t\tdata-index=\"5\"\n\t\t\t\t\t\t\tstyle=\"left:86.548%\"\n\t\t\t\t\t\t\trole=\"tab\"\n\t\t\t\t\t\t\taria-selected=\"false\"\n\t\t\t\t\t\t\taria-label=\"2000s - Seeing and Knowing\">\n\t\t\t\t\t\t\t<span class=\"ns-aih-dot-core\" aria-hidden=\"true\"><\/span>\n\t\t\t\t\t\t\t<span class=\"ns-aih-dot-year\">2000s<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\"\n\t\t\t\t\t\t\tclass=\"ns-aih-dot\"\n\t\t\t\t\t\t\tdata-index=\"6\"\n\t\t\t\t\t\t\tstyle=\"left:93.274%\"\n\t\t\t\t\t\t\trole=\"tab\"\n\t\t\t\t\t\t\taria-selected=\"false\"\n\t\t\t\t\t\t\taria-label=\"2010s - Solving Complex Problems\">\n\t\t\t\t\t\t\t<span class=\"ns-aih-dot-core\" aria-hidden=\"true\"><\/span>\n\t\t\t\t\t\t\t<span class=\"ns-aih-dot-year\">2010s<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\"\n\t\t\t\t\t\t\tclass=\"ns-aih-dot\"\n\t\t\t\t\t\t\tdata-index=\"7\"\n\t\t\t\t\t\t\tstyle=\"left:100%\"\n\t\t\t\t\t\t\trole=\"tab\"\n\t\t\t\t\t\t\taria-selected=\"false\"\n\t\t\t\t\t\t\taria-label=\"2020s - Listening and Speaking\">\n\t\t\t\t\t\t\t<span class=\"ns-aih-dot-core\" aria-hidden=\"true\"><\/span>\n\t\t\t\t\t\t\t<span class=\"ns-aih-dot-year\">2020s<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<button type=\"button\" class=\"ns-aih-playtoken\" aria-pressed=\"true\">\n\t\t\t\t\t<span class=\"ns-aih-icon-play\" aria-hidden=\"true\"><\/span>\n\t\t\t\t\t<span class=\"ns-aih-icon-pause\" aria-hidden=\"true\"><\/span>\n\t\t\t\t\t<span class=\"ns-aih-visually-hidden ns-aih-label-play\">Play<\/span>\n\t\t\t\t\t<span class=\"ns-aih-visually-hidden ns-aih-label-pause\">Pause<\/span>\n\t\t\t\t<\/button>\n\t\t\t<\/div>\n\t\t<\/div>\n\n\t\t<div class=\"ns-aih-pane\">\n\t\t\t<div class=\"ns-aih-slides\">\n\t\t\t\t\t\t\t\t\t<section class=\"ns-aih-slide is-active\" data-index=\"0\" aria-hidden=\"false\">\n\t\t\t\t\t\t<h3 class=\"ns-aih-slide-title\">Fiction<\/h3>\n\t\t\t\t\t\t<div class=\"ns-aih-slide-scene\">\n\t\t\t\t\t\t\t    <section\n        class=\"aihm-scene aihm-scene--automaton\"\n        id=\"ns-aih-OmCoiZk-0-automaton\"\n        data-aihm-scene=\"automaton\"\n        data-autoplay=\"1\"\n        style=\"--aihm-height:260px\"\n    >\n        <div class=\"aihm-frame\">\n            \t<figure class=\"aihm-visual aihm-automaton\" aria-label=\"Close view of the Jaquet-Droz Writer automaton moving its writing hand, quill, and head above the writing surface\">\n\t\t<img\n\t\t\tclass=\"aihm-automaton-image\"\n\t\t\tsrc=\"https:\/\/www.9senses.ai\/wp-content\/plugins\/ninesenses-ai-history\/assets\/images\/automaton-writer-poster-v032.webp\"\n\t\t\tdata-aihm-poster-src=\"https:\/\/www.9senses.ai\/wp-content\/plugins\/ninesenses-ai-history\/assets\/images\/automaton-writer-poster-v032.webp\"\n\t\t\tdata-aihm-motion-src=\"https:\/\/www.9senses.ai\/wp-content\/plugins\/ninesenses-ai-history\/assets\/images\/automaton-writer-motion-v032.webp\"\t\t\talt=\"The Jaquet-Droz Writer automaton with its writing surface, quill, and hands in view\"\n\t\t\tloading=\"lazy\"\n\t\t\tdecoding=\"async\"\n\t\t>\n\t\t<figcaption class=\"aihm-credit\">\n\t\t\t<a class=\"aihm-credit-link\" href=\"https:\/\/commons.wikimedia.org\/wiki\/File:Jaquet_Droz_automata_-_Writer.jpg\" target=\"_blank\" rel=\"noopener noreferrer nofollow\" aria-label=\"View image credit and license details on Wikimedia Commons\">Image credit &amp; license<\/a>\n\t\t<\/figcaption>\n\t<\/figure>\n\t        <\/div>\n    <\/section>\n    \t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t<div class=\"ns-aih-slide-text\">\n\t\t\t\t\t\t\t<p class=\"ns-aih-slide-body\">The idea of &quot;automatons&quot; acting &quot;intelligent&quot; is much older than computers themselves. For example, in E.T.A. Hoffmann&#039;s &quot;The Sandman&quot;, published in 1816, a beautiful girl named Olimpia is introduced. She dances and sings beautifully, but only speaks a few words. In fact, she is an automaton, created by physics professor Spalanzani.<\/p>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t<section class=\"ns-aih-slide\" data-index=\"1\" aria-hidden=\"true\">\n\t\t\t\t\t\t<h3 class=\"ns-aih-slide-title\">The Turing Test<\/h3>\n\t\t\t\t\t\t<div class=\"ns-aih-slide-scene\">\n\t\t\t\t\t\t\t    <section\n        class=\"aihm-scene aihm-scene--turing\"\n        id=\"ns-aih-OmCoiZk-1-turing\"\n        data-aihm-scene=\"turing\"\n        data-autoplay=\"1\"\n        style=\"--aihm-height:260px\"\n    >\n        <div class=\"aihm-frame\">\n            \t<div class=\"aihm-visual aihm-vector-scene aihm-turing-preview\" role=\"img\" aria-label=\"Animated diagram of an interrogator exchanging text messages with two concealed respondents\">\n\t\t<svg class=\"aihm-vector-svg\" viewBox=\"0 0 640 360\" preserveAspectRatio=\"xMidYMid meet\" aria-hidden=\"true\" focusable=\"false\">\n\t\t\t<defs>\n\t\t\t\t<radialGradient id=\"ns-aih-OmCoiZk-1-turing-t-node\" cx=\"34%\" cy=\"28%\" r=\"72%\">\n\t\t\t\t\t<stop offset=\"0\" stop-color=\"#ffffff\" stop-opacity=\".98\"\/>\n\t\t\t\t\t<stop offset=\".22\" stop-color=\"#d6ebff\" stop-opacity=\".96\"\/>\n\t\t\t\t\t<stop offset=\".62\" stop-color=\"#58a7f9\" stop-opacity=\".82\"\/>\n\t\t\t\t\t<stop offset=\"1\" stop-color=\"#58a7f9\" stop-opacity=\"0\"\/>\n\t\t\t\t<\/radialGradient>\n\t\t\t\t<radialGradient id=\"ns-aih-OmCoiZk-1-turing-t-blue\" cx=\"34%\" cy=\"28%\" r=\"72%\">\n\t\t\t\t\t<stop offset=\"0\" stop-color=\"#ffffff\" stop-opacity=\".98\"\/>\n\t\t\t\t\t<stop offset=\".25\" stop-color=\"#bcdfff\" stop-opacity=\".96\"\/>\n\t\t\t\t\t<stop offset=\".66\" stop-color=\"#5e9fe6\" stop-opacity=\".82\"\/>\n\t\t\t\t\t<stop offset=\"1\" stop-color=\"#5e9fe6\" stop-opacity=\"0\"\/>\n\t\t\t\t<\/radialGradient>\n\t\t\t\t<filter id=\"ns-aih-OmCoiZk-1-turing-t-glow\" x=\"-120%\" y=\"-120%\" width=\"340%\" height=\"340%\">\n\t\t\t\t\t<feGaussianBlur stdDeviation=\"4\" result=\"blur\"\/>\n\t\t\t\t\t<feMerge><feMergeNode in=\"blur\"\/><feMergeNode in=\"SourceGraphic\"\/><\/feMerge>\n\t\t\t\t<\/filter>\n\t\t\t\t<linearGradient id=\"ns-aih-OmCoiZk-1-turing-t-screen\" x1=\"0\" y1=\"0\" x2=\"1\" y2=\"1\">\n\t\t\t\t\t<stop offset=\"0\" stop-color=\"#0d1b35\"\/>\n\t\t\t\t\t<stop offset=\"1\" stop-color=\"#071126\"\/>\n\t\t\t\t<\/linearGradient>\n\t\t\t<\/defs>\n\n\t\t\t<rect width=\"640\" height=\"360\" fill=\"#02091b\"\/>\n\t\t\t<circle class=\"aihm-vector-star aihm-vector-star--1\" cx=\"51\" cy=\"46\" r=\"1.7\" fill=\"#8ec5ff\"\/>\n\t\t\t<circle class=\"aihm-vector-star aihm-vector-star--2\" cx=\"588\" cy=\"76\" r=\"1.2\" fill=\"#7dbfff\"\/>\n\t\t\t<circle class=\"aihm-vector-star aihm-vector-star--3\" cx=\"318\" cy=\"30\" r=\"1.4\" fill=\"#8ec5ff\"\/>\n\t\t\t<circle class=\"aihm-vector-star aihm-vector-star--4\" cx=\"94\" cy=\"282\" r=\"1\" fill=\"#7dbfff\"\/>\n\t\t\t<circle class=\"aihm-vector-star aihm-vector-star--5\" cx=\"558\" cy=\"300\" r=\"1.6\" fill=\"#8ec5ff\"\/>\n\n\t\t\t<path id=\"ns-aih-OmCoiZk-1-turing-t-qa\" class=\"aihm-vector-route aihm-vector-route--query\" d=\"M320 288 C286 236 232 210 156 163\"\/>\n\t\t\t<path id=\"ns-aih-OmCoiZk-1-turing-t-qb\" class=\"aihm-vector-route aihm-vector-route--query\" d=\"M320 288 C354 236 408 210 484 163\"\/>\n\t\t\t<path class=\"aihm-vector-route aihm-vector-route--return\" d=\"M167 169 C238 215 282 242 321 294\"\/>\n\t\t\t<path class=\"aihm-vector-route aihm-vector-route--return\" d=\"M473 169 C402 215 358 242 319 294\"\/>\n\n\t\t\t<line class=\"aihm-turing-screenline\" x1=\"35\" y1=\"194\" x2=\"605\" y2=\"194\"\/>\n\t\t\t<rect class=\"aihm-turing-screen-label-bg\" x=\"246\" y=\"179\" width=\"148\" height=\"28\"\/>\n\t\t\t<text class=\"aihm-vector-caption\" x=\"320\" y=\"197\" text-anchor=\"middle\">text only<\/text>\n\n\t\t\t<g class=\"aihm-turing-terminal aihm-turing-terminal--a\">\n\t\t\t\t<rect x=\"72\" y=\"52\" width=\"176\" height=\"104\" fill=\"url(#ns-aih-OmCoiZk-1-turing-t-screen)\"\/>\n\t\t\t\t<rect x=\"72\" y=\"52\" width=\"176\" height=\"104\" class=\"aihm-terminal-outline\"\/>\n\t\t\t\t<circle cx=\"99\" cy=\"80\" r=\"10\" fill=\"url(#ns-aih-OmCoiZk-1-turing-t-node)\" filter=\"url(#ns-aih-OmCoiZk-1-turing-t-glow)\"\/>\n\t\t\t\t<text class=\"aihm-terminal-letter\" x=\"99\" y=\"85\" text-anchor=\"middle\">A<\/text>\n\t\t\t\t<text class=\"aihm-terminal-title\" x=\"121\" y=\"83\">respondent<\/text>\n\t\t\t\t<line class=\"aihm-terminal-rule\" x1=\"92\" y1=\"103\" x2=\"226\" y2=\"103\"\/>\n\t\t\t\t<line class=\"aihm-terminal-rule aihm-terminal-rule--short\" x1=\"92\" y1=\"119\" x2=\"205\" y2=\"119\"\/>\n\t\t\t\t<line class=\"aihm-terminal-rule aihm-terminal-rule--reply\" x1=\"92\" y1=\"135\" x2=\"219\" y2=\"135\"\/>\n\t\t\t<\/g>\n\n\t\t\t<g class=\"aihm-turing-terminal aihm-turing-terminal--b\">\n\t\t\t\t<rect x=\"392\" y=\"52\" width=\"176\" height=\"104\" fill=\"url(#ns-aih-OmCoiZk-1-turing-t-screen)\"\/>\n\t\t\t\t<rect x=\"392\" y=\"52\" width=\"176\" height=\"104\" class=\"aihm-terminal-outline\"\/>\n\t\t\t\t<circle cx=\"419\" cy=\"80\" r=\"10\" fill=\"url(#ns-aih-OmCoiZk-1-turing-t-blue)\" filter=\"url(#ns-aih-OmCoiZk-1-turing-t-glow)\"\/>\n\t\t\t\t<text class=\"aihm-terminal-letter\" x=\"419\" y=\"85\" text-anchor=\"middle\">B<\/text>\n\t\t\t\t<text class=\"aihm-terminal-title\" x=\"441\" y=\"83\">respondent<\/text>\n\t\t\t\t<line class=\"aihm-terminal-rule\" x1=\"412\" y1=\"103\" x2=\"546\" y2=\"103\"\/>\n\t\t\t\t<line class=\"aihm-terminal-rule aihm-terminal-rule--short\" x1=\"412\" y1=\"119\" x2=\"525\" y2=\"119\"\/>\n\t\t\t\t<line class=\"aihm-terminal-rule aihm-terminal-rule--reply\" x1=\"412\" y1=\"135\" x2=\"539\" y2=\"135\"\/>\n\t\t\t<\/g>\n\n\t\t\t<g class=\"aihm-turing-terminal aihm-turing-terminal--judge\">\n\t\t\t\t<rect x=\"226\" y=\"260\" width=\"188\" height=\"70\" fill=\"url(#ns-aih-OmCoiZk-1-turing-t-screen)\"\/>\n\t\t\t\t<rect x=\"226\" y=\"260\" width=\"188\" height=\"70\" class=\"aihm-terminal-outline\"\/>\n\t\t\t\t<text class=\"aihm-terminal-title aihm-terminal-title--judge\" x=\"320\" y=\"283\" text-anchor=\"middle\">interrogator<\/text>\n\t\t\t\t<text class=\"aihm-turing-choice\" x=\"320\" y=\"313\" text-anchor=\"middle\">A&nbsp;&nbsp;&nbsp;?&nbsp;&nbsp;&nbsp;B<\/text>\n\t\t\t<\/g>\n\n\t\t\t<circle class=\"aihm-vector-token aihm-vector-token--query\" r=\"5\" fill=\"url(#ns-aih-OmCoiZk-1-turing-t-node)\" filter=\"url(#ns-aih-OmCoiZk-1-turing-t-glow)\" data-route=\"#ns-aih-OmCoiZk-1-turing-t-qa\" data-cycle=\"9000\" data-start=\"450\" data-end=\"2200\" data-radius=\"5\" data-static=\".48\"\/>\n\t\t\t<circle class=\"aihm-vector-token aihm-vector-token--query\" r=\"5\" fill=\"url(#ns-aih-OmCoiZk-1-turing-t-node)\" filter=\"url(#ns-aih-OmCoiZk-1-turing-t-glow)\" data-route=\"#ns-aih-OmCoiZk-1-turing-t-qb\" data-cycle=\"9000\" data-start=\"800\" data-end=\"2550\" data-radius=\"5\" data-static=\".58\"\/>\n\t\t\t<circle class=\"aihm-vector-token aihm-vector-token--reply\" r=\"4.5\" fill=\"url(#ns-aih-OmCoiZk-1-turing-t-blue)\" filter=\"url(#ns-aih-OmCoiZk-1-turing-t-glow)\" data-route=\"#ns-aih-OmCoiZk-1-turing-t-qa\" data-cycle=\"9000\" data-start=\"3350\" data-end=\"5150\" data-reverse=\"1\" data-radius=\"4.5\" data-static=\".72\"\/>\n\t\t\t<circle class=\"aihm-vector-token aihm-vector-token--reply\" r=\"4.5\" fill=\"url(#ns-aih-OmCoiZk-1-turing-t-blue)\" filter=\"url(#ns-aih-OmCoiZk-1-turing-t-glow)\" data-route=\"#ns-aih-OmCoiZk-1-turing-t-qb\" data-cycle=\"9000\" data-start=\"3650\" data-end=\"5450\" data-reverse=\"1\" data-radius=\"4.5\" data-static=\".64\"\/>\n\t\t<\/svg>\n\t<\/div>\n\t        <\/div>\n    <\/section>\n    \t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t<div class=\"ns-aih-slide-text\">\n\t\t\t\t\t\t\t<p class=\"ns-aih-slide-body\">With the appearance of the first computers in the 1940s, the fascination with the technology quickly led to the idea that they could soon perform as intelligently as humans. In 1950, British mathematician and computer scientist Alan Turing devised a test to evaluate when a computer would be able to emulate a human conversation convincingly. It took almost 65 years for the first simulation to narrowly pass.<\/p>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t<section class=\"ns-aih-slide\" data-index=\"2\" aria-hidden=\"true\">\n\t\t\t\t\t\t<h3 class=\"ns-aih-slide-title\">Eliza<\/h3>\n\t\t\t\t\t\t<div class=\"ns-aih-slide-scene\">\n\t\t\t\t\t\t\t    <section\n        class=\"aihm-scene aihm-scene--eliza\"\n        id=\"ns-aih-OmCoiZk-2-eliza\"\n        data-aihm-scene=\"eliza\"\n        data-autoplay=\"1\"\n        style=\"--aihm-height:260px\"\n    >\n        <div class=\"aihm-frame\">\n            \t<a class=\"aihm-visual aihm-action-surface aihm-eliza-preview\" href=\"#eliza\" aria-label=\"Open the 1966 Eliza conversation\">\n\t\t<div class=\"aihm-eliza-paper\">\n\t\t\t<div class=\"aihm-eliza-log\" aria-hidden=\"true\"><\/div>\n\t\t\t<span class=\"aihm-eliza-cursor\" aria-hidden=\"true\">\u258c<\/span>\n\t\t<\/div>\n\t\t<span class=\"aihm-hitarea\" aria-hidden=\"true\">\n\t\t\t<span class=\"aihm-button\">Talk to Eliza<\/span>\n\t\t<\/span>\n\t<\/a>\n\t        <\/div>\n    <\/section>\n    \t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t<div class=\"ns-aih-slide-text\">\n\t\t\t\t\t\t\t<p class=\"ns-aih-slide-body\">When German-American computer scientist Joseph Weizenbaum created chat program &quot;Eliza&quot; in 1966, simulating the dialogue with a psychologist, it was meant like a playful first attempt at processing natural speech. Even though the logic behind it was very simple, many people considered it intelligent and expected computers to be able to speak like humans soon.<\/p>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aih-slide-link-row\"><a class=\"ns-aih-slide-link\" href=\"#eliza\">Talk to Eliza<span class=\"ns-aih-slide-link-arrow\" aria-hidden=\"true\">&rarr;<\/span><\/a><\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t<section class=\"ns-aih-slide\" data-index=\"3\" aria-hidden=\"true\">\n\t\t\t\t\t\t<h3 class=\"ns-aih-slide-title\">Machine Learning<\/h3>\n\t\t\t\t\t\t<div class=\"ns-aih-slide-scene\">\n\t\t\t\t\t\t\t    <section\n        class=\"aihm-scene aihm-scene--machinelearning\"\n        id=\"ns-aih-OmCoiZk-3-machinelearning\"\n        data-aihm-scene=\"machinelearning\"\n        data-autoplay=\"1\"\n        style=\"--aihm-height:260px\"\n    >\n        <div class=\"aihm-frame\">\n            \t<div class=\"aihm-visual aihm-vector-scene aihm-ml-preview\" role=\"img\" aria-label=\"Animated neural network showing a training example moving forward and an error signal moving backward to adjust weights\">\n\t\t<svg class=\"aihm-vector-svg\" viewBox=\"0 0 640 360\" preserveAspectRatio=\"xMidYMid meet\" aria-hidden=\"true\" focusable=\"false\">\n\t\t\t<defs>\n\t\t\t\t<radialGradient id=\"ns-aih-OmCoiZk-3-machinelearning-ml-green\" cx=\"34%\" cy=\"28%\" r=\"72%\">\n\t\t\t\t\t<stop offset=\"0\" stop-color=\"#ffffff\" stop-opacity=\".98\"\/>\n\t\t\t\t\t<stop offset=\".22\" stop-color=\"#dbeeff\" stop-opacity=\".96\"\/>\n\t\t\t\t\t<stop offset=\".62\" stop-color=\"#58a7f9\" stop-opacity=\".82\"\/>\n\t\t\t\t\t<stop offset=\"1\" stop-color=\"#58a7f9\" stop-opacity=\"0\"\/>\n\t\t\t\t<\/radialGradient>\n\t\t\t\t<radialGradient id=\"ns-aih-OmCoiZk-3-machinelearning-ml-blue\" cx=\"34%\" cy=\"28%\" r=\"72%\">\n\t\t\t\t\t<stop offset=\"0\" stop-color=\"#ffffff\" stop-opacity=\".98\"\/>\n\t\t\t\t\t<stop offset=\".24\" stop-color=\"#c7e5ff\" stop-opacity=\".96\"\/>\n\t\t\t\t\t<stop offset=\".66\" stop-color=\"#62a7ea\" stop-opacity=\".82\"\/>\n\t\t\t\t\t<stop offset=\"1\" stop-color=\"#62a7ea\" stop-opacity=\"0\"\/>\n\t\t\t\t<\/radialGradient>\n\t\t\t\t<radialGradient id=\"ns-aih-OmCoiZk-3-machinelearning-ml-error\" cx=\"34%\" cy=\"28%\" r=\"72%\">\n\t\t\t\t\t<stop offset=\"0\" stop-color=\"#ffffff\" stop-opacity=\".98\"\/>\n\t\t\t\t\t<stop offset=\".24\" stop-color=\"#ffd6c9\" stop-opacity=\".96\"\/>\n\t\t\t\t\t<stop offset=\".66\" stop-color=\"#e77962\" stop-opacity=\".82\"\/>\n\t\t\t\t\t<stop offset=\"1\" stop-color=\"#e77962\" stop-opacity=\"0\"\/>\n\t\t\t\t<\/radialGradient>\n\t\t\t\t<filter id=\"ns-aih-OmCoiZk-3-machinelearning-ml-glow\" x=\"-120%\" y=\"-120%\" width=\"340%\" height=\"340%\">\n\t\t\t\t\t<feGaussianBlur stdDeviation=\"4\" result=\"blur\"\/>\n\t\t\t\t\t<feMerge><feMergeNode in=\"blur\"\/><feMergeNode in=\"SourceGraphic\"\/><\/feMerge>\n\t\t\t\t<\/filter>\n\t\t\t<\/defs>\n\n\t\t\t<rect width=\"640\" height=\"360\" fill=\"#02091b\"\/>\n\t\t\t<circle class=\"aihm-vector-star aihm-vector-star--1\" cx=\"62\" cy=\"52\" r=\"1.2\" fill=\"#8ec5ff\"\/>\n\t\t\t<circle class=\"aihm-vector-star aihm-vector-star--2\" cx=\"600\" cy=\"46\" r=\"1.5\" fill=\"#70b9ff\"\/>\n\t\t\t<circle class=\"aihm-vector-star aihm-vector-star--3\" cx=\"329\" cy=\"24\" r=\"1.1\" fill=\"#8ec5ff\"\/>\n\t\t\t<circle class=\"aihm-vector-star aihm-vector-star--4\" cx=\"575\" cy=\"303\" r=\"1.2\" fill=\"#70b9ff\"\/>\n\n\t\t\t<g class=\"aihm-ml-data-panel\">\n\t\t\t\t<rect x=\"38\" y=\"88\" width=\"128\" height=\"168\" class=\"aihm-ml-panel-bg\"\/>\n\t\t\t\t<text class=\"aihm-vector-caption aihm-ml-heading\" x=\"102\" y=\"58\" text-anchor=\"middle\">training<\/text>\n\t\t\t\t<text class=\"aihm-vector-caption aihm-ml-heading\" x=\"102\" y=\"78\" text-anchor=\"middle\">data<\/text>\n\t\t\t\t<line class=\"aihm-ml-boundary\" x1=\"62\" y1=\"228\" x2=\"144\" y2=\"118\"\/>\n\t\t\t\t<circle cx=\"70\" cy=\"206\" r=\"4\" fill=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-green)\"\/>\n\t\t\t\t<circle cx=\"91\" cy=\"218\" r=\"3.6\" fill=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-green)\"\/>\n\t\t\t\t<circle cx=\"113\" cy=\"190\" r=\"4.2\" fill=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-green)\"\/>\n\t\t\t\t<circle cx=\"98\" cy=\"137\" r=\"3.8\" fill=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-blue)\"\/>\n\t\t\t\t<circle cx=\"127\" cy=\"153\" r=\"4.4\" fill=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-blue)\"\/>\n\t\t\t\t<circle cx=\"142\" cy=\"123\" r=\"3.6\" fill=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-blue)\"\/>\n\t\t\t\t<circle class=\"aihm-ml-sample-ring\" cx=\"113\" cy=\"190\" r=\"11\"\/>\n\t\t\t<\/g>\n\n\t\t\t<text class=\"aihm-vector-caption aihm-ml-heading\" x=\"246\" y=\"52\" text-anchor=\"middle\">input<\/text>\n\t\t\t<text class=\"aihm-vector-caption aihm-ml-heading\" x=\"388\" y=\"52\" text-anchor=\"middle\">hidden<\/text>\n\t\t\t<text class=\"aihm-vector-caption aihm-ml-heading\" x=\"530\" y=\"52\" text-anchor=\"middle\">output<\/text>\n\n\t\t\t<g class=\"aihm-ml-edges\">\n\t\t\t\t<path id=\"ns-aih-OmCoiZk-3-machinelearning-ml-i1h1\" class=\"aihm-ml-edge aihm-ml-edge--a\" d=\"M246 112 C292 112 338 103 388 98\"\/>\n\t\t\t\t<path id=\"ns-aih-OmCoiZk-3-machinelearning-ml-i1h2\" class=\"aihm-ml-edge\" d=\"M246 112 C302 140 337 151 388 166\"\/>\n\t\t\t\t<path id=\"ns-aih-OmCoiZk-3-machinelearning-ml-i2h2\" class=\"aihm-ml-edge aihm-ml-edge--b\" d=\"M246 180 C300 180 340 175 388 166\"\/>\n\t\t\t\t<path id=\"ns-aih-OmCoiZk-3-machinelearning-ml-i2h3\" class=\"aihm-ml-edge\" d=\"M246 180 C300 206 340 221 388 234\"\/>\n\t\t\t\t<path id=\"ns-aih-OmCoiZk-3-machinelearning-ml-i3h2\" class=\"aihm-ml-edge\" d=\"M246 248 C302 216 340 194 388 166\"\/>\n\t\t\t\t<path id=\"ns-aih-OmCoiZk-3-machinelearning-ml-i3h3\" class=\"aihm-ml-edge aihm-ml-edge--c\" d=\"M246 248 C294 248 340 239 388 234\"\/>\n\t\t\t\t<path id=\"ns-aih-OmCoiZk-3-machinelearning-ml-h1o1\" class=\"aihm-ml-edge aihm-ml-edge--a\" d=\"M388 98 C437 103 482 120 530 142\"\/>\n\t\t\t\t<path id=\"ns-aih-OmCoiZk-3-machinelearning-ml-h2o1\" class=\"aihm-ml-edge aihm-ml-edge--b\" d=\"M388 166 C438 160 482 151 530 142\"\/>\n\t\t\t\t<path id=\"ns-aih-OmCoiZk-3-machinelearning-ml-h2o2\" class=\"aihm-ml-edge\" d=\"M388 166 C438 179 482 198 530 218\"\/>\n\t\t\t\t<path id=\"ns-aih-OmCoiZk-3-machinelearning-ml-h3o2\" class=\"aihm-ml-edge aihm-ml-edge--c\" d=\"M388 234 C437 231 482 224 530 218\"\/>\n\t\t\t<\/g>\n\n\t\t\t<g class=\"aihm-ml-nodes\">\n\t\t\t\t<circle cx=\"246\" cy=\"112\" r=\"12\" fill=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-green)\" filter=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-glow)\"\/>\n\t\t\t\t<circle cx=\"246\" cy=\"180\" r=\"10\" fill=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-green)\" filter=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-glow)\"\/>\n\t\t\t\t<circle cx=\"246\" cy=\"248\" r=\"8\" fill=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-green)\" filter=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-glow)\"\/>\n\t\t\t\t<circle cx=\"388\" cy=\"98\" r=\"9\" fill=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-blue)\" filter=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-glow)\"\/>\n\t\t\t\t<circle cx=\"388\" cy=\"166\" r=\"12\" fill=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-blue)\" filter=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-glow)\"\/>\n\t\t\t\t<circle cx=\"388\" cy=\"234\" r=\"8\" fill=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-blue)\" filter=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-glow)\"\/>\n\t\t\t\t<circle cx=\"530\" cy=\"142\" r=\"13\" fill=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-green)\" filter=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-glow)\"\/>\n\t\t\t\t<circle cx=\"530\" cy=\"218\" r=\"9\" fill=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-blue)\" filter=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-glow)\"\/>\n\t\t\t<\/g>\n\n\t\t\t<g class=\"aihm-ml-metric\">\n\t\t\t\t<rect x=\"482\" y=\"278\" width=\"118\" height=\"46\" class=\"aihm-ml-panel-bg\"\/>\n\t\t\t\t<text class=\"aihm-vector-caption aihm-ml-metric-label\" x=\"500\" y=\"299\">error<\/text>\n\t\t\t\t<text class=\"aihm-ml-error-value\" x=\"582\" y=\"311\" text-anchor=\"end\">0.64<\/text>\n\t\t\t<\/g>\n\t\t\t<text class=\"aihm-ml-backprop-label\" x=\"386\" y=\"312\" text-anchor=\"middle\">weight adjustment \u2190 error<\/text>\n\n\t\t\t<circle class=\"aihm-vector-token aihm-vector-token--forward\" r=\"5\" fill=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-green)\" filter=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-glow)\" data-route=\"#ns-aih-OmCoiZk-3-machinelearning-ml-i1h1\" data-cycle=\"8200\" data-start=\"400\" data-end=\"2100\" data-radius=\"5\" data-static=\".38\"\/>\n\t\t\t<circle class=\"aihm-vector-token aihm-vector-token--forward\" r=\"4.5\" fill=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-green)\" filter=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-glow)\" data-route=\"#ns-aih-OmCoiZk-3-machinelearning-ml-i2h2\" data-cycle=\"8200\" data-start=\"650\" data-end=\"2250\" data-radius=\"4.5\" data-static=\".54\"\/>\n\t\t\t<circle class=\"aihm-vector-token aihm-vector-token--forward\" r=\"4\" fill=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-green)\" filter=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-glow)\" data-route=\"#ns-aih-OmCoiZk-3-machinelearning-ml-i3h3\" data-cycle=\"8200\" data-start=\"850\" data-end=\"2400\" data-radius=\"4\" data-static=\".65\"\/>\n\t\t\t<circle class=\"aihm-vector-token aihm-vector-token--forward\" r=\"5\" fill=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-green)\" filter=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-glow)\" data-route=\"#ns-aih-OmCoiZk-3-machinelearning-ml-h1o1\" data-cycle=\"8200\" data-start=\"2100\" data-end=\"3500\" data-radius=\"5\" data-static=\".48\"\/>\n\t\t\t<circle class=\"aihm-vector-token aihm-vector-token--forward\" r=\"4.5\" fill=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-green)\" filter=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-glow)\" data-route=\"#ns-aih-OmCoiZk-3-machinelearning-ml-h2o1\" data-cycle=\"8200\" data-start=\"2250\" data-end=\"3650\" data-radius=\"4.5\" data-static=\".58\"\/>\n\t\t\t<circle class=\"aihm-vector-token aihm-vector-token--error\" r=\"5\" fill=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-error)\" filter=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-glow)\" data-route=\"#ns-aih-OmCoiZk-3-machinelearning-ml-h2o1\" data-cycle=\"8200\" data-start=\"4100\" data-end=\"5500\" data-reverse=\"1\" data-radius=\"5\" data-static=\".72\"\/>\n\t\t\t<circle class=\"aihm-vector-token aihm-vector-token--error\" r=\"4.5\" fill=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-error)\" filter=\"url(#ns-aih-OmCoiZk-3-machinelearning-ml-glow)\" data-route=\"#ns-aih-OmCoiZk-3-machinelearning-ml-i2h2\" data-cycle=\"8200\" data-start=\"5300\" data-end=\"6800\" data-reverse=\"1\" data-radius=\"4.5\" data-static=\".63\"\/>\n\t\t<\/svg>\n\t<\/div>\n\t        <\/div>\n    <\/section>\n    \t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t<div class=\"ns-aih-slide-text\">\n\t\t\t\t\t\t\t<p class=\"ns-aih-slide-body\">With increasingly powerful computers and larger storage capabilities that were able to handle large datasets, the first successful machine learning approaches were introduced. They were based on the ability to autonomously find patterns in data, relate them back to certain events and conditions and suggest or take action.<\/p>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t<section class=\"ns-aih-slide\" data-index=\"4\" aria-hidden=\"true\">\n\t\t\t\t\t\t<h3 class=\"ns-aih-slide-title\">Playing Chess (and Winning)<\/h3>\n\t\t\t\t\t\t<div class=\"ns-aih-slide-scene\">\n\t\t\t\t\t\t\t    <section\n        class=\"aihm-scene aihm-scene--deepblue\"\n        id=\"ns-aih-OmCoiZk-4-deepblue\"\n        data-aihm-scene=\"deepblue\"\n        data-autoplay=\"1\"\n        style=\"--aihm-height:260px\"\n    >\n        <div class=\"aihm-frame\">\n            \t<div class=\"aihm-visual aihm-deepblue-preview\" role=\"group\" aria-label=\"Deep Blue chess preview\">\n\t\t<div class=\"aihm-db-console\">\n\t\t\t<header class=\"aihm-db-head dbct-header\">\n\t\t\t\t<div class=\"dbct-bars\" aria-hidden=\"true\"><i><\/i><i><\/i><i><\/i><i><\/i><i><\/i><i><\/i><i><\/i><i><\/i><\/div>\n\t\t\t\t<div class=\"dbct-title\" role=\"heading\" aria-level=\"3\">DEEP BLUE<small>RS\/6000 SP &nbsp;\u00b7&nbsp; OPERATOR CONSOLE<\/small><\/div>\n\t\t\t\t<div class=\"dbct-hright\">GAME 6 &nbsp;\u00b7&nbsp; NEW YORK 1997<br>ENGINE: <b>DEEP BLUE<\/b><\/div>\n\t\t\t<\/header>\n\t\t\t<div class=\"aihm-db-layout\">\n\t\t\t\t<div class=\"aihm-db-board\" role=\"img\" aria-label=\"Chess board replaying the opening of game six\"><\/div>\n\t\t\t\t<div class=\"aihm-db-telemetry\">\n\t\t\t\t\t<div>MOVE <b class=\"aihm-db-move\">1. e4<\/b><\/div>\n\t\t\t\t\t<div>PLY <b class=\"aihm-db-search\">1 \/ 10<\/b><\/div>\n\t\t\t\t\t<div>RESULT <b>DEEP BLUE 1\u20130<\/b><\/div>\n\t\t\t\t<\/div>\n\t\t\t<\/div>\n\t\t\t<div class=\"aihm-db-controls dbct-controls\">\n\t\t\t\t<button type=\"button\" data-dbct-popup-trigger=\"deepblue\" aria-label=\"Open the Deep Blue chess experience\">Play Chess<\/button>\n\t\t\t<\/div>\n\t\t<\/div>\n\t<\/div>\n\t        <\/div>\n    <\/section>\n    \t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t<div class=\"ns-aih-slide-text\">\n\t\t\t\t\t\t\t<p class=\"ns-aih-slide-body\">In 1997, IBM&#039;s Deep Blue supercomputer won its first match against acting chess champion Garry Kasparov. While mostly driven by sheer power which helped build its game on computing more than 200 million positions a second, it was using machine learning elements (heuristics and minimax optimization techniques) mid-game, which can be considered AI.<\/p>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aih-slide-link-row\"><a class=\"ns-aih-slide-link\" href=\"#deepblue\">Play Chess<span class=\"ns-aih-slide-link-arrow\" aria-hidden=\"true\">&rarr;<\/span><\/a><\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t<section class=\"ns-aih-slide\" data-index=\"5\" aria-hidden=\"true\">\n\t\t\t\t\t\t<h3 class=\"ns-aih-slide-title\">Seeing and Knowing<\/h3>\n\t\t\t\t\t\t<div class=\"ns-aih-slide-scene\">\n\t\t\t\t\t\t\t    <section\n        class=\"aihm-scene aihm-scene--vision\"\n        id=\"ns-aih-OmCoiZk-5-vision\"\n        data-aihm-scene=\"vision\"\n        data-autoplay=\"1\"\n        style=\"--aihm-height:260px\"\n    >\n        <div class=\"aihm-frame\">\n            \t<div class=\"aihm-visual aihm-vision-real\" role=\"img\" aria-label=\"Animated computer-vision analysis of a cherry image: real image, inverted scan, pixelated scan, then the label cherry with contour overlays\">\n\t\t<div class=\"aihm-vision-photo aihm-vision-photo--clear\" style=\"background-image:url(https:\/\/www.9senses.ai\/wp-content\/plugins\/ninesenses-ai-history\/assets\/images\/vision-cherry-clear-v046.webp)\"><\/div>\n\t\t<div class=\"aihm-vision-layer aihm-vision-layer--invert\">\n\t\t\t<div class=\"aihm-vision-photo aihm-vision-photo--invert\" style=\"background-image:url(https:\/\/www.9senses.ai\/wp-content\/plugins\/ninesenses-ai-history\/assets\/images\/vision-cherry-invert-v046.webp)\"><\/div>\n\t\t<\/div>\n\t\t<div class=\"aihm-vision-layer aihm-vision-layer--pixel\">\n\t\t\t<div class=\"aihm-vision-photo aihm-vision-photo--pixel\" style=\"background-image:url(https:\/\/www.9senses.ai\/wp-content\/plugins\/ninesenses-ai-history\/assets\/images\/vision-cherry-pixel-v046.webp)\"><\/div>\n\t\t<\/div>\n\t\t<div class=\"aihm-vision-layer aihm-vision-layer--final\">\n\t\t\t<div class=\"aihm-vision-photo aihm-vision-photo--final\" style=\"background-image:url(https:\/\/www.9senses.ai\/wp-content\/plugins\/ninesenses-ai-history\/assets\/images\/vision-cherry-final-v072.webp)\"><\/div>\n\t\t\t<div class=\"aihm-vision-tag-wrap\" aria-hidden=\"true\">\n\t\t\t\t<div class=\"aihm-vision-tag-box\">cherry<\/div>\n\t\t\t<\/div>\n\t\t<\/div>\n\t\t<div class=\"aihm-vision-grid\" aria-hidden=\"true\"><\/div>\n\t\t<div class=\"aihm-vision-scanline aihm-vision-scanline--invert\" aria-hidden=\"true\"><\/div>\n\t\t<div class=\"aihm-vision-scanline aihm-vision-scanline--pixel\" aria-hidden=\"true\"><\/div>\n\t\t<div class=\"aihm-vision-scanline aihm-vision-scanline--final\" aria-hidden=\"true\"><\/div>\n\t<\/div>\n\t        <\/div>\n    <\/section>\n    \t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t<div class=\"ns-aih-slide-text\">\n\t\t\t\t\t\t\t<p class=\"ns-aih-slide-body\">While optical character recognition (OCR) had been developed long ago, computers became capable of &quot;seeing&quot; in the late 20th and the early 21st century. This is when the first face and object recognition systems were developed. By now, AI is able to routinely identify people and objects and also understand what they are doing.<\/p>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t<section class=\"ns-aih-slide\" data-index=\"6\" aria-hidden=\"true\">\n\t\t\t\t\t\t<h3 class=\"ns-aih-slide-title\">Solving Complex Problems<\/h3>\n\t\t\t\t\t\t<div class=\"ns-aih-slide-scene\">\n\t\t\t\t\t\t\t    <section\n        class=\"aihm-scene aihm-scene--deeplearning\"\n        id=\"ns-aih-OmCoiZk-6-deeplearning\"\n        data-aihm-scene=\"deeplearning\"\n        data-autoplay=\"1\"\n        style=\"--aihm-height:260px\"\n    >\n        <div class=\"aihm-frame\">\n            \t<div class=\"aihm-visual aihm-dl-preview\" role=\"img\" aria-label=\"Animated deep-learning process adapted from the 9Senses machine-learning explainer: data becomes vectors, trains a model, produces output, is checked and feeds back\">\n\t\t<canvas class=\"aihm-dl-canvas\" aria-hidden=\"true\"><\/canvas>\n\t\t<div class=\"aihm-dl-node aihm-dl-node--data\" data-dl-step=\"0\">data<\/div>\n\t\t<div class=\"aihm-dl-node aihm-dl-node--vectors\" data-dl-step=\"1\">vectors<\/div>\n\t\t<div class=\"aihm-dl-core\" data-dl-step=\"2\">model<\/div>\n\t\t<div class=\"aihm-dl-node aihm-dl-node--output\" data-dl-step=\"3\">output<\/div>\n\t\t<div class=\"aihm-dl-node aihm-dl-node--check\" data-dl-step=\"4\">check<\/div>\n\t\t<div class=\"aihm-dl-node aihm-dl-node--feedback\" data-dl-step=\"5\">feedback<\/div>\n\t<\/div>\n\t        <\/div>\n    <\/section>\n    \t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t<div class=\"ns-aih-slide-text\">\n\t\t\t\t\t\t\t<p class=\"ns-aih-slide-body\">The previous decade was the era where all previous efforts in making computers act &quot;intelligently&quot; came together, and where many breakthroughs shifted public attention towards the term &quot;Artificial Intelligence&quot; again, after it had been rarely used since the 1970s. By 2010, normal desktop and laptop computers were strong enough to perform AI tasks.<\/p>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t<section class=\"ns-aih-slide\" data-index=\"7\" aria-hidden=\"true\">\n\t\t\t\t\t\t<h3 class=\"ns-aih-slide-title\">Listening and Speaking<\/h3>\n\t\t\t\t\t\t<div class=\"ns-aih-slide-scene\">\n\t\t\t\t\t\t\t    <section\n        class=\"aihm-scene aihm-scene--language\"\n        id=\"ns-aih-OmCoiZk-7-language\"\n        data-aihm-scene=\"language\"\n        data-autoplay=\"1\"\n        style=\"--aihm-height:260px\"\n    >\n        <div class=\"aihm-frame\">\n            \t<div class=\"aihm-visual aihm-nlp-preview\" role=\"img\" aria-label=\"Animated natural-language-processing diagram adapted from the 9Senses NLP explainer: input becomes tokens, vectors and context, then output is checked\">\n\t\t<canvas class=\"aihm-nlp-canvas\" aria-hidden=\"true\"><\/canvas>\n\t\t<div class=\"aihm-nlp-tokenline\" aria-hidden=\"true\">\n\t\t\t<span>Can<\/span><span>we<\/span><span>ship<\/span><span>?<\/span>\n\t\t<\/div>\n\t\t<div class=\"aihm-nlp-node aihm-nlp-node--input\" data-nlp-step=\"0\">input<\/div>\n\t\t<div class=\"aihm-nlp-node aihm-nlp-node--tokens\" data-nlp-step=\"1\">tokens<\/div>\n\t\t<div class=\"aihm-nlp-node aihm-nlp-node--vectors\" data-nlp-step=\"2\">vectors<\/div>\n\t\t<div class=\"aihm-nlp-core\" data-nlp-step=\"3\">context<\/div>\n\t\t<div class=\"aihm-nlp-node aihm-nlp-node--output\" data-nlp-step=\"4\">output<\/div>\n\t\t<div class=\"aihm-nlp-node aihm-nlp-node--check\" data-nlp-step=\"5\">check<\/div>\n\t\t<div class=\"aihm-nlp-outputline\" aria-hidden=\"true\"><span>yes<\/span><span>\u2014<\/span><span>with<\/span><span>review<\/span><\/div>\n\t\t<div class=\"aihm-nlp-warning\" data-nlp-step=\"5\" aria-hidden=\"true\">confidence <b>\u2260<\/b> correctness<\/div>\n\t<\/div>\n\t        <\/div>\n    <\/section>\n    \t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t<div class=\"ns-aih-slide-text\">\n\t\t\t\t\t\t\t<p class=\"ns-aih-slide-body\">Finally, conversational AI is able to have conversations with humans based on Large Language Models that have been released. Those models are routinely able to pass the Turing Test, which means that they are able to understand and communicate back in natural language.<\/p>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t<\/div>\n\n\t\t<\/div>\n\t<\/div>\n\t<\/div><\/div>\n<\/div>\n<\/div>\n<\/div>\n\n<div class=\"et_pb_section_3 et_pb_section et_section_regular et_block_section\">\n<div class=\"et_pb_row_5 et_pb_row et_flex_row ns-cta\">\n<div class=\"et_pb_column_6 et_pb_column et_block_column et_pb_css_mix_blend_mode_passthrough et_flex_column_24_24 et_flex_column_24_24_tabletWide\">\n<div class=\"et_pb_text_4 et_pb_text et_pb_bg_layout_light et_pb_module et_block_module\"><div class=\"et_pb_text_inner\"><p>But what is AI really? We have asked two conversational AI systems about their definition of Artificial Intelligence and they came back with quite divergent answers. <a href=\"#AIonAI\">Click to see what AI has to say on AI<\/a><\/p>\n<\/div><\/div>\n\n<div class=\"et_pb_text_5 et_pb_text et_pb_bg_layout_light et_pb_module et_block_module\"><div class=\"et_pb_text_inner\"><p>With two differing statements from two AI systems, we are not afraid of creating our own answer. We at 9senses define AI as<strong> \"a computer system that is able to react to an event it has never experienced before in a meaningful way that is adequate to that event, based on the analysis of many similar events from data.\"<\/strong> This ability clearly distinguishes it from traditional computer logic where each event (or combination of events) has only one defined reaction. We explicitly stay away from comparing it with humans, because in some areas, computers <a href=\"\/why-9\">are still eons away from reaching our abilities, while in others, they massively outperform us<\/a>.<\/p>\n<\/div><\/div>\n<\/div>\n\n<div class=\"et_pb_column_7 et_pb_column et-last-child et_block_column et_pb_css_mix_blend_mode_passthrough et_flex_column_24_24 et_flex_column_24_24_tabletWide\">\n<div class=\"et_pb_module et_pb_button_module_wrapper et_pb_button_0_wrapper\"><a class=\"et_pb_button_0 et_pb_button et_pb_bg_layout_light et_pb_module et_block_module\" href=\"#AIonAI\" style=\"text-wrap:balance\">what AI says about AI<\/a><\/div>\n<\/div>\n<\/div>\n<\/div>\n\n<div class=\"et_pb_section_4 et_pb_section et_section_regular et_block_section ns-block ns-tabset\" id=\"Key_fields_of_AI\" lang=\"en\">\n<div class=\"et_pb_row_6 et_pb_row et_block_row\">\n<div class=\"et_pb_column_8 et_pb_column et_pb_column_4_4 et-last-child et_block_column et_pb_css_mix_blend_mode_passthrough\">\n<div class=\"et_pb_text_6 et_pb_text et_pb_bg_layout_light et_pb_module et_block_module\"><div class=\"et_pb_text_inner\"><h2 style=\"display: inline; margin: 0 0.6em 0 0; padding: 0;\">Key Fields of AI<\/h2>\n<p><span>There are various key AI technology areas, here is one of many ways to break them down:<\/span><\/p>\n<\/div><\/div>\n<\/div>\n<\/div>\n\n<div class=\"et_pb_row_7 et_pb_row et_flex_row\">\n<div class=\"et_pb_column_9 et_pb_column et_flex_column et_pb_css_mix_blend_mode_passthrough et_flex_column_24_24 et_flex_column_24_24_tabletWide hovergroup preset--module--divi-column--47yw920uxm\" id=\"trigger-ml\">\n<div class=\"et_pb_blurb_0 et_pb_blurb et_pb_bg_layout_light et_pb_blurb_position_top et_pb_module et_flex_module\"><div class=\"et_pb_blurb_content et_flex_module\"><div class=\"et_pb_blurb_container\"><h3 class=\"et_pb_module_header\">Ma\u00adchine Learn\u00ading<\/h3><div class=\"et_pb_blurb_description\"><p><span>Finding patterns in large datasets and drawing con\u00ad\u00ad\u00adclusions is at the core of most AI applications these days.\u00a0 Machine Lear\u00adn\u00ading provides the statistical methods to make it happen.<\/span><\/p>\n<\/div><\/div><\/div><\/div>\n\n<div class=\"et_pb_icon_0 et_pb_icon et_pb_module et_flex_module\"><span class=\"et_pb_icon_wrap\"><span class=\"et-pb-icon\">3<\/span><\/span><\/div>\n<\/div>\n\n<div class=\"et_pb_column_10 et_pb_column et_flex_column et_pb_css_mix_blend_mode_passthrough et_flex_column_24_24 et_flex_column_24_24_tabletWide hovergroup preset--module--divi-column--47yw920uxm\" id=\"trigger-nlp\">\n<div class=\"et_pb_blurb_1 et_pb_blurb et_pb_bg_layout_light et_pb_blurb_position_top et_pb_module et_flex_module\"><div class=\"et_pb_blurb_content et_flex_module\"><div class=\"et_pb_blurb_container\"><h3 class=\"et_pb_module_header\">Nat\u00adural Lan\u00adguage Pro\u00adcess\u00ading<\/h3><div class=\"et_pb_blurb_description\"><p>Being able to communicate with humans is one of the most recent key AI develop\u00adments that helps interact with computers, for example in customer-facing IT.<\/p>\n<\/div><\/div><\/div><\/div>\n\n<div class=\"et_pb_icon_1 et_pb_icon et_pb_module et_flex_module\"><span class=\"et_pb_icon_wrap\"><span class=\"et-pb-icon\">3<\/span><\/span><\/div>\n<\/div>\n\n<div class=\"et_pb_column_11 et_pb_column et_flex_column et_pb_css_mix_blend_mode_passthrough et_flex_column_24_24 et_flex_column_24_24_tabletWide hovergroup preset--module--divi-column--47yw920uxm\" id=\"trigger-cv\">\n<div class=\"et_pb_blurb_2 et_pb_blurb et_pb_bg_layout_light et_pb_blurb_position_top et_pb_module et_flex_module\"><div class=\"et_pb_blurb_content et_flex_module\"><div class=\"et_pb_blurb_container\"><h3 class=\"et_pb_module_header\">Com\u00adputer Vi\u00adsion<\/h3><div class=\"et_pb_blurb_description\"><p>Finding items and differences in still or moving imagery is something that computers excel at, for example when it comes to surveillance, irre\u00adgularity detection or simply - counting.<\/p>\n<\/div><\/div><\/div><\/div>\n\n<div class=\"et_pb_icon_2 et_pb_icon et_pb_module et_flex_module\"><span class=\"et_pb_icon_wrap\"><span class=\"et-pb-icon\">3<\/span><\/span><\/div>\n<\/div>\n\n<div class=\"et_pb_column_12 et_pb_column et-last-child et_flex_column et_pb_css_mix_blend_mode_passthrough et_flex_column_24_24 et_flex_column_24_24_tabletWide hovergroup preset--module--divi-column--47yw920uxm\" id=\"trigger-robotics\">\n<div class=\"et_pb_blurb_3 et_pb_blurb et_pb_bg_layout_light et_pb_blurb_position_top et_pb_module et_flex_module\"><div class=\"et_pb_blurb_content et_flex_module\"><div class=\"et_pb_blurb_container\"><h3 class=\"et_pb_module_header\">Ro\u00adbot\u00adics<\/h3><div class=\"et_pb_blurb_description\"><p>Creating autonomous sys\u00adtems that perform phy\u00adsical actions, like driving a vehicle based on controlling equipment using sensor in\u00adput and logic, is a key field of AI, albeit a difficult one.<\/p>\n<\/div><\/div><\/div><\/div>\n\n<div class=\"et_pb_icon_3 et_pb_icon et_pb_module et_flex_module\"><span class=\"et_pb_icon_wrap\"><span class=\"et-pb-icon\">3<\/span><\/span><\/div>\n<\/div>\n<\/div>\n<\/div>\n\n<div class=\"et_pb_section_5 et_pb_section et_section_regular et_flex_section ns-panel\" id=\"ml\" style=\"--ns-rail-w:264px; --ns-body-color:#D3D3D3;\">\n<div class=\"et_pb_row_8 et_pb_row et_flex_row\">\n<div class=\"et_pb_column_13 et_pb_column et_block_column et_pb_css_mix_blend_mode_passthrough et_flex_column_24_24 et_flex_column_24_24_tabletWide\">\n<div class=\"et_pb_text_7 et_pb_text et_pb_bg_layout_light et_pb_module et_flex_module\"><div class=\"et_pb_text_inner\"><p>Whichever AI field you look at, whether NLP, Computer Vision or Robotics, it is usually Machine Learning doing the actual work underneath: models learn from historical data and apply what they have learned to data they have never seen before. Or, as Wikipedia puts it: \"Machine Learning is a field of study in artificial intelligence concerned with the development of statistical algorithms that can learn from data and generalize to unseen data; and thus perform tasks without explicit instructions.\"<\/p>\n<\/div><\/div>\n\n<div class=\"et_pb_code_2 et_pb_code et_pb_module\"><div class=\"et_pb_code_inner\">\t<div\n\t\tid=\"ns-aix-1\"\n\t\tclass=\"ns-aix ns-aix-ml\"\n\t\tdata-topic=\"ml\"\n\t\tdata-flow=\"core\"\n\t\t\t\tdata-step=\"0\"\n\t\tdata-interval=\"9000\"\n\t\t\t\tdata-autoplay=\"1\"\n\t\trole=\"region\"\n\t\taria-label=\"Comment fonctionne le Machine Learning\"\n\t\tstyle=\"--ns-aix-accent:#58a7f9;--ns-aix-secondary:#2b6cb0\"\n\t>\n\t\t<header class=\"ns-aix-head\">\n\t\t\t\t\t\t<h3 class=\"ns-aix-title\">Comment fonctionne le Machine Learning<\/h3>\n\t\t\t<p class=\"ns-aix-intro\">Le Machine Learning n\u2019est pas une suite fig\u00e9e de r\u00e8gles logiques. Le syst\u00e8me rep\u00e8re des motifs utiles dans des exemples, les transforme en mod\u00e8le, puis applique ce mod\u00e8le \u00e0 de nouvelles situations.<\/p>\n\t\t\t\t\t\t\t\t<\/header>\n\n\t\t<div class=\"ns-aix-body\">\n\t\t\t<div class=\"ns-aix-copy\">\n\t\t\t\t<div class=\"ns-aix-progress\" aria-hidden=\"true\"><span><\/span><\/div>\n\t\t\t\t\t\t\t\t\t<div\n\t\t\t\t\t\tid=\"ns-aix-1-flow-panel-core\"\n\t\t\t\t\t\tclass=\"ns-aix-step-panel ns-on\"\n\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\taria-hidden=\"false\"\n\t\t\t\t\t\t\t\t\t\t\t>\n\t\t\t\t\t<ol class=\"ns-aix-steps ns-on\" data-flow=\"core\" role=\"list\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step ns-on\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\t\t\tdata-step=\"0\"\n\t\t\t\t\t\t\t\t\tdata-key=\"observe\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"true\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-1-panel-core-0\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">01<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Observer<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Des cas bruts entrent dans le syst\u00e8me.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\t\t\tdata-step=\"1\"\n\t\t\t\t\t\t\t\t\tdata-key=\"encode\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-1-panel-core-1\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">02<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Encoder<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Le sens devient g\u00e9om\u00e9trie.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\t\t\tdata-step=\"2\"\n\t\t\t\t\t\t\t\t\tdata-key=\"train\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-1-panel-core-2\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">03<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Entra\u00eener<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Pr\u00e9dire, mesurer l\u2019erreur, ajuster.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\t\t\tdata-step=\"3\"\n\t\t\t\t\t\t\t\t\tdata-key=\"generalize\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-1-panel-core-3\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">04<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">G\u00e9n\u00e9raliser<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">La structure plut\u00f4t que la m\u00e9morisation.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\t\t\tdata-step=\"4\"\n\t\t\t\t\t\t\t\t\tdata-key=\"infer\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-1-panel-core-4\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">05<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Appliquer<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Une nouvelle entr\u00e9e devient une aide \u00e0 la d\u00e9cision.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\t\t\tdata-step=\"5\"\n\t\t\t\t\t\t\t\t\tdata-key=\"validate\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-1-panel-core-5\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">06<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Valider<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Le mod\u00e8le doit \u00eatre test\u00e9 en dehors de l\u2019entra\u00eenement.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\t\t\tdata-step=\"6\"\n\t\t\t\t\t\t\t\t\tdata-key=\"improve\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-1-panel-core-6\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">07<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Am\u00e9liorer<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Les retours transforment l\u2019exploitation en apprentissage.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t<\/ol>\n\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\n\t\t\t<div class=\"ns-aix-stage\" aria-label=\"Animated vector-map process diagram\">\n\t\t\t\t<canvas class=\"ns-aix-canvas\" aria-hidden=\"true\"><\/canvas>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node ns-on\" data-flow=\"core\" data-step=\"0\" data-key=\"observe\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>donn\u00e9es<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node\" data-flow=\"core\" data-step=\"1\" data-key=\"encode\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>vecteurs<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node\" data-flow=\"core\" data-step=\"2\" data-key=\"train\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>mod\u00e8le<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node\" data-flow=\"core\" data-step=\"3\" data-key=\"generalize\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>structure<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node\" data-flow=\"core\" data-step=\"4\" data-key=\"infer\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>sortie<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node\" data-flow=\"core\" data-step=\"5\" data-key=\"validate\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>v\u00e9rifier<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node\" data-flow=\"core\" data-step=\"6\" data-key=\"improve\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>retour<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\n\t\t\t\t\t<\/div>\n\n\t\t<div class=\"ns-aix-controls\">\n\t\t\t<div class=\"ns-aix-dots\" role=\"navigation\" aria-label=\"Step navigation\">\n\t\t\t\t\t\t\t\t\t<span\n\t\t\t\t\t\tclass=\"ns-aix-dotset ns-on\"\n\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\trole=\"group\"\n\t\t\t\t\t\taria-label=\"Steps for Comment fonctionne le Machine Learning\"\n\t\t\t\t\t\taria-hidden=\"false\"\n\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot ns-on\"\n\t\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\t\tdata-step=\"0\"\n\t\t\t\t\t\t\t\taria-label=\"Step 1: Observer\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-1-panel-core-0\"\n\t\t\t\t\t\t\t\taria-pressed=\"true\"\n\t\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\t\tdata-step=\"1\"\n\t\t\t\t\t\t\t\taria-label=\"Step 2: Encoder\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-1-panel-core-1\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\t\tdata-step=\"2\"\n\t\t\t\t\t\t\t\taria-label=\"Step 3: Entra\u00eener\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-1-panel-core-2\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\t\tdata-step=\"3\"\n\t\t\t\t\t\t\t\taria-label=\"Step 4: G\u00e9n\u00e9raliser\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-1-panel-core-3\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\t\tdata-step=\"4\"\n\t\t\t\t\t\t\t\taria-label=\"Step 5: Appliquer\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-1-panel-core-4\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\t\tdata-step=\"5\"\n\t\t\t\t\t\t\t\taria-label=\"Step 6: Valider\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-1-panel-core-5\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\t\tdata-step=\"6\"\n\t\t\t\t\t\t\t\taria-label=\"Step 7: Am\u00e9liorer\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-1-panel-core-6\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t<\/div>\n\t\t<\/div>\n\n\t\t<div class=\"ns-aix-below\" aria-live=\"polite\">\n\t\t\t<div class=\"ns-aix-explain\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-1-panel-core-0\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel ns-on\"\n\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\tdata-step=\"0\"\n\t\t\t\t\t\t\taria-hidden=\"false\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">01<\/span>Commencer par des exemples<\/h4>\n\t\t\t\t\t\t\t<p>Le Machine Learning commence par des exemples : documents, dossiers, valeurs de capteurs, conversations avec des clients ou \u00e9v\u00e9nements m\u00e9tier. La question d\u00e9cisive n\u2019est pas seulement la quantit\u00e9, mais la capacit\u00e9 des donn\u00e9es \u00e0 repr\u00e9senter les d\u00e9cisions que le mod\u00e8le devra ensuite soutenir \u2014 et si leur collecte et leur nettoyage en valent l\u2019effort.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-1-panel-core-1\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\tdata-step=\"1\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">02<\/span>Transformer les signaux en vecteurs<\/h4>\n\t\t\t\t\t\t\t<p>Textes, images et nombres sont convertis en caract\u00e9ristiques num\u00e9riques ou en vecteurs : des coordonn\u00e9es qui conservent des motifs utiles comme la similarit\u00e9, la fr\u00e9quence, l\u2019intention, le contexte ou le risque. C\u2019est ce qui permet \u00e0 un mod\u00e8le de comparer des \u00e9l\u00e9ments qui ne sont pas litt\u00e9ralement identiques.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-1-panel-core-2\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\tdata-step=\"2\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">03<\/span>Ajuster le mod\u00e8le<\/h4>\n\t\t\t\t\t\t\t<p>Le mod\u00e8le fait une pr\u00e9diction, la compare au r\u00e9sultat attendu, mesure l\u2019erreur et ajuste ses poids internes. Certains mod\u00e8les apprennent ainsi \u00e0 partir d\u2019exemples \u00e9tiquet\u00e9s ; d\u2019autres trouvent des structures sans \u00e9tiquettes, tandis que les grands mod\u00e8les de langage apprennent en grande partie directement \u00e0 partir de texte brut. \u00c0 force de r\u00e9p\u00e9tition, des exemples individuels deviennent un motif r\u00e9utilisable.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-1-panel-core-3\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\tdata-step=\"3\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">04<\/span>Apprendre une structure, pas une liste<\/h4>\n\t\t\t\t\t\t\t<p>Un mod\u00e8le utile ne m\u00e9morise pas simplement ses cas d\u2019entra\u00eenement \u2014 cette d\u00e9faillance s\u2019appelle l\u2019overfitting. Il apprend suffisamment de structure pour bien fonctionner sur de nouveaux cas proches sur le fond, m\u00eame si les mots, le format ou le contexte diff\u00e8rent.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-1-panel-core-4\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\tdata-step=\"4\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">05<\/span>Appliquer le mod\u00e8le entra\u00een\u00e9<\/h4>\n\t\t\t\t\t\t\t<p>Lors de l\u2019inf\u00e9rence, une nouvelle entr\u00e9e traverse le mod\u00e8le entra\u00een\u00e9. La sortie peut \u00eatre un score, une classification, une recommandation, une pr\u00e9vision, une r\u00e9ponse g\u00e9n\u00e9r\u00e9e ou une liste de documents class\u00e9s.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-1-panel-core-5\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\tdata-step=\"5\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">06<\/span>Ma\u00eetriser la qualit\u00e9 et le risque<\/h4>\n\t\t\t\t\t\t\t<p>Des donn\u00e9es de test s\u00e9par\u00e9es, une revue humaine et le suivi permettent de rep\u00e9rer les cas limites, les biais, les hallucinations, la d\u00e9rive et les risques de conformit\u00e9 avant que l\u2019automatisation n\u2019affecte des clients, des collaborateurs ou des d\u00e9cisions r\u00e9glement\u00e9es.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-1-panel-core-6\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\tdata-step=\"6\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">07<\/span>Boucler le retour d\u2019exp\u00e9rience<\/h4>\n\t\t\t\t\t\t\t<p>Les r\u00e9sultats r\u00e9els et les corrections reviennent dans le syst\u00e8me. Mal g\u00e9r\u00e9, ce m\u00e9canisme peut renforcer d\u2019anciens biais. Le mod\u00e8le, la couche de retrieval, les prompts, la cha\u00eene de donn\u00e9es et les r\u00e8gles de gouvernance doivent donc \u00eatre am\u00e9lior\u00e9s d\u00e9lib\u00e9r\u00e9ment plut\u00f4t que laiss\u00e9s \u00e0 la d\u00e9rive.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t<div class=\"ns-aix-aspects\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset ns-on\" data-flow=\"core\" data-step=\"0\" aria-hidden=\"false\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Concept cl\u00e9<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>L\u2019apprentissage supervis\u00e9 travaille \u00e0 partir d\u2019exemples \u00e9tiquet\u00e9s ; l\u2019apprentissage non supervis\u00e9 trouve des structures dans des donn\u00e9es non \u00e9tiquet\u00e9es ; l\u2019apprentissage par renforcement progresse par essais, erreurs et retours.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>En pratique<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Pour pr\u00e9voir les pannes d\u2019une machine, les donn\u00e9es d\u2019exploitation sont rapproch\u00e9es des pannes enregistr\u00e9es afin de fournir au syst\u00e8me des exemples \u00e0 partir desquels apprendre.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"core\" data-step=\"1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Concept cl\u00e9<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Le Machine Learning est avant tout une reconnaissance de motifs dans des ensembles de donn\u00e9es souvent volumineux. L\u2019encodage rend ces motifs mesurables.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>En pratique<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Les repr\u00e9sentations vectorielles permettent aux syst\u00e8mes de recherche, de recommandation et de retrieval de comparer des \u00e9l\u00e9ments similaires sur le fond plut\u00f4t qu\u2019identiques dans leur forme.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"core\" data-step=\"2\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Contexte<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Le Machine Learning remonte aux ann\u00e9es 1950, lorsque des chercheurs ont con\u00e7u des algorithmes capables de reconna\u00eetre des motifs simples dans les donn\u00e9es. Les grands volumes de donn\u00e9es, le mat\u00e9riel performant \u2014 notamment les GPU \u2014 et de meilleurs algorithmes ont ensuite port\u00e9 les avanc\u00e9es du d\u00e9but du XXIe si\u00e8cle.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Comment \u00e7a marche<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Pendant l\u2019entra\u00eenement, l\u2019algorithme produit des pr\u00e9dictions et les compare \u00e0 des r\u00e9sultats connus. R\u00e9p\u00e9t\u00e9 de nombreuses fois, ce processus am\u00e9liore progressivement la pr\u00e9cision.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"core\" data-step=\"3\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Concept cl\u00e9<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Le Machine Learning moderne repose souvent sur des r\u00e9seaux de neurones artificiels \u2014 des syst\u00e8mes math\u00e9matiques librement inspir\u00e9s du cerveau qui s\u2019am\u00e9liorent en apprenant de l\u2019exp\u00e9rience.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Pourquoi c\u2019est important<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Le mod\u00e8le termin\u00e9 doit fonctionner sur de nouvelles donn\u00e9es du m\u00eame type, qu\u2019il n\u2019a jamais vues, et pas simplement reproduire les cas sur lesquels il a \u00e9t\u00e9 entra\u00een\u00e9.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"core\" data-step=\"4\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Ce que le Machine Learning peut faire<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Le Machine Learning alimente moteurs de recherche, syst\u00e8mes de recommandation, assistants vocaux, d\u00e9tection de fraude, analyse d\u2019images et syst\u00e8mes autonomes \u2014 et se montre, dans de nombreuses t\u00e2ches bien d\u00e9limit\u00e9es, plus rapide et plus pr\u00e9cis que l\u2019humain.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>En pratique<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Il peut rep\u00e9rer des tumeurs sur des images m\u00e9dicales, pr\u00e9voir des pannes d\u2019\u00e9quipement, traduire des langues et g\u00e9n\u00e9rer du texte, des images ou du code.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"core\" data-step=\"5\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Pourquoi c\u2019est important<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Un mod\u00e8le ne comprenant pas r\u00e9ellement le contexte \u00e0 un niveau abstrait, il lui faut des consignes claires pendant l\u2019entra\u00eenement et les revues, ainsi que des limites fix\u00e9es par des humains.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Attention<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Un mod\u00e8le opaque peut devenir une bo\u00eete noire : on ne sait plus clairement quels motifs d\u00e9terminent ses d\u00e9cisions. Lorsque les cons\u00e9quences sont importantes, une gouvernance rigoureuse est indispensable.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"core\" data-step=\"6\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Ce que le Machine Learning peut faire<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Le Machine Learning est particuli\u00e8rement fort lorsque de grands volumes de donn\u00e9es structur\u00e9es ou non structur\u00e9es cachent des motifs trop subtils pour une analyse manuelle : d\u00e9couverte de m\u00e9dicaments et mod\u00e9lisation climatique dans la recherche ; pr\u00e9visions, segmentation et optimisation des processus dans l\u2019entreprise.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Attention<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Les r\u00e9sultats d\u00e9pendent fortement de la qualit\u00e9 des donn\u00e9es : de mauvaises donn\u00e9es donnent de mauvais r\u00e9sultats, et des boucles de r\u00e9troaction mal ma\u00eetris\u00e9es renforcent discr\u00e8tement les biais d\u2019hier.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t<\/div>\n\t<\/div><\/div>\n<\/div>\n\n<div class=\"et_pb_column_14 et_pb_column et-last-child et_block_column et_pb_css_mix_blend_mode_passthrough et_flex_column_24_24 et_flex_column_24_24_tablet et_flex_column_24_24_phone et_flex_column_24_24_tabletWide\">\n<div class=\"et_pb_text_8 et_pb_text et_pb_bg_layout_light et_pb_module et_flex_module\"><div class=\"et_pb_text_inner\"><h3>The Challenges<\/h3>\n<p>Machine Learning has important limitations. It heavily depends on the quality and quantity of the underlying data to find the relevant statistical patterns. Bad data leads to incorrect outcomes.<\/p>\n<p>As machine learning - like all \"AI\" - doesn't really understand the context on an abstract level, it needs clear guidance during training and reviews, often coupled with additional boundaries set by humans.<\/p>\n<p>This becomes particularly difficult when the model itself is intransparent and becomes a \"black box\", where it remains unclear what patterns drive decisions made by ML. Thus, careful governance is essential, particularly in areas with high impact.<\/p>\n<\/div><\/div>\n<\/div>\n<\/div>\n\n<div class=\"et_pb_row_9 et_pb_row et_flex_row\" style=\"--ns-caro-top:26px\">\n<div class=\"et_pb_column_15 et_pb_column et-last-child et_flex_column et_pb_css_mix_blend_mode_passthrough et_flex_column_24_24 et_flex_column_24_24_tablet et_flex_column_24_24_phone et_flex_column_24_24_tabletWide\">\n<div class=\"et_pb_code_3 et_pb_code et_pb_module\"><div class=\"et_pb_code_inner\"><div class=\"nsp-frame nsp-frame--transparent\"><div class=\"nsp-frame-head\"><h2>Featured Projects<\/h2><\/div><div class=\"nsp-carousel-wrap\" data-base-width=\"240\" data-loop=\"1\"><button type=\"button\" class=\"nsp-nav nsp-prev\" aria-label=\"Previous\"><span class=\"nsp-nav-icon et-pb-icon\" aria-hidden=\"true\">&#x34;<\/span><\/button><div class=\"nsp-carousel\">\t\t<article class=\"nsp-card\" tabindex=\"0\" role=\"button\" data-id=\"12\" aria-label=\"Intelligent Data Retrieval Agent\">\n\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-card-iconwrap\"><span class=\"nsp-card-icon et-pb-icon\" aria-hidden=\"true\">&#xe0f7;<\/span><\/div>\n\t\t\t\t\t\t<h3 class=\"nsp-card-title\">Intelligent Data Retrieval Agent<\/h3>\n\t\t\t\t\t\t\t\t\t\t<p class=\"nsp-card-excerpt\">Production-ready AI retrieval system using LLMs and semantic search to transform fragmented data into reliable, searchable knowledge.<\/p>\n\t\t\t\t\t\t\t\t\t\t<span class=\"nsp-card-more\">read more\u2026<\/span>\n\t\t\t\t\t<\/article>\n\t\t\t\t<article class=\"nsp-card\" tabindex=\"0\" role=\"button\" data-id=\"13\" aria-label=\"Visual Search Recommendations\">\n\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-card-iconwrap\"><i class=\"nsp-card-icon fa-solid fa-socks\" aria-hidden=\"true\"><\/i><\/div>\n\t\t\t\t\t\t<h3 class=\"nsp-card-title\">Visual Search Recommendations<\/h3>\n\t\t\t\t\t\t\t\t\t\t<p class=\"nsp-card-excerpt\">eCommmerce plugin that enables searching for visually similar products, helping customers to find and compare multiple related items.<\/p>\n\t\t\t\t\t\t\t\t\t\t<span class=\"nsp-card-more\">read more\u2026<\/span>\n\t\t\t\t\t<\/article>\n\t\t\t\t<article class=\"nsp-card\" tabindex=\"0\" role=\"button\" data-id=\"14\" aria-label=\"Visual Assistance for Seniors\">\n\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-card-iconwrap\"><i class=\"nsp-card-icon fa-solid fa-glasses\" aria-hidden=\"true\"><\/i><\/div>\n\t\t\t\t\t\t<h3 class=\"nsp-card-title\">Visual Assistance for Seniors<\/h3>\n\t\t\t\t\t\t\t\t\t\t<p class=\"nsp-card-excerpt\">Visual Assistance App: enabling visual assistance for seniors by helping position determination using Computer Vision and Deep Learning.<\/p>\n\t\t\t\t\t\t\t\t\t\t<span class=\"nsp-card-more\">read more\u2026<\/span>\n\t\t\t\t\t<\/article>\n\t\t\t\t<article class=\"nsp-card\" tabindex=\"0\" role=\"button\" data-id=\"1\" aria-label=\"Electrical Switch Monitor\">\n\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-card-iconwrap\"><i class=\"nsp-card-icon fa-solid fa-toggle-on\" aria-hidden=\"true\"><\/i><\/div>\n\t\t\t\t\t\t<h3 class=\"nsp-card-title\">Electrical Switch Monitor<\/h3>\n\t\t\t\t\t\t\t\t\t\t<p class=\"nsp-card-excerpt\">Public transportation: using AI-driven vision to monitor old-fashioned electrical relays and also to evaluate potential failures for predictive maintenance.<\/p>\n\t\t\t\t\t\t\t\t\t\t<span class=\"nsp-card-more\">read more\u2026<\/span>\n\t\t\t\t\t<\/article>\n\t\t\t\t<article class=\"nsp-card\" tabindex=\"0\" role=\"button\" data-id=\"5\" aria-label=\"Hydropower Plant Operations\">\n\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-card-iconwrap\"><i class=\"nsp-card-icon fa-solid fa-water\" aria-hidden=\"true\"><\/i><\/div>\n\t\t\t\t\t\t<h3 class=\"nsp-card-title\">Hydropower Plant Operations<\/h3>\n\t\t\t\t\t\t\t\t\t\t<p class=\"nsp-card-excerpt\">Create a control and monitoring solution for all plant operations, including predictive maintenance logic and intrusion mon\u00aditoring.<\/p>\n\t\t\t\t\t\t\t\t\t\t<span class=\"nsp-card-more\">read more\u2026<\/span>\n\t\t\t\t\t<\/article>\n\t\t\t\t<article class=\"nsp-card\" tabindex=\"0\" role=\"button\" data-id=\"2\" aria-label=\"Customer Interaction Analysis\">\n\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-card-iconwrap\"><span class=\"nsp-card-icon et-pb-icon\" aria-hidden=\"true\">&#x77;<\/span><\/div>\n\t\t\t\t\t\t<h3 class=\"nsp-card-title\">Customer Interaction Analysis<\/h3>\n\t\t\t\t\t\t\t\t\t\t<p class=\"nsp-card-excerpt\">SaaS project platform: the objective was to evaluate dialogue quality using an AI model to ensure timely intervention and customer care.<\/p>\n\t\t\t\t\t\t\t\t\t\t<span class=\"nsp-card-more\">read more\u2026<\/span>\n\t\t\t\t\t<\/article>\n\t\t\t\t<article class=\"nsp-card\" tabindex=\"0\" role=\"button\" data-id=\"6\" aria-label=\"Motion-sensitive wearables\">\n\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-card-iconwrap\"><i class=\"nsp-card-icon fa-solid fa-shirt\" aria-hidden=\"true\"><\/i><\/div>\n\t\t\t\t\t\t<h3 class=\"nsp-card-title\">Motion-sensitive wearables<\/h3>\n\t\t\t\t\t\t\t\t\t\t<p class=\"nsp-card-excerpt\">Wearable fabric-based devices with embedded microcontrollers and sensitivity for motion, heartbeat, body temperature and sweat detection.<\/p>\n\t\t\t\t\t\t\t\t\t\t<span class=\"nsp-card-more\">read more\u2026<\/span>\n\t\t\t\t\t<\/article>\n\t\t<\/div><button type=\"button\" class=\"nsp-nav nsp-next\" aria-label=\"Next\"><span class=\"nsp-nav-icon et-pb-icon\" aria-hidden=\"true\">&#x35;<\/span><\/button><\/div><script>(function(){\n\t\t\tif (!window.nspComputeLayout) {\n\t\t\t\twindow.NSP_BASE_CARD_WIDTH = 240;\n\t\t\t\twindow.nspIsFluidWidth = function(){ return window.innerWidth > 640; };\n\t\t\t\twindow.nspComputeLayout = function(available, gap, totalCards, baseWidth){\n\t\t\t\t\tvar BASE = baseWidth || window.NSP_BASE_CARD_WIDTH;\n\t\t\t\t\tavailable = Math.max(0, available || 0);\n\t\t\t\t\tgap = Math.max(0, gap || 0);\n\t\t\t\t\ttotalCards = Math.max(0, totalCards || 0);\n\t\t\t\t\tif (!totalCards){\n\t\t\t\t\t\treturn { count: 0, width: BASE, center: true };\n\t\t\t\t\t}\n\t\t\t\t\tvar per = BASE + gap;\n\t\t\t\t\tvar baseCount = Math.max(1, Math.floor((available + gap) \/ per));\n\t\t\t\t\tvar used = baseCount * BASE + (baseCount - 1) * gap;\n\t\t\t\t\tvar remainder = available - used;\n\t\t\t\t\t\/\/ Only create an extra visible slot when there is actually another card to fill it.\n\t\t\t\t\t\/\/ If exactly the base-width row count is present, distribute those cards evenly\n\t\t\t\t\t\/\/ across the row instead of leaving them beside a phantom extra slot.\n\t\t\t\t\tvar fullCount = (totalCards > baseCount && remainder >= (2 \/ 3) * BASE) ? baseCount + 1 : baseCount;\n\t\t\t\t\tfullCount = Math.max(1, fullCount);\n\t\t\t\t\tvar fullWidth = (available - (fullCount - 1) * gap) \/ fullCount;\n\t\t\t\t\tif (!isFinite(fullWidth) || fullWidth <= 0){ fullWidth = BASE; }\n\n\t\t\t\t\t\/\/ One or two cards should never grow just because the row is wide.\n\t\t\t\t\t\/\/ They keep the base width, sit at the left, and leave the rest of the row empty.\n\t\t\t\t\tif (totalCards < 3){\n\t\t\t\t\t\tvar smallMax = (available - (totalCards - 1) * gap) \/ totalCards;\n\t\t\t\t\t\tvar smallWidth = (isFinite(smallMax) && smallMax > 0) ? Math.min(BASE, smallMax) : BASE;\n\t\t\t\t\t\treturn { count: totalCards, width: smallWidth, center: false };\n\t\t\t\t\t}\n\n\t\t\t\t\t\/\/ If there are fewer cards than a full row would hold, keep the full-row\n\t\t\t\t\t\/\/ card width rather than stretching those few across the whole row, and\n\t\t\t\t\t\/\/ leave the remaining slots empty. Left-aligned, as the master grid is.\n\t\t\t\t\tif (totalCards < fullCount){\n\t\t\t\t\t\treturn { count: totalCards, width: fullWidth, center: false };\n\t\t\t\t\t}\n\n\t\t\t\t\tvar count = Math.min(fullCount, totalCards);\n\t\t\t\t\treturn { count: count, width: fullWidth, center: false };\n\t\t\t\t};\n\t\t\t\t\/\/ Centre the chevrons on the media block of a card rather than on\n\t\t\t\t\/\/ the whole card: on people cards that is the square photo\n\t\t\t\t\/\/ (whose height tracks the fluid card width, so it cannot be\n\t\t\t\t\/\/ expressed in CSS), on project cards the icon block. Published\n\t\t\t\t\/\/ as a custom property the stylesheet reads.\n\t\t\t\twindow.nspTrackPadX = function(track){\n\t\t\t\t\tvar cs = window.getComputedStyle(track);\n\t\t\t\t\treturn (parseFloat(cs.paddingLeft) || 0) + (parseFloat(cs.paddingRight) || 0);\n\t\t\t\t};\n\t\t\t\twindow.nspSyncNav = function(wrap){\n\t\t\t\t\t\/\/ People cards anchor on the photo, which is the visual mass of\n\t\t\t\t\t\/\/ the card. Project cards have no photo -- only a small icon\n\t\t\t\t\t\/\/ at the top -- so they anchor on the card panel itself and\n\t\t\t\t\t\/\/ the arrows land on its vertical centre. Measuring against\n\t\t\t\t\t\/\/ the wrap in viewport coordinates absorbs the track padding\n\t\t\t\t\t\/\/ and card margins without restating any of them here.\n\t\t\t\t\tvar media = wrap.querySelector(\".nsp-card-photo\")\n\t\t\t\t\t\t|| wrap.querySelector(\".nsp-card\");\n\t\t\t\t\tif (!media){ return; }\n\t\t\t\t\tvar r = media.getBoundingClientRect();\n\t\t\t\t\tif (!r.height){ return; }\n\t\t\t\t\tvar w = wrap.getBoundingClientRect();\n\t\t\t\t\twrap.style.setProperty(\"--nsp-nav-top\", ((r.top - w.top) + r.height \/ 2) + \"px\");\n\t\t\t\t};\n\t\t\t}\n\t\t\tvar wrap = document.currentScript.previousElementSibling;\n\t\t\tif(!wrap){ return; }\n\t\t\t\/\/ The frame carries the visible panel, so it has to be revealed\n\t\t\t\/\/ together with the wrap -- otherwise an empty framed box paints\n\t\t\t\/\/ before the cards have been sized into it.\n\t\t\tfunction reveal(w){\n\t\t\t\tw.style.visibility = \"visible\";\n\t\t\t\tvar f = w.parentNode;\n\t\t\t\tif (f && f.classList && f.classList.contains(\"nsp-frame\")){ f.style.visibility = \"visible\"; }\n\t\t\t}\n\t\t\tvar baseWidth = parseInt(wrap.getAttribute(\"data-base-width\"), 10) || window.NSP_BASE_CARD_WIDTH;\n\t\t\tvar track = wrap.querySelector(\".nsp-carousel\");\n\t\t\tvar cards = track ? track.querySelectorAll(\".nsp-card\") : [];\n\t\t\tif(!track || !cards.length){ if (window.nspSyncNav){ window.nspSyncNav(wrap); }\n\t\t\treveal(wrap); return; }\n\t\t\t\/\/ Read from the stylesheet rather than restating it: a hard-coded copy\n\t\t\t\/\/ here is exactly how the gutter drifted from the CSS before.\n\t\t\tvar gap = parseFloat(window.getComputedStyle(track).columnGap) || 24;\n\t\t\tif (!window.nspIsFluidWidth()) {\n\t\t\t\t\/\/ Mobile: unchanged -- fixed per-card width (CSS 82vw), just avoid a partial\n\t\t\t\t\/\/ card peeking out at the wrap edge.\n\t\t\t\tvar cardW = cards[0].getBoundingClientRect().width;\n\t\t\t\tif(!cardW){ if (window.nspSyncNav){ window.nspSyncNav(wrap); }\n\t\t\treveal(wrap); return; }\n\t\t\t\t\/\/ clientWidth is the PADDING box, so the track padding has to come\n\t\t\t\t\/\/ off before this is the room the cards actually get.\n\t\t\t\tvar available0 = track.clientWidth - window.nspTrackPadX(track);\n\t\t\t\tvar fitCount = Math.max(1, Math.floor((available0 + gap) \/ (cardW + gap)));\n\t\t\t\tvar count0 = Math.min(fitCount, cards.length);\n\t\t\t\tvar needed0 = count0 * cardW + (count0 - 1) * gap;\n\t\t\t\tvar ws0 = window.getComputedStyle(wrap);\n\t\t\t\tvar padX0 = (parseFloat(ws0.paddingLeft) || 0) + (parseFloat(ws0.paddingRight) || 0);\n\t\t\t\tvar target0 = needed0 + window.nspTrackPadX(track) + (ws0.boxSizing === \"border-box\" ? padX0 : 0);\n\t\t\t\twrap.style.maxWidth = Math.ceil(target0) + \"px\";\n\t\t\t\tif (window.nspSyncNav){ window.nspSyncNav(wrap); }\n\t\t\treveal(wrap);\n\t\t\t\treturn;\n\t\t\t}\n\t\t\t\/\/ Desktop\/tablet: fluid card width, filling the row exactly either way.\n\t\t\t\/\/ clientWidth is the PADDING box -- take the track padding off it.\n\t\t\tvar available = track.clientWidth - window.nspTrackPadX(track);\n\t\t\tvar layout = window.nspComputeLayout(available, gap, cards.length, baseWidth);\n\t\t\tfor (var i = 0; i < cards.length; i++){ cards[i].style.flex = \"0 0 \" + layout.width + \"px\"; }\n\t\t\tif (layout.center) {\n\t\t\t\tvar totalW = layout.count * layout.width + (layout.count - 1) * gap;\n\t\t\t\twrap.style.maxWidth = Math.ceil(totalW + window.nspTrackPadX(track)) + \"px\";\n\t\t\t\twrap.style.marginLeft = \"auto\";\n\t\t\t\twrap.style.marginRight = \"auto\";\n\t\t\t}\n\t\t\tif (window.nspSyncNav){ window.nspSyncNav(wrap); }\n\t\t\treveal(wrap);\n\t\t})();<\/script><\/div><noscript><style>.nsp-carousel-wrap,.nsp-frame{visibility:visible !important;}<\/style><\/noscript><div class=\"nsp-popups\" hidden><template class=\"nsp-popup-tpl\" data-id=\"12\">\t\t<div class=\"nsp-dialog no-image\">\n\t\t\t<button type=\"button\" class=\"nsp-close\" aria-label=\"Close\">&times;<\/button>\n\t\t\t<div class=\"nsp-dlg-h\">\n\t\t\t\t<span class=\"nsp-pop-icon et-pb-icon\" aria-hidden=\"true\">&#xe0f7;<\/span>\t\t\t\t<span class=\"nsp-dlg-h-title\">Intelligent Data Retrieval Agent<\/span>\n\t\t\t<\/div>\n\t\t\t<div class=\"nsp-pop-cols\">\n\t\t\t\t<div class=\"nsp-pop-main\">\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-desc\"><p>Development of a production-ready retrieval agent for querying large, fragmented, and undocumented enterprise data repositories using large language models and semantic search. The solution transformed scattered legacy records into structured, searchable knowledge, enabling users to retrieve relevant information within seconds instead of manually searching across hundreds of documents.<\/p>\n<p>The project combined retrieval-augmented generation (RAG), vector search, and modern LLM technologies to deliver reliable, context-aware information retrieval. Designed with a modular architecture, the system supports scalability, maintainability, and future extensions while ensuring robust retrieval quality across heterogeneous data sources.<\/p>\n<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-role nsp-pop-role--below\"><p>AI Engineer<\/p>\n<p>Python \u2022 LangChain \u2022 OpenAI \u2022 Qdrant \u2022 Semantic Search \u2022 RAG \u2022 Streamlit<\/p>\n<\/div>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\n\t\t\t\t\t\t\t<ul class=\"nsp-chiplist nsp-pop-caps\">\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/machine-learning\">Machine Learning<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/nlp\">Natural Language Processing<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/control-and-automation\">Automation<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t<\/ul>\n\t\t\t\t\t<\/div>\n\t\t<\/template><template class=\"nsp-popup-tpl\" data-id=\"13\">\t\t<div class=\"nsp-dialog has-image\">\n\t\t\t<button type=\"button\" class=\"nsp-close\" aria-label=\"Close\">&times;<\/button>\n\t\t\t<div class=\"nsp-dlg-h\">\n\t\t\t\t\t\t\t\t<span class=\"nsp-dlg-h-title\">Visual Search Recommendations<\/span>\n\t\t\t<\/div>\n\t\t\t<div class=\"nsp-pop-cols\">\n\t\t\t\t<div class=\"nsp-pop-main\">\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-desc\"><p>For VisualSearch, the focus was on developing and deploying e-commerce plugins to enhance recommendation functionalities in web shops. The plugins were successfully launched in the store and happily adopted by customers, demonstrating their practical use in improving e-commerce experiences. Key elements included:<br \/>\n- computation of visual embeddings from appearances of e-commerce products<br \/>\n- building and maintaining a search index using these embeddings<br \/>\n- providing a cloud-based API for Shopware and Prestashop plugins<\/p>\n<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-aside\">\n\t\t\t\t\t\t<img loading=\"lazy\" decoding=\"async\" width=\"369\" height=\"455\" src=\"https:\/\/www.9senses.ai\/wp-content\/uploads\/2026\/07\/visualsearch.png\" class=\"nsp-pop-img\" alt=\"\" srcset=\"https:\/\/www.9senses.ai\/wp-content\/uploads\/2026\/07\/visualsearch.png 369w, https:\/\/www.9senses.ai\/wp-content\/uploads\/2026\/07\/visualsearch-243x300.png 243w, https:\/\/www.9senses.ai\/wp-content\/uploads\/2026\/07\/visualsearch-10x12.png 10w\" sizes=\"(max-width: 369px) 100vw, 369px\" \/>\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-role nsp-pop-role--aside\"><p>Peter acted as the project manager and key data scientist on this project, leveraging AWS services, Python, and deep learning frameworks like Keras. The project integrated cloud-based solutions using CloudFormation, Lambda, and Gateway for scalable and efficient deployment, using SQL and DynamoDB for data management.<\/p>\n<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\n\t\t\t\t\t\t\t<ul class=\"nsp-chiplist nsp-pop-caps\">\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/machine-learning\">Machine Learning<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/vision\">Computer Vision<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t<\/ul>\n\t\t\t\t\t<\/div>\n\t\t<\/template><template class=\"nsp-popup-tpl\" data-id=\"14\">\t\t<div class=\"nsp-dialog has-image\">\n\t\t\t<button type=\"button\" class=\"nsp-close\" aria-label=\"Close\">&times;<\/button>\n\t\t\t<div class=\"nsp-dlg-h\">\n\t\t\t\t\t\t\t\t<span class=\"nsp-dlg-h-title\">Visual Assistance for Seniors<\/span>\n\t\t\t<\/div>\n\t\t\t<div class=\"nsp-pop-cols\">\n\t\t\t\t<div class=\"nsp-pop-main\">\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-desc\"><p>For the Austrian Institute of Technology (as a part of the national research project LARAH), a prototype of an assistance system for visual indoor localization for disabled and elderly persons was developed. By leveraging Computer Vision and Deep Learning, innovative algorithms for position determination were implemented. The project resulted in a functional prototype, including two Android applications for real-time localization. Key elements included:<br \/>\n- visual recognition of persons using Deep-Learning models<br \/>\n- visual reconstruction and localization of indoor environments using Structure-from-Motion and Machine Learning algorithms<\/p>\n<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-aside\">\n\t\t\t\t\t\t<img loading=\"lazy\" decoding=\"async\" width=\"366\" height=\"451\" src=\"https:\/\/www.9senses.ai\/wp-content\/uploads\/2026\/07\/larah.png\" class=\"nsp-pop-img\" alt=\"\" srcset=\"https:\/\/www.9senses.ai\/wp-content\/uploads\/2026\/07\/larah.png 366w, https:\/\/www.9senses.ai\/wp-content\/uploads\/2026\/07\/larah-243x300.png 243w, https:\/\/www.9senses.ai\/wp-content\/uploads\/2026\/07\/larah-10x12.png 10w\" sizes=\"(max-width: 366px) 100vw, 366px\" \/>\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-role nsp-pop-role--aside\"><p>Peter acted as the project manager and key data scientist on this project, developed under Python using Deep Learning models like TensorFlow, custom-developed Structure-from-Motion software and custom camera calibration software. Additionally, as a part of the project, two Android apps were modified and integrated together onto the Robot Operating System on the mobile platform.<\/p>\n<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\n\t\t\t\t\t\t\t<ul class=\"nsp-chiplist nsp-pop-caps\">\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/machine-learning\">Machine Learning<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/vision\">Computer Vision<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t<\/ul>\n\t\t\t\t\t<\/div>\n\t\t<\/template><template class=\"nsp-popup-tpl\" data-id=\"1\">\t\t<div class=\"nsp-dialog has-image\">\n\t\t\t<button type=\"button\" class=\"nsp-close\" aria-label=\"Close\">&times;<\/button>\n\t\t\t<div class=\"nsp-dlg-h\">\n\t\t\t\t\t\t\t\t<span class=\"nsp-dlg-h-title\">Electrical Switch Monitor<\/span>\n\t\t\t<\/div>\n\t\t\t<div class=\"nsp-pop-cols\">\n\t\t\t\t<div class=\"nsp-pop-main\">\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-desc\"><p>For a public transportation network, the objective was to optimize the monitoring of their legacy electrical switchboards. These decade-old items that are often located in very remote areas are prone to failures and tracking of errors was not possible. The objective was to enable real-time tracking and the recognition of upcoming failures from changed switching behavior. The key elements were:<\/p>\n<ul>\n<li>Development of specific hardware configuration with custom housings (3D printed) to mount instead of regular switchboard covers;<\/li>\n<li>Camera control and initial image generation on Raspberry Pi integrated in housing;<\/li>\n<li>Initial scan of switch layout and labels;<\/li>\n<li>Identification of switching operations and registration of new positions;<\/li>\n<li>Identification of irregular switching patterns (delays, other irregularities) to indicate upcoming failures for predictive maintenance;<\/li>\n<li>Update of central database and cloud solution with last state and observed switching patterns;<\/li>\n<\/ul>\n<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-aside\">\n\t\t\t\t\t\t<img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"959\" src=\"https:\/\/www.9senses.ai\/wp-content\/uploads\/2026\/07\/6by6cropped-1024x959.jpg\" class=\"nsp-pop-img\" alt=\"\" srcset=\"https:\/\/www.9senses.ai\/wp-content\/uploads\/2026\/07\/6by6cropped-980x918.jpg 980w, https:\/\/www.9senses.ai\/wp-content\/uploads\/2026\/07\/6by6cropped-480x450.jpg 480w\" sizes=\"(min-width: 0px) and (max-width: 480px) 480px, (min-width: 481px) and (max-width: 980px) 980px, (min-width: 981px) 1024px, 100vw\" \/>\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-role nsp-pop-role--aside\"><p>The solution was implemented using Python on Raspberry Pi devices, backbone and cloud processing were done using a LAMP stack, with PyTorch, TensorFlow and OpenCV.<\/p>\n<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\n\t\t\t\t\t\t\t<ul class=\"nsp-chiplist nsp-pop-caps\">\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/machine-learning\">Machine Learning<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/vision\">Computer Vision<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t<\/ul>\n\t\t\t\t\t<\/div>\n\t\t<\/template><template class=\"nsp-popup-tpl\" data-id=\"5\">\t\t<div class=\"nsp-dialog no-image\">\n\t\t\t<button type=\"button\" class=\"nsp-close\" aria-label=\"Close\">&times;<\/button>\n\t\t\t<div class=\"nsp-dlg-h\">\n\t\t\t\t<i class=\"nsp-pop-icon fa-solid fa-water\" aria-hidden=\"true\"><\/i>\t\t\t\t<span class=\"nsp-dlg-h-title\">Hydropower Plant Operations<\/span>\n\t\t\t<\/div>\n\t\t\t<div class=\"nsp-pop-cols\">\n\t\t\t\t<div class=\"nsp-pop-main\">\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-desc\"><div class=\"et_pb_column_11 et_pb_column et_flex_column et_pb_css_mix_blend_mode_passthrough et_flex_column_6_24 et_flex_column_12_24_tablet et_flex_column_24_24_phone et_flex_column_12_24_tabletWide preset--group--divi-column--divi-box-shadow--default preset--group--divi-column--divi-sizing--hsj9uxo--default\">\n<div class=\"et_pb_blurb_2 et_pb_blurb et_pb_bg_layout_light et_pb_blurb_position_top et_pb_module et_flex_module hovergroup preset--group--divi-blurb--divi-box-shadow--default preset--group--divi-blurb--divi-font-body--h19rs5u--7p5s44libg preset--group--divi-blurb--divi-sizing--hsj9uxo--default\">\n<div class=\"et_pb_blurb_content et_flex_module\">\n<div class=\"et_pb_blurb_container\">\n<div class=\"et_pb_blurb_description\">\n<p>Create an integrated monitoring and surveillance solution for small-scale hydropower plants in remote locations. The solution included a full range of required settings:<\/p>\n<ul>\n<li>real-time monitoring and logging of operations<\/li>\n<li>failure detection and automated<\/li>\n<li>predictive maintenance logic to identify early failure<\/li>\n<li>camera-based intrusion and irregularity detection<\/li>\n<\/ul>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-role nsp-pop-role--below\"><p>The solution was a hybrid solution using Siemens PLCs, combined with small edge computing elements (Raspberry Pi and Arduino). All primary logic (particularly shutdown and load adjustment) was local, but key decisions and aggregations were executed online based on regular data transmission to a cloud-based management and operations suite.<\/p>\n<\/div>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\n\t\t\t\t\t\t\t<ul class=\"nsp-chiplist nsp-pop-caps\">\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/machine-learning\">Machine Learning<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/vision\">Computer Vision<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/automation-and-control\">Infrastructure Management<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t<\/ul>\n\t\t\t\t\t<\/div>\n\t\t<\/template><template class=\"nsp-popup-tpl\" data-id=\"2\">\t\t<div class=\"nsp-dialog no-image\">\n\t\t\t<button type=\"button\" class=\"nsp-close\" aria-label=\"Close\">&times;<\/button>\n\t\t\t<div class=\"nsp-dlg-h\">\n\t\t\t\t<span class=\"nsp-pop-icon et-pb-icon\" aria-hidden=\"true\">&#x77;<\/span>\t\t\t\t<span class=\"nsp-dlg-h-title\">Customer Interaction Analysis<\/span>\n\t\t\t<\/div>\n\t\t\t<div class=\"nsp-pop-cols\">\n\t\t\t\t<div class=\"nsp-pop-main\">\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-desc\"><p>For a project management SaaS solution where customers were matched with freelancers, a custom AI solution was established with the purpose to improve experiences for all parties. Key purposes were to create an early warning system to help customer service intervene in case of issues:<\/p>\n<ul>\n<li>Identification of unusual patterns (delays indicating inaction, intense exchanges);<\/li>\n<li>Flagging of language transgressions on both sides (use of inappropriate language, aggression);<\/li>\n<li>Matching of final ratings with evaluation of flow and dialogue quality to foster a more honest rating culture;<\/li>\n<li>Language style matching to improve future matching of freelancers to clients;<\/li>\n<\/ul>\n<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-role nsp-pop-role--below\"><p>The solution was implemented using Python on a LAMP stack, with self-developed machine learning libraries.<\/p>\n<\/div>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\n\t\t\t\t\t\t\t<ul class=\"nsp-chiplist nsp-pop-caps\">\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/machine-learning\">Machine Learning<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/nlp\">Natural Language Processing<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t<\/ul>\n\t\t\t\t\t<\/div>\n\t\t<\/template><template class=\"nsp-popup-tpl\" data-id=\"6\">\t\t<div class=\"nsp-dialog no-image\">\n\t\t\t<button type=\"button\" class=\"nsp-close\" aria-label=\"Close\">&times;<\/button>\n\t\t\t<div class=\"nsp-dlg-h\">\n\t\t\t\t<i class=\"nsp-pop-icon fa-solid fa-shirt\" aria-hidden=\"true\"><\/i>\t\t\t\t<span class=\"nsp-dlg-h-title\">Motion-sensitive wearables<\/span>\n\t\t\t<\/div>\n\t\t\t<div class=\"nsp-pop-cols\">\n\t\t\t\t<div class=\"nsp-pop-main\">\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-desc\"><p>Arduino-based, with edge ml functionality and periodic link via BLE to connected phone. Machine learning algorithms on cloud detecting various health states and delivering alerts to phone or web app. All logic woven into a cotton fabric wristband.<\/p>\n<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\n\t\t\t\t\t\t\t<ul class=\"nsp-chiplist nsp-pop-caps\">\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/machine-learning\">Machine Learning<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t<\/ul>\n\t\t\t\t\t<\/div>\n\t\t<\/template><\/div><div class=\"nsp-overlay\" id=\"nsp-overlay\" role=\"dialog\" aria-modal=\"true\"><\/div><\/div><\/div>\n<\/div>\n<\/div>\n<\/div>\n\n<div class=\"et_pb_section_6 et_pb_section et_section_regular et_flex_section ns-panel\" id=\"nlp\" style=\"--ns-rail-w:264px; --ns-body-color:#D3D3D3;\">\n<div class=\"et_pb_row_10 et_pb_row et_flex_row\">\n<div class=\"et_pb_column_16 et_pb_column et_block_column et_pb_css_mix_blend_mode_passthrough et_flex_column_24_24 et_flex_column_24_24_tabletWide\">\n<div class=\"et_pb_text_9 et_pb_text et_pb_bg_layout_light et_pb_module et_flex_module\"><div class=\"et_pb_text_inner\"><p>There is nothing more fascinating than being able to converse with computers in normal language: asking questions and receiving meaningful answers. Expected for more than half a century, this only became realistically possible a few years ago with the arrival of the first large language models. These LLMs, like ChatGPT or Gemini, have since moved from research milestone to everyday business tool, from employee support to <a href=\"https:\/\/www.9senses.ai\/customer-interaction\/\">customer interactions<\/a>.<\/p>\n<\/div><\/div>\n\n<div class=\"et_pb_code_4 et_pb_code et_pb_module\"><div class=\"et_pb_code_inner\">\t<div\n\t\tid=\"ns-aix-2\"\n\t\tclass=\"ns-aix ns-aix-nlp\"\n\t\tdata-topic=\"nlp\"\n\t\tdata-flow=\"development\"\n\t\t\t\tdata-step=\"0\"\n\t\tdata-interval=\"9500\"\n\t\tdata-handoff-duration=\"7200\"\t\tdata-autoplay=\"1\"\n\t\trole=\"region\"\n\t\taria-label=\"Comment fonctionne le NLP\"\n\t\tstyle=\"--ns-aix-accent:#58a7f9;--ns-aix-secondary:#2b6cb0\"\n\t>\n\t\t<header class=\"ns-aix-head\">\n\t\t\t\t\t\t<h3 class=\"ns-aix-title\">Comment fonctionne le NLP<\/h3>\n\t\t\t<p class=\"ns-aix-intro\">Le traitement du langage transforme le texte en tokens, vecteurs et probabilit\u00e9s. Le mod\u00e8le est construit une fois lors de l\u2019entra\u00eenement, puis applique ce m\u00e9canisme en direct \u00e0 chaque requ\u00eate. C\u2019est ce m\u00eame fonctionnement qui le rend utile \u2014 et qui explique pourquoi les \u00e9carts entre langues et les erreurs subsistent.<\/p>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-flowgroup\">\n\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-tabs\" role=\"tablist\" aria-label=\"Comment fonctionne le NLP\">\n\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\tid=\"ns-aix-2-tab-development\"\n\t\t\t\t\t\t\tclass=\"ns-aix-tab ns-on\"\n\t\t\t\t\t\t\tdata-flow=\"development\"\n\t\t\t\t\t\t\trole=\"tab\"\n\t\t\t\t\t\t\taria-selected=\"true\"\n\t\t\t\t\t\t\taria-label=\"Afficher le parcours de d\u00e9veloppement et d\u2019entra\u00eenement du NLP\"\n\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\taria-controls=\"ns-aix-2-flow-panel-development\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\tD\u00e9veloppement &amp; entra\u00eenement de base\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\tid=\"ns-aix-2-tab-execution\"\n\t\t\t\t\t\t\tclass=\"ns-aix-tab\"\n\t\t\t\t\t\t\tdata-flow=\"execution\"\n\t\t\t\t\t\t\trole=\"tab\"\n\t\t\t\t\t\t\taria-selected=\"false\"\n\t\t\t\t\t\t\taria-label=\"Afficher le parcours d\u2019impl\u00e9mentation du NLP\"\n\t\t\t\t\t\t\ttabindex=\"-1\"\n\t\t\t\t\t\t\taria-controls=\"ns-aix-2-flow-panel-execution\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\tMise en \u0153uvre\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t<\/div>\n\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t<\/header>\n\n\t\t<div class=\"ns-aix-body\">\n\t\t\t<div class=\"ns-aix-copy\">\n\t\t\t\t<div class=\"ns-aix-progress\" aria-hidden=\"true\"><span><\/span><\/div>\n\t\t\t\t\t\t\t\t\t<div\n\t\t\t\t\t\tid=\"ns-aix-2-flow-panel-development\"\n\t\t\t\t\t\tclass=\"ns-aix-step-panel ns-on\"\n\t\t\t\t\t\tdata-flow=\"development\"\n\t\t\t\t\t\taria-hidden=\"false\"\n\t\t\t\t\t\trole=\"tabpanel\" aria-labelledby=\"ns-aix-2-tab-development\"\t\t\t\t\t>\n\t\t\t\t\t<ol class=\"ns-aix-steps ns-on\" data-flow=\"development\" role=\"list\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step ns-on\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"development\"\n\t\t\t\t\t\t\t\t\tdata-step=\"0\"\n\t\t\t\t\t\t\t\t\tdata-key=\"collect\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"true\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-2-panel-development-0\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">01<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Collecter<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Le corpus d\u00e9finit le point de d\u00e9part.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"development\"\n\t\t\t\t\t\t\t\t\tdata-step=\"1\"\n\t\t\t\t\t\t\t\t\tdata-key=\"split\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-2-panel-development-1\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">02<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">R\u00e8gles de tokenisation<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">La tokenisation n\u2019est pas neutre.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"development\"\n\t\t\t\t\t\t\t\t\tdata-step=\"2\"\n\t\t\t\t\t\t\t\t\tdata-key=\"map\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-2-panel-development-2\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">03<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Carte vectorielle<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Les zones denses se comportent mieux.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"development\"\n\t\t\t\t\t\t\t\t\tdata-step=\"3\"\n\t\t\t\t\t\t\t\t\tdata-key=\"train\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-2-panel-development-3\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">04<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Entra\u00eener<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">La fluidit\u00e9 est optimis\u00e9e statistiquement.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"development\"\n\t\t\t\t\t\t\t\t\tdata-step=\"4\"\n\t\t\t\t\t\t\t\t\tdata-key=\"adapt\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-2-panel-development-4\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">05<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Adapter<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Un bon d\u00e9ploiement demande des donn\u00e9es locales.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"development\"\n\t\t\t\t\t\t\t\t\tdata-step=\"5\"\n\t\t\t\t\t\t\t\t\tdata-key=\"test\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-2-panel-development-5\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">06<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Tester<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Les moyennes masquent les concentrations d\u2019erreurs.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"development\"\n\t\t\t\t\t\t\t\t\tdata-step=\"6\"\n\t\t\t\t\t\t\t\t\tdata-key=\"govern\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-2-panel-development-6\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">07<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Piloter<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Le contr\u00f4le doit \u00eatre con\u00e7u d\u00e8s le d\u00e9part.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t<\/ol>\n\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t<div\n\t\t\t\t\t\tid=\"ns-aix-2-flow-panel-execution\"\n\t\t\t\t\t\tclass=\"ns-aix-step-panel\"\n\t\t\t\t\t\tdata-flow=\"execution\"\n\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\trole=\"tabpanel\" aria-labelledby=\"ns-aix-2-tab-execution\"\t\t\t\t\t>\n\t\t\t\t\t<ol class=\"ns-aix-steps\" data-flow=\"execution\" role=\"list\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"execution\"\n\t\t\t\t\t\t\t\t\tdata-step=\"0\"\n\t\t\t\t\t\t\t\t\tdata-key=\"read\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-2-panel-execution-0\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">01<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Lire<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">La formulation humaine entre dans la cha\u00eene de traitement.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"execution\"\n\t\t\t\t\t\t\t\t\tdata-step=\"1\"\n\t\t\t\t\t\t\t\t\tdata-key=\"tokenize\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-2-panel-execution-1\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">02<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Tokeniser<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Les mots deviennent des fragments.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"execution\"\n\t\t\t\t\t\t\t\t\tdata-step=\"2\"\n\t\t\t\t\t\t\t\t\tdata-key=\"vectorize\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-2-panel-execution-2\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">03<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Vectoriser<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Le sens devient g\u00e9om\u00e9trie.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"execution\"\n\t\t\t\t\t\t\t\t\tdata-step=\"3\"\n\t\t\t\t\t\t\t\t\tdata-key=\"contextualize\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-2-panel-execution-3\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">04<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Contexte<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">L\u2019attention remod\u00e8le les vecteurs.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"execution\"\n\t\t\t\t\t\t\t\t\tdata-step=\"4\"\n\t\t\t\t\t\t\t\t\tdata-key=\"predict\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-2-panel-execution-4\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">05<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Pr\u00e9dire<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">La fluidit\u00e9 se construit token par token.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"execution\"\n\t\t\t\t\t\t\t\t\tdata-step=\"5\"\n\t\t\t\t\t\t\t\t\tdata-key=\"confabulate\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-2-panel-execution-5\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">06<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Risque<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">L\u2019assurance peut d\u00e9passer les preuves.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"execution\"\n\t\t\t\t\t\t\t\t\tdata-step=\"6\"\n\t\t\t\t\t\t\t\t\tdata-key=\"ground\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-2-panel-execution-6\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">07<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">V\u00e9rifier<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Le contr\u00f4le vient de structures ajout\u00e9es.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t<\/ol>\n\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\n\t\t\t<div class=\"ns-aix-stage\" aria-label=\"Animated vector-map process diagram\">\n\t\t\t\t<canvas class=\"ns-aix-canvas\" aria-hidden=\"true\"><\/canvas>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node ns-on\" data-flow=\"development\" data-step=\"0\" data-key=\"collect\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>corpus<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node\" data-flow=\"development\" data-step=\"1\" data-key=\"split\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>tokens<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node\" data-flow=\"development\" data-step=\"2\" data-key=\"map\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>vecteurs<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node\" data-flow=\"development\" data-step=\"3\" data-key=\"train\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>entra\u00eener<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node\" data-flow=\"development\" data-step=\"4\" data-key=\"adapt\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>adapter<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node\" data-flow=\"development\" data-step=\"5\" data-key=\"test\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>test<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node\" data-flow=\"development\" data-step=\"6\" data-key=\"govern\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>piloter<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node ns-hidden\" data-flow=\"execution\" data-step=\"0\" data-key=\"read\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>entr\u00e9e<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node ns-hidden\" data-flow=\"execution\" data-step=\"1\" data-key=\"tokenize\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>tokens<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node ns-hidden\" data-flow=\"execution\" data-step=\"2\" data-key=\"vectorize\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>vecteurs<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node ns-hidden\" data-flow=\"execution\" data-step=\"3\" data-key=\"contextualize\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>contexte<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node ns-hidden\" data-flow=\"execution\" data-step=\"4\" data-key=\"predict\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>sortie<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node ns-hidden\" data-flow=\"execution\" data-step=\"5\" data-key=\"confabulate\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>risque<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node ns-hidden\" data-flow=\"execution\" data-step=\"6\" data-key=\"ground\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>v\u00e9rifier<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\n\t\t\t\t\t\t\t<div class=\"ns-aix-handoff-panel\" role=\"status\" aria-live=\"polite\" aria-hidden=\"true\">\n\t\t\t\t\t<div class=\"ns-aix-handoff-rail\" aria-hidden=\"true\"><\/div>\n\t\t\t\t\t<div class=\"ns-aix-handoff-inner\">\n\t\t\t\t\t\t<span class=\"ns-aix-handoff-kicker\">d\u00e9ploiement<\/span>\n\t\t\t\t\t\t<h4>Le mod\u00e8le passe en production<\/h4>\n\t\t\t\t\t\t<p>Tout ce qui rel\u00e8ve du d\u00e9veloppement n\u2019a lieu qu\u2019une fois, avant le lancement. Ensuite, le m\u00eame mod\u00e8le traite chaque message envoy\u00e9 par un utilisateur \u2014 les \u00e9tapes suivantes se r\u00e9p\u00e8tent \u00e0 chaque requ\u00eate.<\/p>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\n\t\t<div class=\"ns-aix-controls\">\n\t\t\t<div class=\"ns-aix-dots\" role=\"navigation\" aria-label=\"Step navigation\">\n\t\t\t\t\t\t\t\t\t<span\n\t\t\t\t\t\tclass=\"ns-aix-dotset ns-on\"\n\t\t\t\t\t\tdata-flow=\"development\"\n\t\t\t\t\t\trole=\"group\"\n\t\t\t\t\t\taria-label=\"Steps for D\u00e9veloppement &amp; entra\u00eenement de base\"\n\t\t\t\t\t\taria-hidden=\"false\"\n\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot ns-on\"\n\t\t\t\t\t\t\t\tdata-flow=\"development\"\n\t\t\t\t\t\t\t\tdata-step=\"0\"\n\t\t\t\t\t\t\t\taria-label=\"Step 1: Collecter\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-2-panel-development-0\"\n\t\t\t\t\t\t\t\taria-pressed=\"true\"\n\t\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"development\"\n\t\t\t\t\t\t\t\tdata-step=\"1\"\n\t\t\t\t\t\t\t\taria-label=\"Step 2: R\u00e8gles de tokenisation\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-2-panel-development-1\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"development\"\n\t\t\t\t\t\t\t\tdata-step=\"2\"\n\t\t\t\t\t\t\t\taria-label=\"Step 3: Carte vectorielle\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-2-panel-development-2\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"development\"\n\t\t\t\t\t\t\t\tdata-step=\"3\"\n\t\t\t\t\t\t\t\taria-label=\"Step 4: Entra\u00eener\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-2-panel-development-3\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"development\"\n\t\t\t\t\t\t\t\tdata-step=\"4\"\n\t\t\t\t\t\t\t\taria-label=\"Step 5: Adapter\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-2-panel-development-4\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"development\"\n\t\t\t\t\t\t\t\tdata-step=\"5\"\n\t\t\t\t\t\t\t\taria-label=\"Step 6: Tester\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-2-panel-development-5\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"development\"\n\t\t\t\t\t\t\t\tdata-step=\"6\"\n\t\t\t\t\t\t\t\taria-label=\"Step 7: Piloter\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-2-panel-development-6\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t\t<span\n\t\t\t\t\t\tclass=\"ns-aix-dotset\"\n\t\t\t\t\t\tdata-flow=\"execution\"\n\t\t\t\t\t\trole=\"group\"\n\t\t\t\t\t\taria-label=\"Steps for Mise en \u0153uvre\"\n\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"execution\"\n\t\t\t\t\t\t\t\tdata-step=\"0\"\n\t\t\t\t\t\t\t\taria-label=\"Step 1: Lire\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-2-panel-execution-0\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"-1\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"execution\"\n\t\t\t\t\t\t\t\tdata-step=\"1\"\n\t\t\t\t\t\t\t\taria-label=\"Step 2: Tokeniser\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-2-panel-execution-1\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"-1\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"execution\"\n\t\t\t\t\t\t\t\tdata-step=\"2\"\n\t\t\t\t\t\t\t\taria-label=\"Step 3: Vectoriser\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-2-panel-execution-2\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"-1\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"execution\"\n\t\t\t\t\t\t\t\tdata-step=\"3\"\n\t\t\t\t\t\t\t\taria-label=\"Step 4: Contexte\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-2-panel-execution-3\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"-1\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"execution\"\n\t\t\t\t\t\t\t\tdata-step=\"4\"\n\t\t\t\t\t\t\t\taria-label=\"Step 5: Pr\u00e9dire\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-2-panel-execution-4\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"-1\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"execution\"\n\t\t\t\t\t\t\t\tdata-step=\"5\"\n\t\t\t\t\t\t\t\taria-label=\"Step 6: Risque\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-2-panel-execution-5\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"-1\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"execution\"\n\t\t\t\t\t\t\t\tdata-step=\"6\"\n\t\t\t\t\t\t\t\taria-label=\"Step 7: V\u00e9rifier\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-2-panel-execution-6\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"-1\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t<\/div>\n\t\t<\/div>\n\n\t\t<div class=\"ns-aix-below\" aria-live=\"polite\">\n\t\t\t<div class=\"ns-aix-explain\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-2-panel-development-0\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel ns-on\"\n\t\t\t\t\t\t\tdata-flow=\"development\"\n\t\t\t\t\t\t\tdata-step=\"0\"\n\t\t\t\t\t\t\taria-hidden=\"false\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">01<\/span>Collecter le texte<\/h4>\n\t\t\t\t\t\t\t<p>L\u2019entra\u00eenement commence par de grandes collections de textes. Le mod\u00e8le apprend la forme statistique du mat\u00e9riau qu\u2019il voit. Comme les corpus issus du web sont souvent tr\u00e8s domin\u00e9s par l\u2019anglais, la couverture des langues est in\u00e9gale d\u00e8s le d\u00e9part si le projet ne compense pas volontairement ce d\u00e9s\u00e9quilibre.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-2-panel-development-1\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"development\"\n\t\t\t\t\t\t\tdata-step=\"1\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">02<\/span>D\u00e9couper en tokens<\/h4>\n\t\t\t\t\t\t\t<p>Avant l\u2019entra\u00eenement, le texte est d\u00e9coup\u00e9 en tokens \u00e0 l\u2019aide d\u2019un vocabulaire lui-m\u00eame appris \u2014 g\u00e9n\u00e9ralement \u00e0 partir de textes tr\u00e8s anglophones. Certaines langues ont donc besoin de davantage de tokens pour exprimer la m\u00eame id\u00e9e ; les mots compos\u00e9s, les flexions et les \u00e9critures peu repr\u00e9sent\u00e9es consomment plus vite le contexte et affaiblissent le raisonnement en aval.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-2-panel-development-2\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"development\"\n\t\t\t\t\t\t\tdata-step=\"2\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">03<\/span>Construire l\u2019espace vectoriel<\/h4>\n\t\t\t\t\t\t\t<p>Tokens et passages deviennent des vecteurs. Les motifs fr\u00e9quents forment des voisinages denses ; les langues rares et les domaines de niche cr\u00e9ent des zones clairsem\u00e9es o\u00f9 le voisin le plus proche peut d\u00e9j\u00e0 \u00eatre trop \u00e9loign\u00e9.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-2-panel-development-3\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"development\"\n\t\t\t\t\t\t\tdata-step=\"3\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">04<\/span>Entra\u00eener la pr\u00e9diction<\/h4>\n\t\t\t\t\t\t\t<p>Le mod\u00e8le pr\u00e9dit encore et encore des tokens masqu\u00e9s ou les tokens suivants, puis ajuste ses param\u00e8tres lorsqu\u2019il se trompe. Les petits mod\u00e8les sont moins chers et plus rapides, mais ont moins de capacit\u00e9 ; les grands couvrent davantage de motifs et de langues, au prix de plus de calcul. Les deux optimisent la pr\u00e9diction, pas la v\u00e9rit\u00e9.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-2-panel-development-4\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"development\"\n\t\t\t\t\t\t\tdata-step=\"4\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">05<\/span>Adapter localement<\/h4>\n\t\t\t\t\t\t\t<p>Un syst\u00e8me utile en entreprise a souvent besoin d\u2019un fine-tuning propre \u00e0 la langue, d\u2019un alignement par retour humain, d\u2019exemples s\u00e9lectionn\u00e9s, de donn\u00e9es de retrieval et de garde-fous. Cela repr\u00e9sente un travail suppl\u00e9mentaire, surtout hors anglais et dans les domaines sp\u00e9cialis\u00e9s.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-2-panel-development-5\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"development\"\n\t\t\t\t\t\t\tdata-step=\"5\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">06<\/span>Tester les points faibles<\/h4>\n\t\t\t\t\t\t\t<p>L\u2019\u00e9valuation doit viser pr\u00e9cis\u00e9ment les endroits o\u00f9 le mod\u00e8le risque le plus d\u2019\u00e9chouer : langues minoritaires, terminologie sp\u00e9cialis\u00e9e, entit\u00e9s rares, formulations ambigu\u00ebs et questions reposant sur peu de sources.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-2-panel-development-6\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"development\"\n\t\t\t\t\t\t\tdata-step=\"6\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">07<\/span>Piloter le mod\u00e8le<\/h4>\n\t\t\t\t\t\t\t<p>Logs, audits, seuils de refus, revue humaine et cycles de mise \u00e0 jour d\u00e9terminent si l\u2019on obtient un syst\u00e8me ma\u00eetris\u00e9 \u2014 ou une bo\u00eete noire \u00e9loquente aux performances linguistiques in\u00e9gales.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-2-panel-execution-0\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"execution\"\n\t\t\t\t\t\t\tdata-step=\"0\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">01<\/span>Recevoir le texte<\/h4>\n\t\t\t\t\t\t\t<p>Un utilisateur \u00e9crit une question, un document arrive ou un client parle \u00e0 un bot. Le syst\u00e8me re\u00e7oit d\u2019abord des caract\u00e8res, pas du sens.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-2-panel-execution-1\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"execution\"\n\t\t\t\t\t\t\tdata-step=\"1\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">02<\/span>Former les tokens<\/h4>\n\t\t\t\t\t\t\t<p>Le texte est d\u00e9coup\u00e9 en mots, sous-mots, signes de ponctuation ou fragments. Ces tokens sont convertis en identifiants num\u00e9riques avant que le mod\u00e8le puisse les traiter.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-2-panel-execution-2\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"execution\"\n\t\t\t\t\t\t\tdata-step=\"2\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">03<\/span>Utiliser les vecteurs<\/h4>\n\t\t\t\t\t\t\t<p>Chaque token est repr\u00e9sent\u00e9 par un vecteur de grande dimension : une longue liste de nombres. Des contextes proches rapprochent les tokens. C\u2019est de la proximit\u00e9, pas de la v\u00e9rit\u00e9.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-2-panel-execution-3\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"execution\"\n\t\t\t\t\t\t\tdata-step=\"3\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">04<\/span>Appliquer le contexte<\/h4>\n\t\t\t\t\t\t\t<p>Le m\u00e9canisme d\u2019attention du transformer compare chaque token aux tokens qui l\u2019entourent. Le vecteur d\u2019un mot change avec sa phrase, mais le mod\u00e8le apprend toujours des relations \u2014 pas un mod\u00e8le du monde ancr\u00e9 dans la r\u00e9alit\u00e9.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-2-panel-execution-4\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"execution\"\n\t\t\t\t\t\t\tdata-step=\"4\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">05<\/span>Choisir la sortie<\/h4>\n\t\t\t\t\t\t\t<p>Le mod\u00e8le estime quel token devrait suivre. Les r\u00e9ponses fluides \u00e9mergent d\u2019une succession de choix probabilistes ; la v\u00e9rification des faits ne fait pas partie du m\u00e9canisme de g\u00e9n\u00e9ration.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-2-panel-execution-5\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"execution\"\n\t\t\t\t\t\t\tdata-step=\"5\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">06<\/span>Surveiller le risque<\/h4>\n\t\t\t\t\t\t\t<p>Lorsque les \u00e9l\u00e9ments disponibles sont limit\u00e9s, le mod\u00e8le fournit tout de m\u00eame la continuation la plus plausible. C\u2019est pourquoi il peut para\u00eetre s\u00fbr de lui tout en inventant des faits, surtout dans les domaines peu document\u00e9s ou les langues moins bien repr\u00e9sent\u00e9es.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-2-panel-execution-6\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"execution\"\n\t\t\t\t\t\t\tdata-step=\"6\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">07<\/span>Ancrer la r\u00e9ponse<\/h4>\n\t\t\t\t\t\t\t<p>Les meilleurs syst\u00e8mes ralentissent le mod\u00e8le avec du retrieval, des contr\u00f4les de sources, des r\u00e8gles m\u00e9tier ou une revue humaine. Les retours de production alimentent ensuite la conception des prompts, les jeux d\u2019\u00e9valuation, le fine-tuning et la gouvernance. La fiabilit\u00e9 vient de la mise en \u0153uvre, pas du mod\u00e8le de langage seul.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t<div class=\"ns-aix-aspects\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset ns-on\" data-flow=\"development\" data-step=\"0\" aria-hidden=\"false\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Contexte<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Le NLP remonte aux ann\u00e9es 1950 et \u00e0 la traduction automatique fond\u00e9e sur des r\u00e8gles. Les m\u00e9thodes statistiques ont pris le relais dans les ann\u00e9es 1990 et 2000 ; le Deep Learning et les grands mod\u00e8les de langage ont apport\u00e9 les grandes avanc\u00e9es des ann\u00e9es 2010.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Attention<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Un mod\u00e8le reproduit les erreurs et les biais pr\u00e9sents dans ses donn\u00e9es d\u2019entra\u00eenement \u2014 le corpus d\u00e9termine le r\u00e9sultat.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"development\" data-step=\"1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Concept cl\u00e9<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Le NLP combine linguistique et Machine Learning : les syst\u00e8mes analysent la grammaire (syntaxe), le sens (s\u00e9mantique) et parfois l\u2019intention (pragmatique).<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>En pratique<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Une m\u00eame phrase peut consommer des budgets de tokens tr\u00e8s diff\u00e9rents selon la langue \u2014 avec un effet concret sur le co\u00fbt comme sur la qualit\u00e9 des r\u00e9ponses.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"development\" data-step=\"2\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Comment \u00e7a marche<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Les mots et sous-mots sont projet\u00e9s dans un espace vectoriel multidimensionnel qui capte des motifs de sens \u00e0 partir de la mani\u00e8re dont les mots apparaissent ensemble.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Pourquoi c\u2019est important<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Les hallucinations se concentrent l\u00e0 o\u00f9 les sources sont rares : dans les zones clairsem\u00e9es, le voisin statistique le plus proche peut \u00eatre trop \u00e9loign\u00e9 pour repr\u00e9senter un fait, mais le mod\u00e8le r\u00e9pond quand m\u00eame.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"development\" data-step=\"3\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Concept cl\u00e9<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Le NLP moderne repose sur des transformers qui traitent la langue en analysant les relations entre les mots d\u2019une phrase, plut\u00f4t qu\u2019en suivant des r\u00e8gles grammaticales rigides.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Attention<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>La g\u00e9n\u00e9ration \u00e9chantillonne des probabilit\u00e9s. Un m\u00eame prompt peut donc produire des r\u00e9ponses sensiblement diff\u00e9rentes d\u2019une session \u00e0 l\u2019autre \u2014 consid\u00e9rez un verdict isol\u00e9 de l\u2019IA comme un tirage, pas comme une opinion stable.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"development\" data-step=\"4\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Concept cl\u00e9<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Les petits comme les grands mod\u00e8les de langage sont d\u2019abord entra\u00een\u00e9s sur la langue g\u00e9n\u00e9rale et le raisonnement, puis affin\u00e9s pour des applications pr\u00e9cises et compl\u00e9t\u00e9s par des donn\u00e9es sp\u00e9cifiques accessibles par retrieval.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>En pratique<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>L\u2019automatisation du service client, l\u2019analyse de documents juridiques ou de dossiers m\u00e9dicaux d\u00e9pendent toutes de cette couche d\u2019adaptation, pas seulement du mod\u00e8le de base.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"development\" data-step=\"5\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Pourquoi c\u2019est important<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Les taux d\u2019erreur mesur\u00e9s augmentent fortement \u00e0 mesure que les sujets sont moins document\u00e9s \u2014 d\u2019environ 1 % pour le r\u00e9sum\u00e9 de documents courts jusqu\u2019\u00e0 une majorit\u00e9 des r\u00e9ponses pour certaines questions sp\u00e9cialis\u00e9es de niche.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>En pratique<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>\u00c9valuez le syst\u00e8me avec des experts du domaine, sur vos propres documents et dans vos propres langues, pas seulement \u00e0 l\u2019aide de benchmarks publics qui r\u00e9compensent la performance moyenne.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"development\" data-step=\"6\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Concept cl\u00e9<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>La supervision humaine et l\u2019\u00e9valuation critique sont indispensables ; l\u2019entra\u00eenement et l\u2019exploitation de grands mod\u00e8les soul\u00e8vent aussi des questions de confidentialit\u00e9, de d\u00e9sinformation et d\u2019usage abusif.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Attention<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Sans logs ni revue, le syst\u00e8me reste fluide mais sans v\u00e9ritable responsabilit\u00e9 \u2014 et ses performances linguistiques restent in\u00e9gales l\u00e0 o\u00f9 personne ne les mesure.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"execution\" data-step=\"0\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Ce que le NLP peut faire<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Traduire des langues, r\u00e9sumer de longs textes, extraire des informations cl\u00e9s, analyser le sentiment et alimenter des agents conversationnels \u2014 avec beaucoup moins de travail manuel dans les activit\u00e9s fortement textuelles.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>En pratique<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>La reconnaissance et la synth\u00e8se vocales \u00e9tendent la m\u00eame cha\u00eene de traitement aux interactions orales avec les machines.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"execution\" data-step=\"1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Comment \u00e7a marche<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Le texte n\u2019entre ni ne sort jamais d\u2019un mod\u00e8le de langage sous forme de texte. Il est d\u00e9coup\u00e9 en fragments de sous-mots, chacun associ\u00e9 \u00e0 un identifiant entier.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Pourquoi c\u2019est important<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Tout ce que le mod\u00e8le \u00ab sait \u00bb de votre texte passe par ces fragments \u2014 rien d\u2019autre.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"execution\" data-step=\"2\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Concept cl\u00e9<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Chaque vecteur contient g\u00e9n\u00e9ralement des milliers de nombres. L\u2019entra\u00eenement ajuste des milliards de param\u00e8tres pour que les tokens issus de contextes similaires obtiennent des vecteurs math\u00e9matiquement proches.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Attention<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>\u00c0 aucun moment le syst\u00e8me ne v\u00e9rifie quelque chose par rapport \u00e0 la v\u00e9rit\u00e9 : il v\u00e9rifie la proximit\u00e9. Deux orthographes d\u2019un nom sont presque identiques sous forme de vecteurs ; la variante statistiquement la plus probable l\u2019emporte.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"execution\" data-step=\"3\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Comment \u00e7a marche<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>L\u2019attention modifie le vecteur d\u2019un mot selon la phrase \u2014 c\u2019est ce qui permet au m\u00eame mot de prendre des sens diff\u00e9rents selon le contexte.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Pourquoi c\u2019est important<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Le mod\u00e8le apprend des relations statistiques entre les mots et les expressions, pas une compr\u00e9hension ancr\u00e9e du monde qu\u2019ils d\u00e9crivent.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"execution\" data-step=\"4\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Comment \u00e7a marche<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Le mod\u00e8le choisit successivement un prochain token probable jusqu\u2019\u00e0 compl\u00e9ter la r\u00e9ponse. De petites diff\u00e9rences statistiques au d\u00e9but peuvent entra\u00eener une r\u00e9ponse finale tr\u00e8s diff\u00e9rente.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Attention<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>L\u2019assurance fait partie du style de g\u00e9n\u00e9ration et n\u2019a aucun lien avec la validit\u00e9 du contenu : le syst\u00e8me para\u00eet tout aussi convaincu lorsqu\u2019il devine.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"execution\" data-step=\"5\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Pourquoi c\u2019est important<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Les erreurs se concentrent pr\u00e9cis\u00e9ment sur les sujets que les utilisateurs sont le moins capables de v\u00e9rifier. Plus votre propre expertise est faible, plus votre scepticisme devrait \u00eatre \u00e9lev\u00e9.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>En pratique<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Poser plusieurs fois la m\u00eame question dans de nouvelles sessions et comparer les r\u00e9ponses montre o\u00f9 le mod\u00e8le est sur un terrain solide \u2014 et o\u00f9 il ne l\u2019est pas.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"execution\" data-step=\"6\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Concept cl\u00e9<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>La fiabilit\u00e9 est une propri\u00e9t\u00e9 de la mise en \u0153uvre : retrieval, v\u00e9rification des sources, r\u00e8gles m\u00e9tier et contr\u00f4le humain s\u2019ajoutent autour du mod\u00e8le ; ils ne se trouvent pas \u00e0 l\u2019int\u00e9rieur.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>En pratique<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Lorsque la similarit\u00e9 du retrieval est faible, un syst\u00e8me bien con\u00e7u privil\u00e9gie la prudence : il renonce \u00e0 r\u00e9pondre ou signale explicitement une faible confiance au lieu de deviner.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t<\/div>\n\t<\/div><\/div>\n<\/div>\n\n<div class=\"et_pb_column_17 et_pb_column et-last-child et_block_column et_pb_css_mix_blend_mode_passthrough et_flex_column_24_24 et_flex_column_24_24_tablet et_flex_column_24_24_phone et_flex_column_24_24_tabletWide\">\n<div class=\"et_pb_text_10 et_pb_text et_pb_bg_layout_light et_pb_module et_flex_module\"><div class=\"et_pb_text_inner\"><h3>The Challenges<\/h3>\n<p data-start=\"3434\" data-end=\"3776\">No matter how \"human\" they sound, NLP systems face important system-defined limitations. They can easily produce fluent but factually incorrect or misleading information. They also reproduce any falsehood or bias in the information available when trained.<\/p>\n<p data-start=\"3434\" data-end=\"3776\">As NLP systems are solely based on statistical patterns and have only limited contextual understanding, they can struggle with reasoning, and consistency. Also, they are solely based on the input provided during training and feedback during operations. This is particularly problematic with large open models.<\/p>\n<p data-start=\"3778\" data-end=\"4217\" data-is-last-node=\"\" data-is-only-node=\"\">Training large models requires significant resources and raises concerns about privacy, misinformation, and misuse. Human supervision and critical evaluation are thus essential.<\/p>\n<\/div><\/div>\n<\/div>\n<\/div>\n\n<div class=\"et_pb_row_11 et_pb_row et_flex_row\" style=\"--ns-caro-top:26px\">\n<div class=\"et_pb_column_18 et_pb_column et-last-child et_flex_column et_pb_css_mix_blend_mode_passthrough et_flex_column_24_24 et_flex_column_24_24_tablet et_flex_column_24_24_phone et_flex_column_24_24_tabletWide\">\n<div class=\"et_pb_code_5 et_pb_code et_pb_module\"><div class=\"et_pb_code_inner\"><div class=\"nsp-frame nsp-frame--transparent\"><div class=\"nsp-frame-head\"><h2>Featured Projects<\/h2><\/div><div class=\"nsp-carousel-wrap\" data-base-width=\"240\" data-loop=\"1\"><button type=\"button\" class=\"nsp-nav nsp-prev\" aria-label=\"Previous\"><span class=\"nsp-nav-icon et-pb-icon\" aria-hidden=\"true\">&#x34;<\/span><\/button><div class=\"nsp-carousel\">\t\t<article class=\"nsp-card\" tabindex=\"0\" role=\"button\" data-id=\"3\" aria-label=\"RAG-driven Legal Chatbot\">\n\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-card-iconwrap\"><i class=\"nsp-card-icon fa-solid fa-section\" aria-hidden=\"true\"><\/i><\/div>\n\t\t\t\t\t\t<h3 class=\"nsp-card-title\">RAG-driven Legal Chatbot<\/h3>\n\t\t\t\t\t\t\t\t\t\t<p class=\"nsp-card-excerpt\">A small-to-medium language model reliably answering German legal questions based on a strong RAG pipeline.<\/p>\n\t\t\t\t\t\t\t\t\t\t<span class=\"nsp-card-more\">read more\u2026<\/span>\n\t\t\t\t\t<\/article>\n\t\t\t\t<article class=\"nsp-card\" tabindex=\"0\" role=\"button\" data-id=\"7\" aria-label=\"Generative AI Audit Framework\">\n\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-card-iconwrap\"><span class=\"nsp-card-icon et-pb-icon\" aria-hidden=\"true\">&#xe0f7;<\/span><\/div>\n\t\t\t\t\t\t<h3 class=\"nsp-card-title\">Generative AI Audit Framework<\/h3>\n\t\t\t\t\t\t\t\t\t\t<p class=\"nsp-card-excerpt\">Development of a structured Generative AI audit framework. This included establishing a methodology for Level 1 and Level 2 GenAI audits<\/p>\n\t\t\t\t\t\t\t\t\t\t<span class=\"nsp-card-more\">read more\u2026<\/span>\n\t\t\t\t\t<\/article>\n\t\t\t\t<article class=\"nsp-card\" tabindex=\"0\" role=\"button\" data-id=\"12\" aria-label=\"Intelligent Data Retrieval Agent\">\n\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-card-iconwrap\"><span class=\"nsp-card-icon et-pb-icon\" aria-hidden=\"true\">&#xe0f7;<\/span><\/div>\n\t\t\t\t\t\t<h3 class=\"nsp-card-title\">Intelligent Data Retrieval Agent<\/h3>\n\t\t\t\t\t\t\t\t\t\t<p class=\"nsp-card-excerpt\">Production-ready AI retrieval system using LLMs and semantic search to transform fragmented data into reliable, searchable knowledge.<\/p>\n\t\t\t\t\t\t\t\t\t\t<span class=\"nsp-card-more\">read more\u2026<\/span>\n\t\t\t\t\t<\/article>\n\t\t\t\t<article class=\"nsp-card\" tabindex=\"0\" role=\"button\" data-id=\"23\" aria-label=\"RAG-Based Enterprise Knowledge Assistant\">\n\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-card-iconwrap\"><i class=\"nsp-card-icon fa-regular fa-folder-open\" aria-hidden=\"true\"><\/i><\/div>\n\t\t\t\t\t\t<h3 class=\"nsp-card-title\">RAG-Based Enterprise Knowledge Assistant<\/h3>\n\t\t\t\t\t\t\t\t\t\t<p class=\"nsp-card-excerpt\">Designed RAG-based AI assistants that turn fragmented documents and expert knowledge into accessible, context-aware enterprise knowledge.<\/p>\n\t\t\t\t\t\t\t\t\t\t<span class=\"nsp-card-more\">read more\u2026<\/span>\n\t\t\t\t\t<\/article>\n\t\t\t\t<article class=\"nsp-card\" tabindex=\"0\" role=\"button\" data-id=\"2\" aria-label=\"Customer Interaction Analysis\">\n\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-card-iconwrap\"><span class=\"nsp-card-icon et-pb-icon\" aria-hidden=\"true\">&#x77;<\/span><\/div>\n\t\t\t\t\t\t<h3 class=\"nsp-card-title\">Customer Interaction Analysis<\/h3>\n\t\t\t\t\t\t\t\t\t\t<p class=\"nsp-card-excerpt\">SaaS project platform: the objective was to evaluate dialogue quality using an AI model to ensure timely intervention and customer care.<\/p>\n\t\t\t\t\t\t\t\t\t\t<span class=\"nsp-card-more\">read more\u2026<\/span>\n\t\t\t\t\t<\/article>\n\t\t<\/div><button type=\"button\" class=\"nsp-nav nsp-next\" aria-label=\"Next\"><span class=\"nsp-nav-icon et-pb-icon\" aria-hidden=\"true\">&#x35;<\/span><\/button><\/div><script>(function(){\n\t\t\tif (!window.nspComputeLayout) {\n\t\t\t\twindow.NSP_BASE_CARD_WIDTH = 240;\n\t\t\t\twindow.nspIsFluidWidth = function(){ return window.innerWidth > 640; };\n\t\t\t\twindow.nspComputeLayout = function(available, gap, totalCards, baseWidth){\n\t\t\t\t\tvar BASE = baseWidth || window.NSP_BASE_CARD_WIDTH;\n\t\t\t\t\tavailable = Math.max(0, available || 0);\n\t\t\t\t\tgap = Math.max(0, gap || 0);\n\t\t\t\t\ttotalCards = Math.max(0, totalCards || 0);\n\t\t\t\t\tif (!totalCards){\n\t\t\t\t\t\treturn { count: 0, width: BASE, center: true };\n\t\t\t\t\t}\n\t\t\t\t\tvar per = BASE + gap;\n\t\t\t\t\tvar baseCount = Math.max(1, Math.floor((available + gap) \/ per));\n\t\t\t\t\tvar used = baseCount * BASE + (baseCount - 1) * gap;\n\t\t\t\t\tvar remainder = available - used;\n\t\t\t\t\t\/\/ Only create an extra visible slot when there is actually another card to fill it.\n\t\t\t\t\t\/\/ If exactly the base-width row count is present, distribute those cards evenly\n\t\t\t\t\t\/\/ across the row instead of leaving them beside a phantom extra slot.\n\t\t\t\t\tvar fullCount = (totalCards > baseCount && remainder >= (2 \/ 3) * BASE) ? baseCount + 1 : baseCount;\n\t\t\t\t\tfullCount = Math.max(1, fullCount);\n\t\t\t\t\tvar fullWidth = (available - (fullCount - 1) * gap) \/ fullCount;\n\t\t\t\t\tif (!isFinite(fullWidth) || fullWidth <= 0){ fullWidth = BASE; }\n\n\t\t\t\t\t\/\/ One or two cards should never grow just because the row is wide.\n\t\t\t\t\t\/\/ They keep the base width, sit at the left, and leave the rest of the row empty.\n\t\t\t\t\tif (totalCards < 3){\n\t\t\t\t\t\tvar smallMax = (available - (totalCards - 1) * gap) \/ totalCards;\n\t\t\t\t\t\tvar smallWidth = (isFinite(smallMax) && smallMax > 0) ? Math.min(BASE, smallMax) : BASE;\n\t\t\t\t\t\treturn { count: totalCards, width: smallWidth, center: false };\n\t\t\t\t\t}\n\n\t\t\t\t\t\/\/ If there are fewer cards than a full row would hold, keep the full-row\n\t\t\t\t\t\/\/ card width rather than stretching those few across the whole row, and\n\t\t\t\t\t\/\/ leave the remaining slots empty. Left-aligned, as the master grid is.\n\t\t\t\t\tif (totalCards < fullCount){\n\t\t\t\t\t\treturn { count: totalCards, width: fullWidth, center: false };\n\t\t\t\t\t}\n\n\t\t\t\t\tvar count = Math.min(fullCount, totalCards);\n\t\t\t\t\treturn { count: count, width: fullWidth, center: false };\n\t\t\t\t};\n\t\t\t\t\/\/ Centre the chevrons on the media block of a card rather than on\n\t\t\t\t\/\/ the whole card: on people cards that is the square photo\n\t\t\t\t\/\/ (whose height tracks the fluid card width, so it cannot be\n\t\t\t\t\/\/ expressed in CSS), on project cards the icon block. Published\n\t\t\t\t\/\/ as a custom property the stylesheet reads.\n\t\t\t\twindow.nspTrackPadX = function(track){\n\t\t\t\t\tvar cs = window.getComputedStyle(track);\n\t\t\t\t\treturn (parseFloat(cs.paddingLeft) || 0) + (parseFloat(cs.paddingRight) || 0);\n\t\t\t\t};\n\t\t\t\twindow.nspSyncNav = function(wrap){\n\t\t\t\t\t\/\/ People cards anchor on the photo, which is the visual mass of\n\t\t\t\t\t\/\/ the card. Project cards have no photo -- only a small icon\n\t\t\t\t\t\/\/ at the top -- so they anchor on the card panel itself and\n\t\t\t\t\t\/\/ the arrows land on its vertical centre. Measuring against\n\t\t\t\t\t\/\/ the wrap in viewport coordinates absorbs the track padding\n\t\t\t\t\t\/\/ and card margins without restating any of them here.\n\t\t\t\t\tvar media = wrap.querySelector(\".nsp-card-photo\")\n\t\t\t\t\t\t|| wrap.querySelector(\".nsp-card\");\n\t\t\t\t\tif (!media){ return; }\n\t\t\t\t\tvar r = media.getBoundingClientRect();\n\t\t\t\t\tif (!r.height){ return; }\n\t\t\t\t\tvar w = wrap.getBoundingClientRect();\n\t\t\t\t\twrap.style.setProperty(\"--nsp-nav-top\", ((r.top - w.top) + r.height \/ 2) + \"px\");\n\t\t\t\t};\n\t\t\t}\n\t\t\tvar wrap = document.currentScript.previousElementSibling;\n\t\t\tif(!wrap){ return; }\n\t\t\t\/\/ The frame carries the visible panel, so it has to be revealed\n\t\t\t\/\/ together with the wrap -- otherwise an empty framed box paints\n\t\t\t\/\/ before the cards have been sized into it.\n\t\t\tfunction reveal(w){\n\t\t\t\tw.style.visibility = \"visible\";\n\t\t\t\tvar f = w.parentNode;\n\t\t\t\tif (f && f.classList && f.classList.contains(\"nsp-frame\")){ f.style.visibility = \"visible\"; }\n\t\t\t}\n\t\t\tvar baseWidth = parseInt(wrap.getAttribute(\"data-base-width\"), 10) || window.NSP_BASE_CARD_WIDTH;\n\t\t\tvar track = wrap.querySelector(\".nsp-carousel\");\n\t\t\tvar cards = track ? track.querySelectorAll(\".nsp-card\") : [];\n\t\t\tif(!track || !cards.length){ if (window.nspSyncNav){ window.nspSyncNav(wrap); }\n\t\t\treveal(wrap); return; }\n\t\t\t\/\/ Read from the stylesheet rather than restating it: a hard-coded copy\n\t\t\t\/\/ here is exactly how the gutter drifted from the CSS before.\n\t\t\tvar gap = parseFloat(window.getComputedStyle(track).columnGap) || 24;\n\t\t\tif (!window.nspIsFluidWidth()) {\n\t\t\t\t\/\/ Mobile: unchanged -- fixed per-card width (CSS 82vw), just avoid a partial\n\t\t\t\t\/\/ card peeking out at the wrap edge.\n\t\t\t\tvar cardW = cards[0].getBoundingClientRect().width;\n\t\t\t\tif(!cardW){ if (window.nspSyncNav){ window.nspSyncNav(wrap); }\n\t\t\treveal(wrap); return; }\n\t\t\t\t\/\/ clientWidth is the PADDING box, so the track padding has to come\n\t\t\t\t\/\/ off before this is the room the cards actually get.\n\t\t\t\tvar available0 = track.clientWidth - window.nspTrackPadX(track);\n\t\t\t\tvar fitCount = Math.max(1, Math.floor((available0 + gap) \/ (cardW + gap)));\n\t\t\t\tvar count0 = Math.min(fitCount, cards.length);\n\t\t\t\tvar needed0 = count0 * cardW + (count0 - 1) * gap;\n\t\t\t\tvar ws0 = window.getComputedStyle(wrap);\n\t\t\t\tvar padX0 = (parseFloat(ws0.paddingLeft) || 0) + (parseFloat(ws0.paddingRight) || 0);\n\t\t\t\tvar target0 = needed0 + window.nspTrackPadX(track) + (ws0.boxSizing === \"border-box\" ? padX0 : 0);\n\t\t\t\twrap.style.maxWidth = Math.ceil(target0) + \"px\";\n\t\t\t\tif (window.nspSyncNav){ window.nspSyncNav(wrap); }\n\t\t\treveal(wrap);\n\t\t\t\treturn;\n\t\t\t}\n\t\t\t\/\/ Desktop\/tablet: fluid card width, filling the row exactly either way.\n\t\t\t\/\/ clientWidth is the PADDING box -- take the track padding off it.\n\t\t\tvar available = track.clientWidth - window.nspTrackPadX(track);\n\t\t\tvar layout = window.nspComputeLayout(available, gap, cards.length, baseWidth);\n\t\t\tfor (var i = 0; i < cards.length; i++){ cards[i].style.flex = \"0 0 \" + layout.width + \"px\"; }\n\t\t\tif (layout.center) {\n\t\t\t\tvar totalW = layout.count * layout.width + (layout.count - 1) * gap;\n\t\t\t\twrap.style.maxWidth = Math.ceil(totalW + window.nspTrackPadX(track)) + \"px\";\n\t\t\t\twrap.style.marginLeft = \"auto\";\n\t\t\t\twrap.style.marginRight = \"auto\";\n\t\t\t}\n\t\t\tif (window.nspSyncNav){ window.nspSyncNav(wrap); }\n\t\t\treveal(wrap);\n\t\t})();<\/script><\/div><noscript><style>.nsp-carousel-wrap,.nsp-frame{visibility:visible !important;}<\/style><\/noscript><div class=\"nsp-popups\" hidden><template class=\"nsp-popup-tpl\" data-id=\"3\">\t\t<div class=\"nsp-dialog no-image\">\n\t\t\t<button type=\"button\" class=\"nsp-close\" aria-label=\"Close\">&times;<\/button>\n\t\t\t<div class=\"nsp-dlg-h\">\n\t\t\t\t<i class=\"nsp-pop-icon fa-solid fa-section\" aria-hidden=\"true\"><\/i>\t\t\t\t<span class=\"nsp-dlg-h-title\">RAG-driven Legal Chatbot<\/span>\n\t\t\t<\/div>\n\t\t\t<div class=\"nsp-pop-cols\">\n\t\t\t\t<div class=\"nsp-pop-main\">\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-desc\"><p>An AI-supported RAG-Chatbot developed for legal and administrative workflows. Designed to support caseworkers in navigating complex regulations - currently focused on German Social Welfare - the system provides fast, contextual access to relevant legal information and assists in decision-making for applications and case management. The modular architecture allows seamless expansion into additional legal domains and regulatory frameworks.<\/p>\n<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\n\t\t\t\t\t\t\t<ul class=\"nsp-chiplist nsp-pop-caps\">\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/nlp\">Natural Language Processing<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t<\/ul>\n\t\t\t\t\t<\/div>\n\t\t<\/template><template class=\"nsp-popup-tpl\" data-id=\"7\">\t\t<div class=\"nsp-dialog no-image\">\n\t\t\t<button type=\"button\" class=\"nsp-close\" aria-label=\"Close\">&times;<\/button>\n\t\t\t<div class=\"nsp-dlg-h\">\n\t\t\t\t<span class=\"nsp-pop-icon et-pb-icon\" aria-hidden=\"true\">&#xe0f7;<\/span>\t\t\t\t<span class=\"nsp-dlg-h-title\">Generative AI Audit Framework<\/span>\n\t\t\t<\/div>\n\t\t\t<div class=\"nsp-pop-cols\">\n\t\t\t\t<div class=\"nsp-pop-main\">\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-desc\"><p>Development of a multidimensional Blackbox Chatbot Audit framework for evaluating chatbot user experience and business value in customer service environments. The audit methodology combines structured use-case testing with qualitative and quantitative evaluation dimensions, including answer quality, response speed, dialogue quality, and user interface assessment.<\/p>\n<p>The framework also incorporates hallucination testing and edge-case analysis to assess robustness and real-world usability. The project included extensive market and user-frustration research, methodology development, pilot implementation, and iterative testing and retesting phases. The resulting audit framework is used to evaluate chatbot performance, identify optimization potential, and assess user retention likelihood and overall business impact.<\/p>\n<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\n\t\t\t\t\t\t\t<ul class=\"nsp-chiplist nsp-pop-caps\">\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/governance-and-ethics\">Governance<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/nlp\">Natural Language Processing<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t<\/ul>\n\t\t\t\t\t<\/div>\n\t\t<\/template><template class=\"nsp-popup-tpl\" data-id=\"12\">\t\t<div class=\"nsp-dialog no-image\">\n\t\t\t<button type=\"button\" class=\"nsp-close\" aria-label=\"Close\">&times;<\/button>\n\t\t\t<div class=\"nsp-dlg-h\">\n\t\t\t\t<span class=\"nsp-pop-icon et-pb-icon\" aria-hidden=\"true\">&#xe0f7;<\/span>\t\t\t\t<span class=\"nsp-dlg-h-title\">Intelligent Data Retrieval Agent<\/span>\n\t\t\t<\/div>\n\t\t\t<div class=\"nsp-pop-cols\">\n\t\t\t\t<div class=\"nsp-pop-main\">\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-desc\"><p>Development of a production-ready retrieval agent for querying large, fragmented, and undocumented enterprise data repositories using large language models and semantic search. The solution transformed scattered legacy records into structured, searchable knowledge, enabling users to retrieve relevant information within seconds instead of manually searching across hundreds of documents.<\/p>\n<p>The project combined retrieval-augmented generation (RAG), vector search, and modern LLM technologies to deliver reliable, context-aware information retrieval. Designed with a modular architecture, the system supports scalability, maintainability, and future extensions while ensuring robust retrieval quality across heterogeneous data sources.<\/p>\n<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-role nsp-pop-role--below\"><p>AI Engineer<\/p>\n<p>Python \u2022 LangChain \u2022 OpenAI \u2022 Qdrant \u2022 Semantic Search \u2022 RAG \u2022 Streamlit<\/p>\n<\/div>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\n\t\t\t\t\t\t\t<ul class=\"nsp-chiplist nsp-pop-caps\">\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/machine-learning\">Machine Learning<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/nlp\">Natural Language Processing<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/control-and-automation\">Automation<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t<\/ul>\n\t\t\t\t\t<\/div>\n\t\t<\/template><template class=\"nsp-popup-tpl\" data-id=\"23\">\t\t<div class=\"nsp-dialog no-image\">\n\t\t\t<button type=\"button\" class=\"nsp-close\" aria-label=\"Close\">&times;<\/button>\n\t\t\t<div class=\"nsp-dlg-h\">\n\t\t\t\t<i class=\"nsp-pop-icon fa-regular fa-folder-open\" aria-hidden=\"true\"><\/i>\t\t\t\t<span class=\"nsp-dlg-h-title\">RAG-Based Enterprise Knowledge Assistant<\/span>\n\t\t\t<\/div>\n\t\t\t<div class=\"nsp-pop-cols\">\n\t\t\t\t<div class=\"nsp-pop-main\">\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-desc\"><p>Gerold designed and developed concepts and prototypes for RAG-based AI assistants that make fragmented organizational knowledge accessible through natural-language interaction.<\/p>\n<p>The solutions combine internal documents, structured information and expert knowledge with Large Language Models, semantic search and retrieval-augmented generation. The focus is not only on the underlying technology, but on creating a reliable end-to-end solution: from identifying and structuring relevant knowledge sources to retrieval architecture, user experience, access concepts and governance.<\/p>\n<p>The work demonstrates how Generative AI can transform static document repositories into practical knowledge systems that support employees in finding information faster, preserving expert knowledge and making better-informed decisions.<\/p>\n<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-role nsp-pop-role--below\"><p>AI Strategy &amp; Solution Design, Generative AI, LLMs, RAG, Prompt Design, Knowledge Management<\/p>\n<\/div>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\n\t\t\t\t\t\t\t<ul class=\"nsp-chiplist nsp-pop-caps\">\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/ai-strategy\">AI Strategy<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/consulting\">AI Transformation<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/nlp\">Natural Language Processing<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t<\/ul>\n\t\t\t\t\t<\/div>\n\t\t<\/template><template class=\"nsp-popup-tpl\" data-id=\"2\">\t\t<div class=\"nsp-dialog no-image\">\n\t\t\t<button type=\"button\" class=\"nsp-close\" aria-label=\"Close\">&times;<\/button>\n\t\t\t<div class=\"nsp-dlg-h\">\n\t\t\t\t<span class=\"nsp-pop-icon et-pb-icon\" aria-hidden=\"true\">&#x77;<\/span>\t\t\t\t<span class=\"nsp-dlg-h-title\">Customer Interaction Analysis<\/span>\n\t\t\t<\/div>\n\t\t\t<div class=\"nsp-pop-cols\">\n\t\t\t\t<div class=\"nsp-pop-main\">\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-desc\"><p>For a project management SaaS solution where customers were matched with freelancers, a custom AI solution was established with the purpose to improve experiences for all parties. Key purposes were to create an early warning system to help customer service intervene in case of issues:<\/p>\n<ul>\n<li>Identification of unusual patterns (delays indicating inaction, intense exchanges);<\/li>\n<li>Flagging of language transgressions on both sides (use of inappropriate language, aggression);<\/li>\n<li>Matching of final ratings with evaluation of flow and dialogue quality to foster a more honest rating culture;<\/li>\n<li>Language style matching to improve future matching of freelancers to clients;<\/li>\n<\/ul>\n<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-role nsp-pop-role--below\"><p>The solution was implemented using Python on a LAMP stack, with self-developed machine learning libraries.<\/p>\n<\/div>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\n\t\t\t\t\t\t\t<ul class=\"nsp-chiplist nsp-pop-caps\">\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/machine-learning\">Machine Learning<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/nlp\">Natural Language Processing<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t<\/ul>\n\t\t\t\t\t<\/div>\n\t\t<\/template><\/div><div class=\"nsp-overlay\" id=\"nsp-overlay\" role=\"dialog\" aria-modal=\"true\"><\/div><\/div><\/div>\n<\/div>\n<\/div>\n<\/div>\n\n<div class=\"et_pb_section_7 et_pb_section et_section_regular et_flex_section ns-panel\" id=\"cv\" style=\"--ns-rail-w:264px; --ns-body-color:#D3D3D3;\">\n<div class=\"et_pb_row_12 et_pb_row et_flex_row\">\n<div class=\"et_pb_column_19 et_pb_column et_block_column et_pb_css_mix_blend_mode_passthrough et_flex_column_24_24 et_flex_column_24_24_tabletWide\">\n<div class=\"et_pb_text_11 et_pb_text et_pb_bg_layout_light et_pb_module et_flex_module\"><div class=\"et_pb_text_inner\"><p>As formidable as the interplay of the human eye, brain and muscles is, it evolved to focus on what matters and filter out the rest. Machine vision has no such filter: it inspects every pixel with the same attention, frame after frame, around the clock. It never gets tired, and it works reliably in environments where humans are unsafe or uncomfortable.<\/p>\n<\/div><\/div>\n\n<div class=\"et_pb_code_6 et_pb_code et_pb_module\"><div class=\"et_pb_code_inner\">\t<div\n\t\tid=\"ns-aix-3\"\n\t\tclass=\"ns-aix ns-aix-cv\"\n\t\tdata-topic=\"cv\"\n\t\tdata-flow=\"core\"\n\t\t\t\tdata-step=\"0\"\n\t\tdata-interval=\"9000\"\n\t\t\t\tdata-autoplay=\"1\"\n\t\trole=\"region\"\n\t\taria-label=\"Comment fonctionne la vision par ordinateur\"\n\t\tstyle=\"--ns-aix-accent:#58a7f9;--ns-aix-secondary:#2b6cb0\"\n\t>\n\t\t<header class=\"ns-aix-head\">\n\t\t\t\t\t\t<h3 class=\"ns-aix-title\">Comment fonctionne la vision par ordinateur<\/h3>\n\t\t\t<p class=\"ns-aix-intro\">La vision par ordinateur transforme les signaux visuels en donn\u00e9es num\u00e9riques \u2014 une image est une grille de nombres \u2014 puis en extrait les motifs utiles \u00e0 la t\u00e2che et produit des r\u00e9sultats structur\u00e9s : \u00e9tiquettes, positions, masques, texte, mesures ou mouvements. Elle ne voit pas comme un humain ; elle estime des probabilit\u00e9s \u00e0 partir d\u2019indices visuels appris pour une t\u00e2che donn\u00e9e.<\/p>\n\t\t\t\t\t\t\t\t<\/header>\n\n\t\t<div class=\"ns-aix-body\">\n\t\t\t<div class=\"ns-aix-copy\">\n\t\t\t\t<div class=\"ns-aix-progress\" aria-hidden=\"true\"><span><\/span><\/div>\n\t\t\t\t\t\t\t\t\t<div\n\t\t\t\t\t\tid=\"ns-aix-3-flow-panel-core\"\n\t\t\t\t\t\tclass=\"ns-aix-step-panel ns-on\"\n\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\taria-hidden=\"false\"\n\t\t\t\t\t\t\t\t\t\t\t>\n\t\t\t\t\t<ol class=\"ns-aix-steps ns-on\" data-flow=\"core\" role=\"list\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step ns-on\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\t\t\tdata-step=\"0\"\n\t\t\t\t\t\t\t\t\tdata-key=\"capture\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"true\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-3-panel-core-0\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">01<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Capturer<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">La lumi\u00e8re et d\u2019autres signaux deviennent des entr\u00e9es.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\t\t\tdata-step=\"1\"\n\t\t\t\t\t\t\t\t\tdata-key=\"encode\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-3-panel-core-1\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">02<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Encoder<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Pixels, canaux et images.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\t\t\tdata-step=\"2\"\n\t\t\t\t\t\t\t\t\tdata-key=\"prepare\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-3-panel-core-2\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">03<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Pr\u00e9parer<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Corriger le bruit, l\u2019\u00e9chelle et le point de vue.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\t\t\tdata-step=\"3\"\n\t\t\t\t\t\t\t\t\tdata-key=\"learn\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-3-panel-core-3\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">04<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Apprendre<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Les contours deviennent formes et objets.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\t\t\tdata-step=\"4\"\n\t\t\t\t\t\t\t\t\tdata-key=\"recognize\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-3-panel-core-4\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">05<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Reconna\u00eetre<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Classer, d\u00e9tecter et rep\u00e9rer les anomalies.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\t\t\tdata-step=\"5\"\n\t\t\t\t\t\t\t\t\tdata-key=\"map\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-3-panel-core-5\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">06<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Cartographier<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Masques, points de rep\u00e8re, profondeur et texte.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\t\t\tdata-step=\"6\"\n\t\t\t\t\t\t\t\t\tdata-key=\"track\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-3-panel-core-6\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">07<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Suivre<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Identit\u00e9 et mouvement persistent d\u2019une image \u00e0 l\u2019autre.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\t\t\tdata-step=\"7\"\n\t\t\t\t\t\t\t\t\tdata-key=\"act\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-3-panel-core-7\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">08<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Agir<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Mesurer, compter, alerter ou piloter.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\t\t\tdata-step=\"8\"\n\t\t\t\t\t\t\t\t\tdata-key=\"validate\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-3-panel-core-8\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">09<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Valider<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Performance, biais, d\u00e9rive et supervision.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t<\/ol>\n\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\n\t\t\t<div class=\"ns-aix-stage\" aria-label=\"Animated vector-map process diagram\">\n\t\t\t\t<canvas class=\"ns-aix-canvas\" aria-hidden=\"true\"><\/canvas>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node ns-on\" data-flow=\"core\" data-step=\"0\" data-key=\"capture\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>capteur<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node\" data-flow=\"core\" data-step=\"1\" data-key=\"encode\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>pixels<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node\" data-flow=\"core\" data-step=\"2\" data-key=\"prepare\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>pr\u00e9parer<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node\" data-flow=\"core\" data-step=\"3\" data-key=\"learn\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>caract\u00e9ristiques<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node\" data-flow=\"core\" data-step=\"4\" data-key=\"recognize\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>d\u00e9tecter<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node\" data-flow=\"core\" data-step=\"5\" data-key=\"map\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>carte<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node\" data-flow=\"core\" data-step=\"6\" data-key=\"track\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>suivre<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node\" data-flow=\"core\" data-step=\"7\" data-key=\"act\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>agir<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node\" data-flow=\"core\" data-step=\"8\" data-key=\"validate\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>v\u00e9rifier<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\n\t\t\t\t\t<\/div>\n\n\t\t<div class=\"ns-aix-controls\">\n\t\t\t<div class=\"ns-aix-dots\" role=\"navigation\" aria-label=\"Step navigation\">\n\t\t\t\t\t\t\t\t\t<span\n\t\t\t\t\t\tclass=\"ns-aix-dotset ns-on\"\n\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\trole=\"group\"\n\t\t\t\t\t\taria-label=\"Steps for Comment fonctionne la vision par ordinateur\"\n\t\t\t\t\t\taria-hidden=\"false\"\n\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot ns-on\"\n\t\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\t\tdata-step=\"0\"\n\t\t\t\t\t\t\t\taria-label=\"Step 1: Capturer\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-3-panel-core-0\"\n\t\t\t\t\t\t\t\taria-pressed=\"true\"\n\t\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\t\tdata-step=\"1\"\n\t\t\t\t\t\t\t\taria-label=\"Step 2: Encoder\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-3-panel-core-1\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\t\tdata-step=\"2\"\n\t\t\t\t\t\t\t\taria-label=\"Step 3: Pr\u00e9parer\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-3-panel-core-2\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\t\tdata-step=\"3\"\n\t\t\t\t\t\t\t\taria-label=\"Step 4: Apprendre\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-3-panel-core-3\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\t\tdata-step=\"4\"\n\t\t\t\t\t\t\t\taria-label=\"Step 5: Reconna\u00eetre\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-3-panel-core-4\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\t\tdata-step=\"5\"\n\t\t\t\t\t\t\t\taria-label=\"Step 6: Cartographier\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-3-panel-core-5\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\t\tdata-step=\"6\"\n\t\t\t\t\t\t\t\taria-label=\"Step 7: Suivre\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-3-panel-core-6\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\t\tdata-step=\"7\"\n\t\t\t\t\t\t\t\taria-label=\"Step 8: Agir\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-3-panel-core-7\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\t\tdata-step=\"8\"\n\t\t\t\t\t\t\t\taria-label=\"Step 9: Valider\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-3-panel-core-8\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t<\/div>\n\t\t<\/div>\n\n\t\t<div class=\"ns-aix-below\" aria-live=\"polite\">\n\t\t\t<div class=\"ns-aix-explain\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-3-panel-core-0\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel ns-on\"\n\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\tdata-step=\"0\"\n\t\t\t\t\t\t\taria-hidden=\"false\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">01<\/span>Mesurer la sc\u00e8ne<\/h4>\n\t\t\t\t\t\t\t<p>La vision par ordinateur commence par un capteur : cam\u00e9ra, scanner, microscope, satellite, appareil \u00e0 rayons X, cam\u00e9ra thermique ou capteur de profondeur. L\u2019objectif, l\u2019angle de vue, l\u2019exposition, la r\u00e9solution et la fr\u00e9quence d\u2019images d\u00e9terminent les informations qui entrent dans le syst\u00e8me. Un d\u00e9tail jamais captur\u00e9 ne peut pas \u00eatre reconstruit de fa\u00e7on fiable par la suite.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-3-panel-core-1\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\tdata-step=\"1\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">02<\/span>Encoder en pixels<\/h4>\n\t\t\t\t\t\t\t<p>Une image num\u00e9rique est une grille de pixels. Chaque pixel stocke des valeurs de canal \u2014 rouge, vert et bleu, intensit\u00e9 en niveaux de gris, r\u00e9ponse infrarouge ou profondeur. La vid\u00e9o ajoute le temps sous forme d\u2019une succession d\u2019images. R\u00e9solution, profondeur de couleur et compression d\u00e9terminent la quantit\u00e9 d\u2019information visuelle conserv\u00e9e ou perdue.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-3-panel-core-2\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\tdata-step=\"2\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">03<\/span>Pr\u00e9parer l\u2019entr\u00e9e<\/h4>\n\t\t\t\t\t\t\t<p>Les images peuvent \u00eatre redimensionn\u00e9es, recadr\u00e9es, d\u00e9bruit\u00e9es, accentu\u00e9es, corrig\u00e9es des distorsions optiques ou de perspective, et normalis\u00e9es en couleur ou en \u00e9clairage. Les donn\u00e9es d\u2019entra\u00eenement sont souvent enrichies par des variations r\u00e9alistes. La pr\u00e9paration doit correspondre aux conditions d\u2019exploitation.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-3-panel-core-3\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\tdata-step=\"3\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">04<\/span>Apprendre les motifs visuels<\/h4>\n\t\t\t\t\t\t\t<p>Les syst\u00e8mes classiques de vision s\u2019appuient sur des contours, angles et mod\u00e8les con\u00e7us \u00e0 la main. Les r\u00e9seaux convolutifs et les Vision Transformers modernes apprennent des hi\u00e9rarchies de caract\u00e9ristiques \u00e0 partir d\u2019exemples : de simples contrastes se combinent en textures, formes, parties d\u2019objets et relations spatiales. Cet apprentissage a lieu une fois, avant le d\u00e9ploiement.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-3-panel-core-4\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\tdata-step=\"4\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">05<\/span>Reconna\u00eetre objets et anomalies<\/h4>\n\t\t\t\t\t\t\t<p>La classification attribue une \u00e9tiquette \u00e0 une image enti\u00e8re. La d\u00e9tection d\u2019objets rep\u00e8re des instances individuelles et leur position, g\u00e9n\u00e9ralement avec des bo\u00eetes englobantes et des scores de confiance. La d\u00e9tection d\u2019anomalies apprend plut\u00f4t \u00e0 quoi ressemblent des donn\u00e9es visuelles normales et signale les \u00e9carts. Un score de confiance \u00e9lev\u00e9 ne signifie pas une forte probabilit\u00e9 d\u2019avoir raison.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-3-panel-core-5\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\tdata-step=\"5\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">06<\/span>Cartographier r\u00e9gions, profondeur et texte<\/h4>\n\t\t\t\t\t\t\t<p>La segmentation attribue une cat\u00e9gorie ou une identit\u00e9 d\u2019objet aux pixels. Les mod\u00e8les de points cl\u00e9s et de pose localisent des rep\u00e8res ; les mod\u00e8les de profondeur estiment la distance et la structure tridimensionnelle ; les syst\u00e8mes OCR et d\u2019analyse de mise en page reconstituent caract\u00e8res, champs et ordre de lecture.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-3-panel-core-6\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\tdata-step=\"6\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">07<\/span>Suivre les objets dans le temps<\/h4>\n\t\t\t\t\t\t\t<p>Les syst\u00e8mes vid\u00e9o relient les d\u00e9tections d\u2019une image \u00e0 l\u2019autre pour conserver une identit\u00e9 stable. Les trajectoires permettent ensuite d\u2019estimer direction, vitesse, flux, temps de pr\u00e9sence ou actions, ou encore de compter les objets qui franchissent une limite.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-3-panel-core-7\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\tdata-step=\"7\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">08<\/span>Transformer la perception en action<\/h4>\n\t\t\t\t\t\t\t<p>La sortie de la vision par ordinateur devient utile lorsqu\u2019un autre processus l\u2019exploite : compter ou trier un objet, alimenter un tableau de donn\u00e9es, alerter un op\u00e9rateur, d\u00e9clencher une maintenance ou guider un robot. Des seuils transforment les probabilit\u00e9s en actions. Lorsqu\u2019il s\u2019agit de personnes, protection des donn\u00e9es et consentement limitent ce que le syst\u00e8me peut capter, conserver ou utiliser pour agir.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-3-panel-core-8\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"core\"\n\t\t\t\t\t\t\tdata-step=\"8\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">09<\/span>Tester dans le monde r\u00e9el<\/h4>\n\t\t\t\t\t\t\t<p>Un syst\u00e8me de vision doit \u00eatre test\u00e9 sur des donn\u00e9es s\u00e9par\u00e9es, dans diff\u00e9rentes conditions d\u2019\u00e9clairage et de m\u00e9t\u00e9o, avec plusieurs positions de cam\u00e9ra, groupes d\u00e9mographiques, cas rares et perturbations volontaires. Pr\u00e9cision, rappel, faux positifs, faux n\u00e9gatifs, latence et calibration comptent tous. Le monitoring et les corrections humaines permettent de d\u00e9tecter la d\u00e9rive et d\u2019am\u00e9liorer le syst\u00e8me en s\u00e9curit\u00e9.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t<div class=\"ns-aix-aspects\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset ns-on\" data-flow=\"core\" data-step=\"0\" aria-hidden=\"false\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Contexte<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Les premi\u00e8res applications de vision par ordinateur apparaissent dans les ann\u00e9es 1960 et 1970 avec des programmes capables de reconna\u00eetre des formes simples \u2014 et avec la reconnaissance optique de caract\u00e8res (OCR) pour les documents dactylographi\u00e9s, parmi les premiers usages pratiques.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>En pratique<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Le choix de la cam\u00e9ra, son emplacement et l\u2019\u00e9clairage d\u00e9cident souvent de la r\u00e9ussite d\u2019un projet avant m\u00eame qu\u2019un mod\u00e8le soit entra\u00een\u00e9.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"core\" data-step=\"1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Concept cl\u00e9<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Les syst\u00e8mes de vision extraient du sens de pixels repr\u00e9sent\u00e9s par des valeurs num\u00e9riques et rep\u00e8rent des motifs dans ces nombres pour identifier objets, formes, textures et mouvements.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Pourquoi c\u2019est important<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>La compression et une faible r\u00e9solution \u00e9liminent discr\u00e8tement des informations dont le mod\u00e8le pourrait avoir besoin plus tard \u2014 ce qui est perdu ici l\u2019est d\u00e9finitivement.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"core\" data-step=\"2\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Comment \u00e7a marche<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Pendant l\u2019entra\u00eenement, le mod\u00e8le voit de nombreux exemples \u00e9tiquet\u00e9s, fait des pr\u00e9dictions, les compare aux bonnes \u00e9tiquettes et ajuste ses param\u00e8tres internes pour r\u00e9duire les erreurs.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Attention<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Un nettoyage trop pouss\u00e9 peut supprimer pr\u00e9cis\u00e9ment le d\u00e9faut ou le signal que le syst\u00e8me doit d\u00e9tecter.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"core\" data-step=\"3\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Contexte<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Les progr\u00e8s se sont acc\u00e9l\u00e9r\u00e9s dans les ann\u00e9es 2000 gr\u00e2ce \u00e0 de meilleurs mat\u00e9riels et de grands jeux de donn\u00e9es ; une avanc\u00e9e majeure a eu lieu en 2012, lorsque le Deep Learning a fortement am\u00e9lior\u00e9 la pr\u00e9cision de reconnaissance.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Concept cl\u00e9<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Les r\u00e9seaux neuronaux profonds apprennent directement des motifs visuels \u00e0 partir de grandes collections d\u2019images \u00e9tiquet\u00e9es, sans d\u00e9pendre de r\u00e8gles \u00e9crites \u00e0 la main.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"core\" data-step=\"4\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Ce que la vision par ordinateur peut faire<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Rep\u00e9rer des irr\u00e9gularit\u00e9s sur une ligne de production, d\u00e9tecter des intrus, classer des objets ou reconna\u00eetre des visages \u2014 et souvent voir des diff\u00e9rences que l\u2019\u0153il humain ne remarque pas.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Attention<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>La pr\u00e9cision peut varier entre groupes d\u00e9mographiques lorsque les donn\u00e9es d\u2019entra\u00eenement sont biais\u00e9es \u2014 un risque connu et s\u00e9rieux en reconnaissance faciale.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"core\" data-step=\"5\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Ce que la vision par ordinateur peut faire<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Combin\u00e9 \u00e0 l\u2019IA, l\u2019OCR peut traiter presque n\u2019importe quel document m\u00e9tier, remplir correctement des tableaux de donn\u00e9es et aider \u00e0 d\u00e9cider quoi faire des informations re\u00e7ues.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>En pratique<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Segmentation, points de rep\u00e8re et profondeur servent \u00e0 des t\u00e2ches de mesure, d\u2019inspection, de navigation et de traitement documentaire.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"core\" data-step=\"6\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Ce que la vision par ordinateur peut faire<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Compter des personnes dans une foule, suivre des objets de nuit ou \u00e0 grande vitesse, mesurer des flux \u2014 souvent en rempla\u00e7ant des capteurs co\u00fbteux ou des heures de travail humain.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Attention<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Occlusion, flou de mouvement, d\u00e9placement de la cam\u00e9ra ou r\u00e9apparition d\u2019un objet peuvent rompre un suivi et fausser les comptages.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"core\" data-step=\"7\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Concept cl\u00e9<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Le syst\u00e8me ne \u00ab reconna\u00eet \u00bb pas les objets au sens humain : il calcule des probabilit\u00e9s \u00e0 partir de motifs visuels appris. Leur faire aveugl\u00e9ment confiance peut produire des erreurs dangereuses.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Pourquoi c\u2019est important<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Le co\u00fbt d\u2019un \u00e9v\u00e9nement manqu\u00e9 par rapport \u00e0 celui d\u2019une fausse alerte doit d\u00e9terminer quand l\u2019automatisation agit et quand une personne v\u00e9rifie.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"core\" data-step=\"8\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Attention<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Les syst\u00e8mes de vision ont du mal avec les objets inconnus, les conditions inhabituelles et les chevauchements. La d\u00e9rive appara\u00eet lorsque le monde change autour d\u2019un mod\u00e8le fig\u00e9.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Pourquoi c\u2019est important<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Dans les usages r\u00e9glement\u00e9s ou li\u00e9s \u00e0 la s\u00e9curit\u00e9, des r\u00e9sultats de test document\u00e9s sur diff\u00e9rentes conditions et diff\u00e9rents groupes d\u00e9mographiques transforment une d\u00e9monstration fonctionnelle en syst\u00e8me digne de confiance et auditable.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t<\/div>\n\t<\/div><\/div>\n<\/div>\n\n<div class=\"et_pb_column_20 et_pb_column et-last-child et_block_column et_pb_css_mix_blend_mode_passthrough et_flex_column_24_24 et_flex_column_24_24_tablet et_flex_column_24_24_phone et_flex_column_24_24_tabletWide\">\n<div class=\"et_pb_text_12 et_pb_text et_pb_bg_layout_light et_pb_module et_flex_module\"><div class=\"et_pb_text_inner\"><h3>The Challenges<\/h3>\n<p>Despite their capabilities often surpassing human vision, Computer Vision systems have limitations. They struggle with unknown information, unfamiliar conditions, or overlapping objects. Equally, while they can even detect the smallest changes, learning what is relevant and what isn\u2019t can be hard.<\/p>\n<p>Another concern is bias, for example, when it comes to facial recognition. Often, due to biased training data, their accuracy varies across different ethnic groups. And ultimately, they do not \u201crecognize\u201d items, but only statistical patterns. This can create dangerous errors, for example in facial recognition. As with all tools, careful oversight and governance are required.<\/p>\n<p>Additionally, the challenges emerging from image generation create entirely new ethical and regulatory problems.<\/p>\n<\/div><\/div>\n<\/div>\n<\/div>\n\n<div class=\"et_pb_row_13 et_pb_row et_flex_row\" style=\"--ns-caro-top:26px\">\n<div class=\"et_pb_column_21 et_pb_column et-last-child et_flex_column et_pb_css_mix_blend_mode_passthrough et_flex_column_24_24 et_flex_column_24_24_tablet et_flex_column_24_24_phone et_flex_column_24_24_tabletWide\">\n<div class=\"et_pb_code_7 et_pb_code et_pb_module\"><div class=\"et_pb_code_inner\"><div class=\"nsp-frame nsp-frame--transparent\"><div class=\"nsp-frame-head\"><h2>Featured Projects<\/h2><\/div><div class=\"nsp-carousel-wrap\" data-base-width=\"240\" data-loop=\"1\"><button type=\"button\" class=\"nsp-nav nsp-prev\" aria-label=\"Previous\"><span class=\"nsp-nav-icon et-pb-icon\" aria-hidden=\"true\">&#x34;<\/span><\/button><div class=\"nsp-carousel\">\t\t<article class=\"nsp-card\" tabindex=\"0\" role=\"button\" data-id=\"10\" aria-label=\"International Process Digitalization in Facility Management\">\n\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-card-iconwrap\"><span class=\"nsp-card-icon et-pb-icon\" aria-hidden=\"true\">&#xe035;<\/span><\/div>\n\t\t\t\t\t\t<h3 class=\"nsp-card-title\">International Process Digitalization in Facility Management<\/h3>\n\t\t\t\t\t\t\t\t\t\t<p class=\"nsp-card-excerpt\">Transformation and standardization of international facility management processes through the introduction of scalable structures.<\/p>\n\t\t\t\t\t\t\t\t\t\t<span class=\"nsp-card-more\">read more\u2026<\/span>\n\t\t\t\t\t<\/article>\n\t\t\t\t<article class=\"nsp-card\" tabindex=\"0\" role=\"button\" data-id=\"11\" aria-label=\"ERP\/CRM Product Development &amp; Process Digitalization\">\n\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-card-iconwrap\"><span class=\"nsp-card-icon et-pb-icon\" aria-hidden=\"true\">&#xe00d;<\/span><\/div>\n\t\t\t\t\t\t<h3 class=\"nsp-card-title\">ERP\/CRM Product Development &amp; Process Digitalization<\/h3>\n\t\t\t\t\t\t\t\t\t\t<p class=\"nsp-card-excerpt\">Development and implementation of a modular ERP\/CRM system to digitalize central business processes.<\/p>\n\t\t\t\t\t\t\t\t\t\t<span class=\"nsp-card-more\">read more\u2026<\/span>\n\t\t\t\t\t<\/article>\n\t\t\t\t<article class=\"nsp-card\" tabindex=\"0\" role=\"button\" data-id=\"13\" aria-label=\"Visual Search Recommendations\">\n\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-card-iconwrap\"><i class=\"nsp-card-icon fa-solid fa-socks\" aria-hidden=\"true\"><\/i><\/div>\n\t\t\t\t\t\t<h3 class=\"nsp-card-title\">Visual Search Recommendations<\/h3>\n\t\t\t\t\t\t\t\t\t\t<p class=\"nsp-card-excerpt\">eCommmerce plugin that enables searching for visually similar products, helping customers to find and compare multiple related items.<\/p>\n\t\t\t\t\t\t\t\t\t\t<span class=\"nsp-card-more\">read more\u2026<\/span>\n\t\t\t\t\t<\/article>\n\t\t\t\t<article class=\"nsp-card\" tabindex=\"0\" role=\"button\" data-id=\"14\" aria-label=\"Visual Assistance for Seniors\">\n\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-card-iconwrap\"><i class=\"nsp-card-icon fa-solid fa-glasses\" aria-hidden=\"true\"><\/i><\/div>\n\t\t\t\t\t\t<h3 class=\"nsp-card-title\">Visual Assistance for Seniors<\/h3>\n\t\t\t\t\t\t\t\t\t\t<p class=\"nsp-card-excerpt\">Visual Assistance App: enabling visual assistance for seniors by helping position determination using Computer Vision and Deep Learning.<\/p>\n\t\t\t\t\t\t\t\t\t\t<span class=\"nsp-card-more\">read more\u2026<\/span>\n\t\t\t\t\t<\/article>\n\t\t\t\t<article class=\"nsp-card\" tabindex=\"0\" role=\"button\" data-id=\"1\" aria-label=\"Electrical Switch Monitor\">\n\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-card-iconwrap\"><i class=\"nsp-card-icon fa-solid fa-toggle-on\" aria-hidden=\"true\"><\/i><\/div>\n\t\t\t\t\t\t<h3 class=\"nsp-card-title\">Electrical Switch Monitor<\/h3>\n\t\t\t\t\t\t\t\t\t\t<p class=\"nsp-card-excerpt\">Public transportation: using AI-driven vision to monitor old-fashioned electrical relays and also to evaluate potential failures for predictive maintenance.<\/p>\n\t\t\t\t\t\t\t\t\t\t<span class=\"nsp-card-more\">read more\u2026<\/span>\n\t\t\t\t\t<\/article>\n\t\t\t\t<article class=\"nsp-card\" tabindex=\"0\" role=\"button\" data-id=\"5\" aria-label=\"Hydropower Plant Operations\">\n\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-card-iconwrap\"><i class=\"nsp-card-icon fa-solid fa-water\" aria-hidden=\"true\"><\/i><\/div>\n\t\t\t\t\t\t<h3 class=\"nsp-card-title\">Hydropower Plant Operations<\/h3>\n\t\t\t\t\t\t\t\t\t\t<p class=\"nsp-card-excerpt\">Create a control and monitoring solution for all plant operations, including predictive maintenance logic and intrusion mon\u00aditoring.<\/p>\n\t\t\t\t\t\t\t\t\t\t<span class=\"nsp-card-more\">read more\u2026<\/span>\n\t\t\t\t\t<\/article>\n\t\t<\/div><button type=\"button\" class=\"nsp-nav nsp-next\" aria-label=\"Next\"><span class=\"nsp-nav-icon et-pb-icon\" aria-hidden=\"true\">&#x35;<\/span><\/button><\/div><script>(function(){\n\t\t\tif (!window.nspComputeLayout) {\n\t\t\t\twindow.NSP_BASE_CARD_WIDTH = 240;\n\t\t\t\twindow.nspIsFluidWidth = function(){ return window.innerWidth > 640; };\n\t\t\t\twindow.nspComputeLayout = function(available, gap, totalCards, baseWidth){\n\t\t\t\t\tvar BASE = baseWidth || window.NSP_BASE_CARD_WIDTH;\n\t\t\t\t\tavailable = Math.max(0, available || 0);\n\t\t\t\t\tgap = Math.max(0, gap || 0);\n\t\t\t\t\ttotalCards = Math.max(0, totalCards || 0);\n\t\t\t\t\tif (!totalCards){\n\t\t\t\t\t\treturn { count: 0, width: BASE, center: true };\n\t\t\t\t\t}\n\t\t\t\t\tvar per = BASE + gap;\n\t\t\t\t\tvar baseCount = Math.max(1, Math.floor((available + gap) \/ per));\n\t\t\t\t\tvar used = baseCount * BASE + (baseCount - 1) * gap;\n\t\t\t\t\tvar remainder = available - used;\n\t\t\t\t\t\/\/ Only create an extra visible slot when there is actually another card to fill it.\n\t\t\t\t\t\/\/ If exactly the base-width row count is present, distribute those cards evenly\n\t\t\t\t\t\/\/ across the row instead of leaving them beside a phantom extra slot.\n\t\t\t\t\tvar fullCount = (totalCards > baseCount && remainder >= (2 \/ 3) * BASE) ? baseCount + 1 : baseCount;\n\t\t\t\t\tfullCount = Math.max(1, fullCount);\n\t\t\t\t\tvar fullWidth = (available - (fullCount - 1) * gap) \/ fullCount;\n\t\t\t\t\tif (!isFinite(fullWidth) || fullWidth <= 0){ fullWidth = BASE; }\n\n\t\t\t\t\t\/\/ One or two cards should never grow just because the row is wide.\n\t\t\t\t\t\/\/ They keep the base width, sit at the left, and leave the rest of the row empty.\n\t\t\t\t\tif (totalCards < 3){\n\t\t\t\t\t\tvar smallMax = (available - (totalCards - 1) * gap) \/ totalCards;\n\t\t\t\t\t\tvar smallWidth = (isFinite(smallMax) && smallMax > 0) ? Math.min(BASE, smallMax) : BASE;\n\t\t\t\t\t\treturn { count: totalCards, width: smallWidth, center: false };\n\t\t\t\t\t}\n\n\t\t\t\t\t\/\/ If there are fewer cards than a full row would hold, keep the full-row\n\t\t\t\t\t\/\/ card width rather than stretching those few across the whole row, and\n\t\t\t\t\t\/\/ leave the remaining slots empty. Left-aligned, as the master grid is.\n\t\t\t\t\tif (totalCards < fullCount){\n\t\t\t\t\t\treturn { count: totalCards, width: fullWidth, center: false };\n\t\t\t\t\t}\n\n\t\t\t\t\tvar count = Math.min(fullCount, totalCards);\n\t\t\t\t\treturn { count: count, width: fullWidth, center: false };\n\t\t\t\t};\n\t\t\t\t\/\/ Centre the chevrons on the media block of a card rather than on\n\t\t\t\t\/\/ the whole card: on people cards that is the square photo\n\t\t\t\t\/\/ (whose height tracks the fluid card width, so it cannot be\n\t\t\t\t\/\/ expressed in CSS), on project cards the icon block. Published\n\t\t\t\t\/\/ as a custom property the stylesheet reads.\n\t\t\t\twindow.nspTrackPadX = function(track){\n\t\t\t\t\tvar cs = window.getComputedStyle(track);\n\t\t\t\t\treturn (parseFloat(cs.paddingLeft) || 0) + (parseFloat(cs.paddingRight) || 0);\n\t\t\t\t};\n\t\t\t\twindow.nspSyncNav = function(wrap){\n\t\t\t\t\t\/\/ People cards anchor on the photo, which is the visual mass of\n\t\t\t\t\t\/\/ the card. Project cards have no photo -- only a small icon\n\t\t\t\t\t\/\/ at the top -- so they anchor on the card panel itself and\n\t\t\t\t\t\/\/ the arrows land on its vertical centre. Measuring against\n\t\t\t\t\t\/\/ the wrap in viewport coordinates absorbs the track padding\n\t\t\t\t\t\/\/ and card margins without restating any of them here.\n\t\t\t\t\tvar media = wrap.querySelector(\".nsp-card-photo\")\n\t\t\t\t\t\t|| wrap.querySelector(\".nsp-card\");\n\t\t\t\t\tif (!media){ return; }\n\t\t\t\t\tvar r = media.getBoundingClientRect();\n\t\t\t\t\tif (!r.height){ return; }\n\t\t\t\t\tvar w = wrap.getBoundingClientRect();\n\t\t\t\t\twrap.style.setProperty(\"--nsp-nav-top\", ((r.top - w.top) + r.height \/ 2) + \"px\");\n\t\t\t\t};\n\t\t\t}\n\t\t\tvar wrap = document.currentScript.previousElementSibling;\n\t\t\tif(!wrap){ return; }\n\t\t\t\/\/ The frame carries the visible panel, so it has to be revealed\n\t\t\t\/\/ together with the wrap -- otherwise an empty framed box paints\n\t\t\t\/\/ before the cards have been sized into it.\n\t\t\tfunction reveal(w){\n\t\t\t\tw.style.visibility = \"visible\";\n\t\t\t\tvar f = w.parentNode;\n\t\t\t\tif (f && f.classList && f.classList.contains(\"nsp-frame\")){ f.style.visibility = \"visible\"; }\n\t\t\t}\n\t\t\tvar baseWidth = parseInt(wrap.getAttribute(\"data-base-width\"), 10) || window.NSP_BASE_CARD_WIDTH;\n\t\t\tvar track = wrap.querySelector(\".nsp-carousel\");\n\t\t\tvar cards = track ? track.querySelectorAll(\".nsp-card\") : [];\n\t\t\tif(!track || !cards.length){ if (window.nspSyncNav){ window.nspSyncNav(wrap); }\n\t\t\treveal(wrap); return; }\n\t\t\t\/\/ Read from the stylesheet rather than restating it: a hard-coded copy\n\t\t\t\/\/ here is exactly how the gutter drifted from the CSS before.\n\t\t\tvar gap = parseFloat(window.getComputedStyle(track).columnGap) || 24;\n\t\t\tif (!window.nspIsFluidWidth()) {\n\t\t\t\t\/\/ Mobile: unchanged -- fixed per-card width (CSS 82vw), just avoid a partial\n\t\t\t\t\/\/ card peeking out at the wrap edge.\n\t\t\t\tvar cardW = cards[0].getBoundingClientRect().width;\n\t\t\t\tif(!cardW){ if (window.nspSyncNav){ window.nspSyncNav(wrap); }\n\t\t\treveal(wrap); return; }\n\t\t\t\t\/\/ clientWidth is the PADDING box, so the track padding has to come\n\t\t\t\t\/\/ off before this is the room the cards actually get.\n\t\t\t\tvar available0 = track.clientWidth - window.nspTrackPadX(track);\n\t\t\t\tvar fitCount = Math.max(1, Math.floor((available0 + gap) \/ (cardW + gap)));\n\t\t\t\tvar count0 = Math.min(fitCount, cards.length);\n\t\t\t\tvar needed0 = count0 * cardW + (count0 - 1) * gap;\n\t\t\t\tvar ws0 = window.getComputedStyle(wrap);\n\t\t\t\tvar padX0 = (parseFloat(ws0.paddingLeft) || 0) + (parseFloat(ws0.paddingRight) || 0);\n\t\t\t\tvar target0 = needed0 + window.nspTrackPadX(track) + (ws0.boxSizing === \"border-box\" ? padX0 : 0);\n\t\t\t\twrap.style.maxWidth = Math.ceil(target0) + \"px\";\n\t\t\t\tif (window.nspSyncNav){ window.nspSyncNav(wrap); }\n\t\t\treveal(wrap);\n\t\t\t\treturn;\n\t\t\t}\n\t\t\t\/\/ Desktop\/tablet: fluid card width, filling the row exactly either way.\n\t\t\t\/\/ clientWidth is the PADDING box -- take the track padding off it.\n\t\t\tvar available = track.clientWidth - window.nspTrackPadX(track);\n\t\t\tvar layout = window.nspComputeLayout(available, gap, cards.length, baseWidth);\n\t\t\tfor (var i = 0; i < cards.length; i++){ cards[i].style.flex = \"0 0 \" + layout.width + \"px\"; }\n\t\t\tif (layout.center) {\n\t\t\t\tvar totalW = layout.count * layout.width + (layout.count - 1) * gap;\n\t\t\t\twrap.style.maxWidth = Math.ceil(totalW + window.nspTrackPadX(track)) + \"px\";\n\t\t\t\twrap.style.marginLeft = \"auto\";\n\t\t\t\twrap.style.marginRight = \"auto\";\n\t\t\t}\n\t\t\tif (window.nspSyncNav){ window.nspSyncNav(wrap); }\n\t\t\treveal(wrap);\n\t\t})();<\/script><\/div><noscript><style>.nsp-carousel-wrap,.nsp-frame{visibility:visible !important;}<\/style><\/noscript><div class=\"nsp-popups\" hidden><template class=\"nsp-popup-tpl\" data-id=\"10\">\t\t<div class=\"nsp-dialog no-image\">\n\t\t\t<button type=\"button\" class=\"nsp-close\" aria-label=\"Close\">&times;<\/button>\n\t\t\t<div class=\"nsp-dlg-h\">\n\t\t\t\t<span class=\"nsp-pop-icon et-pb-icon\" aria-hidden=\"true\">&#xe035;<\/span>\t\t\t\t<span class=\"nsp-dlg-h-title\">International Process Digitalization in Facility Management<\/span>\n\t\t\t<\/div>\n\t\t\t<div class=\"nsp-pop-cols\">\n\t\t\t\t<div class=\"nsp-pop-main\">\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-desc\"><p>Supported the transformation and standardization of international facility management processes by designing scalable end-to-end process and service structures. <\/p>\n<p>The project included gathering and harmonizing requirements across multiple country organizations, translating business needs into digital process and system solutions, and coordinating international rollouts including SIT, UAT, training, and change management. <\/p>\n<p>Consistent process modeling and documentation using BPMN 2.0 and SAP Signavio helped establish sustainable governance, transparency, and operational efficiency.<\/p>\n<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-role nsp-pop-role--below\"><p>SAP Signavio, BPMN 2.0, ERP Systems, Digital Workflow Platforms, Interface Integration, SIT\/UAT, Requirements Management<\/p>\n<\/div>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\n\t\t\t\t\t\t\t<ul class=\"nsp-chiplist nsp-pop-caps\">\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/vision\">Computer Vision<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/control-and-automation\">Automation<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/automation-and-control\">Infrastructure Management<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\tHuman-Technology Interaction\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t<\/ul>\n\t\t\t\t\t<\/div>\n\t\t<\/template><template class=\"nsp-popup-tpl\" data-id=\"11\">\t\t<div class=\"nsp-dialog no-image\">\n\t\t\t<button type=\"button\" class=\"nsp-close\" aria-label=\"Close\">&times;<\/button>\n\t\t\t<div class=\"nsp-dlg-h\">\n\t\t\t\t<span class=\"nsp-pop-icon et-pb-icon\" aria-hidden=\"true\">&#xe00d;<\/span>\t\t\t\t<span class=\"nsp-dlg-h-title\">ERP\/CRM Product Development &amp; Process Digitalization<\/span>\n\t\t\t<\/div>\n\t\t\t<div class=\"nsp-pop-cols\">\n\t\t\t\t<div class=\"nsp-pop-main\">\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-desc\"><p>Led the development and implementation of a modular ERP\/CRM system to digitalize central business processes.<\/p>\n<p>The project included end-to-end product ownership, requirements analysis, prioritization, and scaling of the system, including mobile solutions and extensions.<\/p>\n<p>Responsibilities also covered the management of cross-functional development teams, the establishment of testing, quality, and operations processes, and the introduction of ITIL-based change and incident structures. Governance, KPI, PMO, and documentation standards were developed to support sustainable product and project management, while product strategy and stakeholder alignment were managed at leadership level.<\/p>\n<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-role nsp-pop-role--below\"><p>ERP\/CRM Systems, Mobile Solutions, Agile Product Development, Requirements Management, UAT, ITIL, Change &amp; Incident Management, KPI\/PMO Structures, Stakeholder Management<\/p>\n<\/div>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\n\t\t\t\t\t\t\t<ul class=\"nsp-chiplist nsp-pop-caps\">\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/vision\">Computer Vision<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/control-and-automation\">Automation<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/automation-and-control\">Infrastructure Management<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/data-and-knowledge-management\/\">Data Warehouse<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\tHuman-Technology Interaction\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t<\/ul>\n\t\t\t\t\t<\/div>\n\t\t<\/template><template class=\"nsp-popup-tpl\" data-id=\"13\">\t\t<div class=\"nsp-dialog has-image\">\n\t\t\t<button type=\"button\" class=\"nsp-close\" aria-label=\"Close\">&times;<\/button>\n\t\t\t<div class=\"nsp-dlg-h\">\n\t\t\t\t\t\t\t\t<span class=\"nsp-dlg-h-title\">Visual Search Recommendations<\/span>\n\t\t\t<\/div>\n\t\t\t<div class=\"nsp-pop-cols\">\n\t\t\t\t<div class=\"nsp-pop-main\">\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-desc\"><p>For VisualSearch, the focus was on developing and deploying e-commerce plugins to enhance recommendation functionalities in web shops. The plugins were successfully launched in the store and happily adopted by customers, demonstrating their practical use in improving e-commerce experiences. Key elements included:<br \/>\n- computation of visual embeddings from appearances of e-commerce products<br \/>\n- building and maintaining a search index using these embeddings<br \/>\n- providing a cloud-based API for Shopware and Prestashop plugins<\/p>\n<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-aside\">\n\t\t\t\t\t\t<img loading=\"lazy\" decoding=\"async\" width=\"369\" height=\"455\" src=\"https:\/\/www.9senses.ai\/wp-content\/uploads\/2026\/07\/visualsearch.png\" class=\"nsp-pop-img\" alt=\"\" srcset=\"https:\/\/www.9senses.ai\/wp-content\/uploads\/2026\/07\/visualsearch.png 369w, https:\/\/www.9senses.ai\/wp-content\/uploads\/2026\/07\/visualsearch-243x300.png 243w, https:\/\/www.9senses.ai\/wp-content\/uploads\/2026\/07\/visualsearch-10x12.png 10w\" sizes=\"(max-width: 369px) 100vw, 369px\" \/>\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-role nsp-pop-role--aside\"><p>Peter acted as the project manager and key data scientist on this project, leveraging AWS services, Python, and deep learning frameworks like Keras. The project integrated cloud-based solutions using CloudFormation, Lambda, and Gateway for scalable and efficient deployment, using SQL and DynamoDB for data management.<\/p>\n<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\n\t\t\t\t\t\t\t<ul class=\"nsp-chiplist nsp-pop-caps\">\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/machine-learning\">Machine Learning<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/vision\">Computer Vision<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t<\/ul>\n\t\t\t\t\t<\/div>\n\t\t<\/template><template class=\"nsp-popup-tpl\" data-id=\"14\">\t\t<div class=\"nsp-dialog has-image\">\n\t\t\t<button type=\"button\" class=\"nsp-close\" aria-label=\"Close\">&times;<\/button>\n\t\t\t<div class=\"nsp-dlg-h\">\n\t\t\t\t\t\t\t\t<span class=\"nsp-dlg-h-title\">Visual Assistance for Seniors<\/span>\n\t\t\t<\/div>\n\t\t\t<div class=\"nsp-pop-cols\">\n\t\t\t\t<div class=\"nsp-pop-main\">\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-desc\"><p>For the Austrian Institute of Technology (as a part of the national research project LARAH), a prototype of an assistance system for visual indoor localization for disabled and elderly persons was developed. By leveraging Computer Vision and Deep Learning, innovative algorithms for position determination were implemented. The project resulted in a functional prototype, including two Android applications for real-time localization. Key elements included:<br \/>\n- visual recognition of persons using Deep-Learning models<br \/>\n- visual reconstruction and localization of indoor environments using Structure-from-Motion and Machine Learning algorithms<\/p>\n<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-aside\">\n\t\t\t\t\t\t<img loading=\"lazy\" decoding=\"async\" width=\"366\" height=\"451\" src=\"https:\/\/www.9senses.ai\/wp-content\/uploads\/2026\/07\/larah.png\" class=\"nsp-pop-img\" alt=\"\" srcset=\"https:\/\/www.9senses.ai\/wp-content\/uploads\/2026\/07\/larah.png 366w, https:\/\/www.9senses.ai\/wp-content\/uploads\/2026\/07\/larah-243x300.png 243w, https:\/\/www.9senses.ai\/wp-content\/uploads\/2026\/07\/larah-10x12.png 10w\" sizes=\"(max-width: 366px) 100vw, 366px\" \/>\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-role nsp-pop-role--aside\"><p>Peter acted as the project manager and key data scientist on this project, developed under Python using Deep Learning models like TensorFlow, custom-developed Structure-from-Motion software and custom camera calibration software. Additionally, as a part of the project, two Android apps were modified and integrated together onto the Robot Operating System on the mobile platform.<\/p>\n<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\n\t\t\t\t\t\t\t<ul class=\"nsp-chiplist nsp-pop-caps\">\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/machine-learning\">Machine Learning<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/vision\">Computer Vision<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t<\/ul>\n\t\t\t\t\t<\/div>\n\t\t<\/template><template class=\"nsp-popup-tpl\" data-id=\"1\">\t\t<div class=\"nsp-dialog has-image\">\n\t\t\t<button type=\"button\" class=\"nsp-close\" aria-label=\"Close\">&times;<\/button>\n\t\t\t<div class=\"nsp-dlg-h\">\n\t\t\t\t\t\t\t\t<span class=\"nsp-dlg-h-title\">Electrical Switch Monitor<\/span>\n\t\t\t<\/div>\n\t\t\t<div class=\"nsp-pop-cols\">\n\t\t\t\t<div class=\"nsp-pop-main\">\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-desc\"><p>For a public transportation network, the objective was to optimize the monitoring of their legacy electrical switchboards. These decade-old items that are often located in very remote areas are prone to failures and tracking of errors was not possible. The objective was to enable real-time tracking and the recognition of upcoming failures from changed switching behavior. The key elements were:<\/p>\n<ul>\n<li>Development of specific hardware configuration with custom housings (3D printed) to mount instead of regular switchboard covers;<\/li>\n<li>Camera control and initial image generation on Raspberry Pi integrated in housing;<\/li>\n<li>Initial scan of switch layout and labels;<\/li>\n<li>Identification of switching operations and registration of new positions;<\/li>\n<li>Identification of irregular switching patterns (delays, other irregularities) to indicate upcoming failures for predictive maintenance;<\/li>\n<li>Update of central database and cloud solution with last state and observed switching patterns;<\/li>\n<\/ul>\n<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-aside\">\n\t\t\t\t\t\t<img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"959\" src=\"https:\/\/www.9senses.ai\/wp-content\/uploads\/2026\/07\/6by6cropped-1024x959.jpg\" class=\"nsp-pop-img\" alt=\"\" srcset=\"https:\/\/www.9senses.ai\/wp-content\/uploads\/2026\/07\/6by6cropped-980x918.jpg 980w, https:\/\/www.9senses.ai\/wp-content\/uploads\/2026\/07\/6by6cropped-480x450.jpg 480w\" sizes=\"(min-width: 0px) and (max-width: 480px) 480px, (min-width: 481px) and (max-width: 980px) 980px, (min-width: 981px) 1024px, 100vw\" \/>\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-role nsp-pop-role--aside\"><p>The solution was implemented using Python on Raspberry Pi devices, backbone and cloud processing were done using a LAMP stack, with PyTorch, TensorFlow and OpenCV.<\/p>\n<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\n\t\t\t\t\t\t\t<ul class=\"nsp-chiplist nsp-pop-caps\">\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/machine-learning\">Machine Learning<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/vision\">Computer Vision<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t<\/ul>\n\t\t\t\t\t<\/div>\n\t\t<\/template><template class=\"nsp-popup-tpl\" data-id=\"5\">\t\t<div class=\"nsp-dialog no-image\">\n\t\t\t<button type=\"button\" class=\"nsp-close\" aria-label=\"Close\">&times;<\/button>\n\t\t\t<div class=\"nsp-dlg-h\">\n\t\t\t\t<i class=\"nsp-pop-icon fa-solid fa-water\" aria-hidden=\"true\"><\/i>\t\t\t\t<span class=\"nsp-dlg-h-title\">Hydropower Plant Operations<\/span>\n\t\t\t<\/div>\n\t\t\t<div class=\"nsp-pop-cols\">\n\t\t\t\t<div class=\"nsp-pop-main\">\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-desc\"><div class=\"et_pb_column_11 et_pb_column et_flex_column et_pb_css_mix_blend_mode_passthrough et_flex_column_6_24 et_flex_column_12_24_tablet et_flex_column_24_24_phone et_flex_column_12_24_tabletWide preset--group--divi-column--divi-box-shadow--default preset--group--divi-column--divi-sizing--hsj9uxo--default\">\n<div class=\"et_pb_blurb_2 et_pb_blurb et_pb_bg_layout_light et_pb_blurb_position_top et_pb_module et_flex_module hovergroup preset--group--divi-blurb--divi-box-shadow--default preset--group--divi-blurb--divi-font-body--h19rs5u--7p5s44libg preset--group--divi-blurb--divi-sizing--hsj9uxo--default\">\n<div class=\"et_pb_blurb_content et_flex_module\">\n<div class=\"et_pb_blurb_container\">\n<div class=\"et_pb_blurb_description\">\n<p>Create an integrated monitoring and surveillance solution for small-scale hydropower plants in remote locations. The solution included a full range of required settings:<\/p>\n<ul>\n<li>real-time monitoring and logging of operations<\/li>\n<li>failure detection and automated<\/li>\n<li>predictive maintenance logic to identify early failure<\/li>\n<li>camera-based intrusion and irregularity detection<\/li>\n<\/ul>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"nsp-pop-role nsp-pop-role--below\"><p>The solution was a hybrid solution using Siemens PLCs, combined with small edge computing elements (Raspberry Pi and Arduino). All primary logic (particularly shutdown and load adjustment) was local, but key decisions and aggregations were executed online based on regular data transmission to a cloud-based management and operations suite.<\/p>\n<\/div>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\n\t\t\t\t\t\t\t<ul class=\"nsp-chiplist nsp-pop-caps\">\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/machine-learning\">Machine Learning<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/vision\">Computer Vision<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t<li class=\"nsp-chip\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<a href=\"\/automation-and-control\">Infrastructure Management<\/a>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t<\/ul>\n\t\t\t\t\t<\/div>\n\t\t<\/template><\/div><div class=\"nsp-overlay\" id=\"nsp-overlay\" role=\"dialog\" aria-modal=\"true\"><\/div><\/div><\/div>\n<\/div>\n<\/div>\n<\/div>\n\n<div class=\"et_pb_section_8 et_pb_section et_section_regular et_flex_section ns-panel\" id=\"robotics\" style=\"--ns-rail-w:264px; --ns-body-color:#D3D3D3;\">\n<div class=\"et_pb_row_14 et_pb_row et_flex_row\">\n<div class=\"et_pb_column_22 et_pb_column et_block_column et_pb_css_mix_blend_mode_passthrough et_flex_column_24_24 et_flex_column_24_24_tabletWide\">\n<div class=\"et_pb_text_13 et_pb_text et_pb_bg_layout_light et_pb_module et_flex_module\"><div class=\"et_pb_text_inner\"><p><span>Of all AI fields, robotics is the one where software has consequences in the physical world: sensor data has to be turned into safe, precise motion in real time, and there is no undo button. That step from calculating to acting is what makes the field so demanding, and why progress here is slower than in purely digital AI.<\/span><\/p>\n<\/div><\/div>\n\n<div class=\"et_pb_code_8 et_pb_code et_pb_module\"><div class=\"et_pb_code_inner\">\t<div\n\t\tid=\"ns-aix-4\"\n\t\tclass=\"ns-aix ns-aix-rob ns-aix-rob3d\"\n\t\tdata-topic=\"rob\"\n\t\tdata-flow=\"industrial\"\n\t\t\t\t\t\tdata-journey=\"industrial\"\n\t\tdata-active-key=\"task\"\n\t\t\t\tdata-step=\"0\"\n\t\tdata-interval=\"9000\"\n\t\t\t\tdata-autoplay=\"1\"\n\t\trole=\"region\"\n\t\taria-label=\"Comment fonctionne la robotique\"\n\t\tstyle=\"--ns-aix-accent:#58a7f9;--ns-aix-secondary:#2b6cb0\"\n\t>\n\t\t<header class=\"ns-aix-head\">\n\t\t\t\t\t\t<h3 class=\"ns-aix-title\">Comment fonctionne la robotique<\/h3>\n\t\t\t<p class=\"ns-aix-intro\">En robotique, l\u2019IA et l\u2019automatisation deviennent physiques. Les capteurs mesurent le monde, le logiciel estime ce qui se passe, les contr\u00f4leurs choisissent des mouvements s\u00fbrs et les actionneurs d\u00e9placent des objets r\u00e9els. La boucle ne fonctionne que si s\u00e9curit\u00e9, v\u00e9rification et gouvernance sont pr\u00e9vues d\u00e8s la conception.<\/p>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-flowgroup ns-aix-flowgroup-shaped\">\n\t\t\t\t\t\t\t\t\t\t\t<svg class=\"ns-aix-flowshape\" aria-hidden=\"true\" focusable=\"false\" preserveAspectRatio=\"none\">\n\t\t\t\t\t\t\t<path class=\"ns-aix-flowshape-fill\" d=\"\" \/>\n\t\t\t\t\t\t\t<path class=\"ns-aix-flowshape-glow\" d=\"\" \/>\n\t\t\t\t\t\t<\/svg>\n\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-tabs\" role=\"tablist\" aria-label=\"Comment fonctionne la robotique\">\n\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\tid=\"ns-aix-4-tab-industrial\"\n\t\t\t\t\t\t\tclass=\"ns-aix-tab ns-on\"\n\t\t\t\t\t\t\tdata-flow=\"industrial\"\n\t\t\t\t\t\t\trole=\"tab\"\n\t\t\t\t\t\t\taria-selected=\"true\"\n\t\t\t\t\t\t\taria-label=\"Afficher le parcours de la cellule robotis\u00e9e\"\n\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\taria-controls=\"ns-aix-4-flow-panel-industrial\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\tRobot industriel\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\tid=\"ns-aix-4-tab-autonomous\"\n\t\t\t\t\t\t\tclass=\"ns-aix-tab\"\n\t\t\t\t\t\t\tdata-flow=\"autonomous\"\n\t\t\t\t\t\t\trole=\"tab\"\n\t\t\t\t\t\t\taria-selected=\"false\"\n\t\t\t\t\t\t\taria-label=\"Afficher le parcours du robot autonome\"\n\t\t\t\t\t\t\ttabindex=\"-1\"\n\t\t\t\t\t\t\taria-controls=\"ns-aix-4-flow-panel-autonomous\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\tRobot autonome\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t<\/div>\n\n\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-flow-intros\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t<p\n\t\t\t\t\t\t\t\tid=\"ns-aix-4-flow-intro-industrial\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-flow-intro ns-on\"\n\t\t\t\t\t\t\t\tdata-flow=\"industrial\"\n\t\t\t\t\t\t\t\taria-hidden=\"false\"\n\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\tUn robot industriel peut \u00eatre tr\u00e8s performant parce que sa cellule de travail est volontairement contrainte. Montages, coordonn\u00e9es, outils, limites de vitesse et zones de s\u00e9curit\u00e9 r\u00e9duisent l\u2019incertitude avant que le contr\u00f4leur ne mette le robot en mouvement.\t\t\t\t\t\t\t<\/p>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<p\n\t\t\t\t\t\t\t\tid=\"ns-aix-4-flow-intro-autonomous\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-flow-intro\"\n\t\t\t\t\t\t\t\tdata-flow=\"autonomous\"\n\t\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\tL\u2019autonomie devient plus difficile lorsque ces limites disparaissent. Un v\u00e9hicule, un rover ou un robot marcheur doit estimer sa propre position, interpr\u00e9ter une sc\u00e8ne qui change et agir en s\u00e9curit\u00e9 malgr\u00e9 l\u2019incertitude.\t\t\t\t\t\t\t<\/p>\n\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t<\/header>\n\n\t\t<div class=\"ns-aix-body\">\n\t\t\t<div class=\"ns-aix-copy\">\n\t\t\t\t<div class=\"ns-aix-progress\" aria-hidden=\"true\"><span><\/span><\/div>\n\t\t\t\t\t\t\t\t\t<div\n\t\t\t\t\t\tid=\"ns-aix-4-flow-panel-industrial\"\n\t\t\t\t\t\tclass=\"ns-aix-step-panel ns-on\"\n\t\t\t\t\t\tdata-flow=\"industrial\"\n\t\t\t\t\t\taria-hidden=\"false\"\n\t\t\t\t\t\trole=\"tabpanel\" aria-labelledby=\"ns-aix-4-tab-industrial\"\t\t\t\t\t>\n\t\t\t\t\t<ol class=\"ns-aix-steps ns-on\" data-flow=\"industrial\" role=\"list\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step ns-on\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"industrial\"\n\t\t\t\t\t\t\t\t\tdata-step=\"0\"\n\t\t\t\t\t\t\t\t\tdata-key=\"task\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"true\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-4-panel-industrial-0\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">01<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">T\u00e2che<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Une t\u00e2che clairement d\u00e9limit\u00e9e rend l\u2019automatisation possible.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"industrial\"\n\t\t\t\t\t\t\t\t\tdata-step=\"1\"\n\t\t\t\t\t\t\t\t\tdata-key=\"sense\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-4-panel-industrial-1\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">02<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Percevoir<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Cam\u00e9ras, encodeurs et signaux de force entrent dans la boucle.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"industrial\"\n\t\t\t\t\t\t\t\t\tdata-step=\"2\"\n\t\t\t\t\t\t\t\t\tdata-key=\"locate\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-4-panel-industrial-2\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">03<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Localiser<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">La position n\u2019a de sens que dans un rep\u00e8re commun.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"industrial\"\n\t\t\t\t\t\t\t\t\tdata-step=\"3\"\n\t\t\t\t\t\t\t\t\tdata-key=\"plan\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-4-panel-industrial-3\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">04<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Planifier<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Une trajectoire est choisie avant le d\u00e9but du mouvement.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"industrial\"\n\t\t\t\t\t\t\t\t\tdata-step=\"4\"\n\t\t\t\t\t\t\t\t\tdata-key=\"control\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-4-panel-industrial-4\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">05<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Commander<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Les actionneurs transforment les commandes en mouvement.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"industrial\"\n\t\t\t\t\t\t\t\t\tdata-step=\"5\"\n\t\t\t\t\t\t\t\t\tdata-key=\"verify\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-4-panel-industrial-5\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">06<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">V\u00e9rifier<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">La r\u00e9ussite se mesure, elle ne se suppose pas.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"industrial\"\n\t\t\t\t\t\t\t\t\tdata-step=\"6\"\n\t\t\t\t\t\t\t\t\tdata-key=\"protect\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-4-panel-industrial-6\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">07<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Prot\u00e9ger<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">L\u2019action physique exige des limites explicites.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"industrial\"\n\t\t\t\t\t\t\t\t\tdata-step=\"7\"\n\t\t\t\t\t\t\t\t\tdata-key=\"improve\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-4-panel-industrial-7\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">08<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Am\u00e9liorer<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Les logs soutiennent la maintenance et l\u2019optimisation.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t<\/ol>\n\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t<div\n\t\t\t\t\t\tid=\"ns-aix-4-flow-panel-autonomous\"\n\t\t\t\t\t\tclass=\"ns-aix-step-panel\"\n\t\t\t\t\t\tdata-flow=\"autonomous\"\n\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\trole=\"tabpanel\" aria-labelledby=\"ns-aix-4-tab-autonomous\"\t\t\t\t\t>\n\t\t\t\t\t<ol class=\"ns-aix-steps\" data-flow=\"autonomous\" role=\"list\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"autonomous\"\n\t\t\t\t\t\t\t\t\tdata-step=\"0\"\n\t\t\t\t\t\t\t\t\tdata-key=\"mission\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-4-panel-autonomous-0\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">01<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Mission<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Une mission comprend des r\u00e8gles, pas seulement une destination.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"autonomous\"\n\t\t\t\t\t\t\t\t\tdata-step=\"1\"\n\t\t\t\t\t\t\t\t\tdata-key=\"perceive\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-4-panel-autonomous-1\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">02<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Percevoir<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Plusieurs capteurs r\u00e9duisent les angles morts.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"autonomous\"\n\t\t\t\t\t\t\t\t\tdata-step=\"2\"\n\t\t\t\t\t\t\t\t\tdata-key=\"localize\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-4-panel-autonomous-2\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">03<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Localiser<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">La carte n\u2019est utile qu\u2019avec une estimation en temps r\u00e9el de la pose.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"autonomous\"\n\t\t\t\t\t\t\t\t\tdata-step=\"3\"\n\t\t\t\t\t\t\t\t\tdata-key=\"model\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-4-panel-autonomous-3\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">04<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Mod\u00e9liser<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Objets, espace et risque sont estim\u00e9s.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"autonomous\"\n\t\t\t\t\t\t\t\t\tdata-step=\"4\"\n\t\t\t\t\t\t\t\t\tdata-key=\"plan\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-4-panel-autonomous-4\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">05<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Planifier<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Plusieurs trajectoires possibles sont compar\u00e9es sous contraintes.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"autonomous\"\n\t\t\t\t\t\t\t\t\tdata-step=\"5\"\n\t\t\t\t\t\t\t\t\tdata-key=\"act\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-4-panel-autonomous-5\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">06<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Agir<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Le contr\u00f4le du mouvement s\u2019adapte en continu.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"autonomous\"\n\t\t\t\t\t\t\t\t\tdata-step=\"6\"\n\t\t\t\t\t\t\t\t\tdata-key=\"respond\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-4-panel-autonomous-6\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">07<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">R\u00e9agir<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Le mode de repli fait partie de l\u2019autonomie.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<li>\n\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\t\tclass=\"ns-aix-step\"\n\t\t\t\t\t\t\t\t\tdata-flow=\"autonomous\"\n\t\t\t\t\t\t\t\t\tdata-step=\"7\"\n\t\t\t\t\t\t\t\t\tdata-key=\"govern\"\n\t\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\t\taria-controls=\"ns-aix-4-panel-autonomous-7\"\n\t\t\t\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-num\">08<\/span>\n\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-text\">\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-label\">Piloter<\/span>\n\t\t\t\t\t\t\t\t\t\t<span class=\"ns-aix-step-micro\">Logs, limites et intervention humaine comptent.<\/span>\n\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t<\/li>\n\t\t\t\t\t\t\t\t\t\t\t<\/ol>\n\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\n\t\t\t<div class=\"ns-aix-stage\" aria-label=\"Animated vector-map process diagram\">\n\t\t\t\t<canvas class=\"ns-aix-canvas\" aria-hidden=\"true\"><\/canvas>\n\t\t\t\t\t\t\t\t\t<canvas class=\"ns-aix-canvas3d\" aria-hidden=\"true\"><\/canvas>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node ns-on\" data-flow=\"industrial\" data-step=\"0\" data-key=\"task\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>t\u00e2che<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node\" data-flow=\"industrial\" data-step=\"1\" data-key=\"sense\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>capter<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node\" data-flow=\"industrial\" data-step=\"2\" data-key=\"locate\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>rep\u00e8re<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node\" data-flow=\"industrial\" data-step=\"3\" data-key=\"plan\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>plan<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node\" data-flow=\"industrial\" data-step=\"4\" data-key=\"control\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>agir<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node\" data-flow=\"industrial\" data-step=\"5\" data-key=\"verify\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>v\u00e9rifier<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node\" data-flow=\"industrial\" data-step=\"6\" data-key=\"protect\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>s\u00e9curit\u00e9<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node\" data-flow=\"industrial\" data-step=\"7\" data-key=\"improve\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>logs<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node ns-hidden\" data-flow=\"autonomous\" data-step=\"0\" data-key=\"mission\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>mission<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node ns-hidden\" data-flow=\"autonomous\" data-step=\"1\" data-key=\"perceive\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>capter<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node ns-hidden\" data-flow=\"autonomous\" data-step=\"2\" data-key=\"localize\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>pose<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node ns-hidden\" data-flow=\"autonomous\" data-step=\"3\" data-key=\"model\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>sc\u00e8ne<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node ns-hidden\" data-flow=\"autonomous\" data-step=\"4\" data-key=\"plan\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>plan<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node ns-hidden\" data-flow=\"autonomous\" data-step=\"5\" data-key=\"act\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>agir<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node ns-hidden\" data-flow=\"autonomous\" data-step=\"6\" data-key=\"respond\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>mode de repli<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t<button type=\"button\" class=\"ns-aix-node ns-hidden\" data-flow=\"autonomous\" data-step=\"7\" data-key=\"govern\" tabindex=\"-1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t<span>piloter<\/span>\n\t\t\t\t\t\t<\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\n\t\t\t\t\t<\/div>\n\n\t\t<div class=\"ns-aix-controls\">\n\t\t\t<div class=\"ns-aix-dots\" role=\"navigation\" aria-label=\"Step navigation\">\n\t\t\t\t\t\t\t\t\t<span\n\t\t\t\t\t\tclass=\"ns-aix-dotset ns-on\"\n\t\t\t\t\t\tdata-flow=\"industrial\"\n\t\t\t\t\t\trole=\"group\"\n\t\t\t\t\t\taria-label=\"Steps for Robot industriel\"\n\t\t\t\t\t\taria-hidden=\"false\"\n\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot ns-on\"\n\t\t\t\t\t\t\t\tdata-flow=\"industrial\"\n\t\t\t\t\t\t\t\tdata-step=\"0\"\n\t\t\t\t\t\t\t\taria-label=\"Step 1: T\u00e2che\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-4-panel-industrial-0\"\n\t\t\t\t\t\t\t\taria-pressed=\"true\"\n\t\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"industrial\"\n\t\t\t\t\t\t\t\tdata-step=\"1\"\n\t\t\t\t\t\t\t\taria-label=\"Step 2: Percevoir\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-4-panel-industrial-1\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"industrial\"\n\t\t\t\t\t\t\t\tdata-step=\"2\"\n\t\t\t\t\t\t\t\taria-label=\"Step 3: Localiser\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-4-panel-industrial-2\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"industrial\"\n\t\t\t\t\t\t\t\tdata-step=\"3\"\n\t\t\t\t\t\t\t\taria-label=\"Step 4: Planifier\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-4-panel-industrial-3\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"industrial\"\n\t\t\t\t\t\t\t\tdata-step=\"4\"\n\t\t\t\t\t\t\t\taria-label=\"Step 5: Commander\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-4-panel-industrial-4\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"industrial\"\n\t\t\t\t\t\t\t\tdata-step=\"5\"\n\t\t\t\t\t\t\t\taria-label=\"Step 6: V\u00e9rifier\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-4-panel-industrial-5\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"industrial\"\n\t\t\t\t\t\t\t\tdata-step=\"6\"\n\t\t\t\t\t\t\t\taria-label=\"Step 7: Prot\u00e9ger\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-4-panel-industrial-6\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"industrial\"\n\t\t\t\t\t\t\t\tdata-step=\"7\"\n\t\t\t\t\t\t\t\taria-label=\"Step 8: Am\u00e9liorer\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-4-panel-industrial-7\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"0\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t\t<span\n\t\t\t\t\t\tclass=\"ns-aix-dotset\"\n\t\t\t\t\t\tdata-flow=\"autonomous\"\n\t\t\t\t\t\trole=\"group\"\n\t\t\t\t\t\taria-label=\"Steps for Robot autonome\"\n\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"autonomous\"\n\t\t\t\t\t\t\t\tdata-step=\"0\"\n\t\t\t\t\t\t\t\taria-label=\"Step 1: Mission\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-4-panel-autonomous-0\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"-1\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"autonomous\"\n\t\t\t\t\t\t\t\tdata-step=\"1\"\n\t\t\t\t\t\t\t\taria-label=\"Step 2: Percevoir\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-4-panel-autonomous-1\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"-1\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"autonomous\"\n\t\t\t\t\t\t\t\tdata-step=\"2\"\n\t\t\t\t\t\t\t\taria-label=\"Step 3: Localiser\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-4-panel-autonomous-2\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"-1\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"autonomous\"\n\t\t\t\t\t\t\t\tdata-step=\"3\"\n\t\t\t\t\t\t\t\taria-label=\"Step 4: Mod\u00e9liser\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-4-panel-autonomous-3\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"-1\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"autonomous\"\n\t\t\t\t\t\t\t\tdata-step=\"4\"\n\t\t\t\t\t\t\t\taria-label=\"Step 5: Planifier\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-4-panel-autonomous-4\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"-1\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"autonomous\"\n\t\t\t\t\t\t\t\tdata-step=\"5\"\n\t\t\t\t\t\t\t\taria-label=\"Step 6: Agir\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-4-panel-autonomous-5\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"-1\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"autonomous\"\n\t\t\t\t\t\t\t\tdata-step=\"6\"\n\t\t\t\t\t\t\t\taria-label=\"Step 7: R\u00e9agir\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-4-panel-autonomous-6\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"-1\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<button\n\t\t\t\t\t\t\t\ttype=\"button\"\n\t\t\t\t\t\t\t\tclass=\"ns-aix-dot\"\n\t\t\t\t\t\t\t\tdata-flow=\"autonomous\"\n\t\t\t\t\t\t\t\tdata-step=\"7\"\n\t\t\t\t\t\t\t\taria-label=\"Step 8: Piloter\"\n\t\t\t\t\t\t\t\taria-controls=\"ns-aix-4-panel-autonomous-7\"\n\t\t\t\t\t\t\t\taria-pressed=\"false\"\n\t\t\t\t\t\t\t\ttabindex=\"-1\"\n\t\t\t\t\t\t\t><\/button>\n\t\t\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t<\/div>\n\t\t<\/div>\n\n\t\t<div class=\"ns-aix-below\" aria-live=\"polite\">\n\t\t\t<div class=\"ns-aix-explain\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-4-panel-industrial-0\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel ns-on\"\n\t\t\t\t\t\t\tdata-flow=\"industrial\"\n\t\t\t\t\t\t\tdata-step=\"0\"\n\t\t\t\t\t\t\taria-hidden=\"false\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">01<\/span>D\u00e9finir la t\u00e2che physique<\/h4>\n\t\t\t\t\t\t\t<p>La robotique industrielle commence par une t\u00e2che d\u00e9finie : saisir, d\u00e9poser, souder, inspecter, visser, emballer ou trier. L\u2019environnement est con\u00e7u autour de cette t\u00e2che afin que le robot n\u2019ait pas \u00e0 comprendre le monde entier. Pi\u00e8ce, montage, outil et trajectoire autoris\u00e9e sont d\u00e9finis avant le d\u00e9but de l\u2019automatisation.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-4-panel-industrial-1\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"industrial\"\n\t\t\t\t\t\t\tdata-step=\"1\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">02<\/span>Mesurer la cellule robotis\u00e9e<\/h4>\n\t\t\t\t\t\t\t<p>Les capteurs indiquent au contr\u00f4leur si la r\u00e9alit\u00e9 correspond au programme. Les cam\u00e9ras localisent les pi\u00e8ces, les encodeurs donnent la position des articulations, les capteurs de force d\u00e9tectent les contacts et les capteurs de s\u00e9curit\u00e9 rep\u00e8rent personnes ou obstacles. Sans capteurs, le robot ne peut que r\u00e9p\u00e9ter des mouvements fixes en esp\u00e9rant que la sc\u00e8ne n\u2019a pas chang\u00e9.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-4-panel-industrial-2\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"industrial\"\n\t\t\t\t\t\t\tdata-step=\"2\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">03<\/span>Convertir les mesures en coordonn\u00e9es<\/h4>\n\t\t\t\t\t\t\t<p>Le robot doit savoir o\u00f9 se trouve la pi\u00e8ce par rapport \u00e0 la cam\u00e9ra, au convoyeur, au montage, \u00e0 l\u2019outil et \u00e0 sa propre base. La calibration convertit les donn\u00e9es des capteurs dans un syst\u00e8me de coordonn\u00e9es commun.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-4-panel-industrial-3\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"industrial\"\n\t\t\t\t\t\t\tdata-step=\"3\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">04<\/span>Choisir une trajectoire s\u00fbre<\/h4>\n\t\t\t\t\t\t\t<p>La planification du mouvement calcule comment le robot doit passer de sa pose actuelle \u00e0 la pose cible. Elle doit respecter les limites des articulations, la charge utile, la port\u00e9e, l\u2019orientation de l\u2019outil, les montages, le temps de cycle et les zones de collision.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-4-panel-industrial-4\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"industrial\"\n\t\t\t\t\t\t\tdata-step=\"4\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">05<\/span>D\u00e9placer avec retour d\u2019information<\/h4>\n\t\t\t\t\t\t\t<p>Moteurs, entra\u00eenements, vannes et pr\u00e9henseurs ex\u00e9cutent le mouvement pr\u00e9vu. Les boucles de contr\u00f4le comparent plusieurs fois par seconde le mouvement demand\u00e9 au mouvement r\u00e9el et corrigent position, vitesse, couple ou force. C\u2019est ici que le logiciel devient comportement physique.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-4-panel-industrial-5\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"industrial\"\n\t\t\t\t\t\t\tdata-step=\"5\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">06<\/span>V\u00e9rifier le r\u00e9sultat<\/h4>\n\t\t\t\t\t\t\t<p>Un robot ne doit pas supposer que la t\u00e2che a r\u00e9ussi. Cam\u00e9ras, courbes de couple, traces de force, contr\u00f4les de poids ou postes qualit\u00e9 en aval confirment que la pi\u00e8ce a bien \u00e9t\u00e9 saisie, d\u00e9pos\u00e9e, serr\u00e9e, mesur\u00e9e ou \u00e9cart\u00e9e.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-4-panel-industrial-6\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"industrial\"\n\t\t\t\t\t\t\tdata-step=\"6\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">07<\/span>Limiter ce qui peut se produire<\/h4>\n\t\t\t\t\t\t\t<p>La robotique cr\u00e9e des risques physiques. Zones de s\u00e9curit\u00e9, limites de vitesse, r\u00e9glages de force en mode collaboratif, arr\u00eats d\u2019urgence, modes de maintenance et intervention humaine d\u00e9finissent ce que la machine est autoris\u00e9e \u00e0 faire. La s\u00e9curit\u00e9 n\u2019est pas un ajout \u00e0 la robotique : elle fait partie du syst\u00e8me de commande.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-4-panel-industrial-7\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"industrial\"\n\t\t\t\t\t\t\tdata-step=\"7\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">08<\/span>Apprendre de l\u2019exploitation<\/h4>\n\t\t\t\t\t\t\t<p>Temps de cycle, prises rat\u00e9es, r\u00e9sultats qualit\u00e9, traces des capteurs et donn\u00e9es de maintenance peuvent am\u00e9liorer le syst\u00e8me \u2014 par de meilleurs outils, des r\u00e8gles mises \u00e0 jour, de la maintenance pr\u00e9dictive ou des mod\u00e8les de Machine Learning. Toute am\u00e9lioration doit \u00eatre valid\u00e9e par un ing\u00e9nieur, consign\u00e9e et requalifi\u00e9e avant d\u2019avoir un effet sur le mouvement.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-4-panel-autonomous-0\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"autonomous\"\n\t\t\t\t\t\t\tdata-step=\"0\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">01<\/span>Fixer l\u2019objectif et les limites<\/h4>\n\t\t\t\t\t\t\t<p>Un robot autonome a besoin de plus qu\u2019une destination. Il lui faut un domaine op\u00e9rationnel clairement d\u00e9fini : o\u00f9 il peut se d\u00e9placer, \u00e0 quelle distance des personnes, quand il doit ralentir, quelles conditions sortent de ses limites de conception et quand un humain doit reprendre la main.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-4-panel-autonomous-1\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"autonomous\"\n\t\t\t\t\t\t\tdata-step=\"1\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">02<\/span>Percevoir l\u2019environnement<\/h4>\n\t\t\t\t\t\t\t<p>Cam\u00e9ras, lidar, radar, ultrasons, GPS, capteurs inertiels, encodeurs de roue ou capteurs tactiles ne fournissent chacun qu\u2019une partie de l\u2019information. Tous ont leurs modes de d\u00e9faillance : \u00e9blouissement, pluie, poussi\u00e8re, occultation, reflets, faible luminosit\u00e9, GPS d\u00e9grad\u00e9 ou vibrations.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-4-panel-autonomous-2\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"autonomous\"\n\t\t\t\t\t\t\tdata-step=\"2\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">03<\/span>Estimer la position et l\u2019incertitude<\/h4>\n\t\t\t\t\t\t\t<p>Le robot doit estimer o\u00f9 il se trouve, \u00e0 quelle vitesse il se d\u00e9place et \u00e0 quel point cette estimation est fiable, en comparant les donn\u00e9es de capteurs en direct \u00e0 une carte enregistr\u00e9e ou en construisant cette carte au fur et \u00e0 mesure. Lorsque la confiance dans la localisation baisse, le comportement s\u00fbr doit devenir plus prudent.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-4-panel-autonomous-3\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"autonomous\"\n\t\t\t\t\t\t\tdata-step=\"3\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">04<\/span>Construire un mod\u00e8le de sc\u00e8ne exploitable<\/h4>\n\t\t\t\t\t\t\t<p>Les donn\u00e9es des capteurs deviennent un mod\u00e8le structur\u00e9 du monde : espace libre, obstacles, personnes, v\u00e9hicules, portes, bordures, escaliers, terrain et objets en mouvement. Le syst\u00e8me doit \u00e9galement pr\u00e9voir o\u00f9 les acteurs mobiles sont susceptibles d\u2019aller ensuite.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-4-panel-autonomous-4\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"autonomous\"\n\t\t\t\t\t\t\tdata-step=\"4\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">05<\/span>Choisir le prochain mouvement<\/h4>\n\t\t\t\t\t\t\t<p>La planification se fait \u00e0 plusieurs niveaux : itin\u00e9raire, trajectoire locale, \u00e9vitement d\u2019obstacles, choix de la vitesse et comportement de r\u00e9cup\u00e9ration. Le syst\u00e8me ne doit pas seulement se demander s\u2019il peut atteindre l\u2019objectif, mais s\u2019il peut le faire en s\u00e9curit\u00e9 compte tenu de l\u2019incertitude actuelle.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-4-panel-autonomous-5\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"autonomous\"\n\t\t\t\t\t\t\tdata-step=\"5\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">06<\/span>Commander le corps<\/h4>\n\t\t\t\t\t\t\t<p>Le plan devient mouvement par les moteurs, freins, direction, roues, jambes ou articulations. Un v\u00e9hicule commande direction, acc\u00e9l\u00e9ration et freinage. Un robot marcheur doit en plus garder l\u2019\u00e9quilibre, placer ses pieds et g\u00e9rer des contacts irr\u00e9guliers. Le retour d\u2019information maintient le mouvement r\u00e9el au plus pr\u00e8s du mouvement pr\u00e9vu.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-4-panel-autonomous-6\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"autonomous\"\n\t\t\t\t\t\t\tdata-step=\"6\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">07<\/span>G\u00e9rer les impr\u00e9vus<\/h4>\n\t\t\t\t\t\t\t<p>La robotique en environnement ouvert est mise \u00e0 l\u2019\u00e9preuve par les impr\u00e9vus : une personne entre dans la trajectoire, un capteur est aveugl\u00e9, le sol devient glissant ou un autre v\u00e9hicule se comporte de fa\u00e7on impr\u00e9visible. Une autonomie s\u00fbre sait ralentir, s\u2019arr\u00eater, recalculer, demander de l\u2019aide ou rendre la main \u00e0 un humain.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t<section\n\t\t\t\t\t\t\tid=\"ns-aix-4-panel-autonomous-7\"\n\t\t\t\t\t\t\tclass=\"ns-aix-panel\"\n\t\t\t\t\t\t\tdata-flow=\"autonomous\"\n\t\t\t\t\t\t\tdata-step=\"7\"\n\t\t\t\t\t\t\taria-hidden=\"true\"\n\t\t\t\t\t\t>\n\t\t\t\t\t\t\t<h4><span class=\"ns-aix-panel-num\">08<\/span>Prouver que l\u2019autonomie est ma\u00eetris\u00e9e<\/h4>\n\t\t\t\t\t\t\t<p>Les robots autonomes ont besoin de logs, de dossiers de s\u00e9curit\u00e9, de limites d\u2019utilisation, d\u2019un contr\u00f4le des mises \u00e0 jour, de cybers\u00e9curit\u00e9, de r\u00e8gles de protection des donn\u00e9es, d\u2019une possibilit\u00e9 d\u2019intervention humaine et d\u2019une responsabilit\u00e9 clairement d\u00e9finie. Le comportement pilot\u00e9 par l\u2019IA \u00e9tant plus difficile \u00e0 retracer qu\u2019une logique classique, la gouvernance doit entourer le robot avant qu\u2019on lui confie des actions physiques.<\/p>\n\t\t\t\t\t\t<\/section>\n\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t<div class=\"ns-aix-aspects\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset ns-on\" data-flow=\"industrial\" data-step=\"0\" aria-hidden=\"false\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Contexte<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Les premiers robots industriels ont rejoint les cha\u00eenes de production dans les ann\u00e9es 1960 pour des t\u00e2ches r\u00e9p\u00e9titives comme le soudage ou l\u2019assemblage. Les capteurs et la planification ont ensuite transform\u00e9 des bras fixes en syst\u00e8mes adaptatifs.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Concept cl\u00e9<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>La robotique associe m\u00e9canique, \u00e9lectronique et logiciel : des capteurs pour percevoir, des actionneurs pour agir et un syst\u00e8me de commande pour coordonner l\u2019ensemble.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"industrial\" data-step=\"1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Comment \u00e7a marche<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Un robot ex\u00e9cute en continu un cycle perception-planification-action, plusieurs fois par seconde.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>En pratique<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Cam\u00e9ras, encodeurs, capteurs de force et capteurs de s\u00e9curit\u00e9 r\u00e9pondent chacun \u00e0 une question diff\u00e9rente sur la sc\u00e8ne \u2014 ensemble, ils remplacent les suppositions par des donn\u00e9es.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"industrial\" data-step=\"2\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Concept cl\u00e9<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>La calibration relie cam\u00e9ra, convoyeur, outil et base du robot dans un m\u00eame syst\u00e8me de coordonn\u00e9es \u2014 le pont entre perception et mouvement.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Attention<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Une petite erreur de calibration peut entra\u00eener une prise rat\u00e9e, une mauvaise soudure, une collision ou un d\u00e9faut qualit\u00e9.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"industrial\" data-step=\"3\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Comment \u00e7a marche<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Les trajectoires possibles sont v\u00e9rifi\u00e9es par rapport aux limites des articulations, \u00e0 la charge utile, \u00e0 la port\u00e9e, \u00e0 l\u2019orientation de l\u2019outil, aux montages et aux zones de collision avant tout mouvement r\u00e9el.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Pourquoi c\u2019est important<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>La bonne trajectoire n\u2019est pas simplement la plus courte ; c\u2019est celle qui peut \u00eatre ex\u00e9cut\u00e9e de fa\u00e7on s\u00fbre et r\u00e9p\u00e9table.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"industrial\" data-step=\"4\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Comment \u00e7a marche<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Le contr\u00f4le servo tourne des milliers de fois par seconde : chaque cycle mesure l\u2019\u00e9tat des articulations et corrige les entra\u00eenements. C\u2019est ce qui rend les mouvements industriels fluides et r\u00e9p\u00e9tables.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>En pratique<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Sur les robots industriels simples, les mouvements sont pr\u00e9programm\u00e9s et r\u00e9p\u00e9t\u00e9s avec pr\u00e9cision ; les syst\u00e8mes avanc\u00e9s s\u2019adaptent aux conditions changeantes gr\u00e2ce au Machine Learning et au feedback en temps r\u00e9el.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"industrial\" data-step=\"5\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Pourquoi c\u2019est important<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>La v\u00e9rification transforme une automatisation aveugle en processus ma\u00eetris\u00e9 : chaque cycle produit des preuves plut\u00f4t que des suppositions.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>En pratique<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Un contr\u00f4le \u00e9chou\u00e9 peut d\u00e9clencher une nouvelle tentative automatique, \u00e9carter la pi\u00e8ce ou arr\u00eater la ligne. La r\u00e9action \u00e0 une mauvaise information est con\u00e7ue \u00e0 l\u2019avance, pas improvis\u00e9e.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"industrial\" data-step=\"6\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Concept cl\u00e9<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>L\u2019enveloppe de s\u00e9curit\u00e9 prime sur la vitesse de production : zones, limites, arr\u00eats d\u2019urgence et intervention humaine sont int\u00e9gr\u00e9s directement au contr\u00f4leur.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Attention<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>L\u2019usure m\u00e9canique, les limites des batteries et les d\u00e9faillances du syst\u00e8me cr\u00e9ent des risques physiques pour les personnes et les \u00e9quipements et exigent une maintenance continue.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"industrial\" data-step=\"7\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Ce que la robotique peut faire<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>T\u00e2ches r\u00e9p\u00e9titives, dangereuses ou de haute pr\u00e9cision ; transport en entrep\u00f4t ; aspirateurs domestiques ; exploration de lieux dangereux ou \u00e9loign\u00e9s.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Attention<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>L\u2019autonomie en environnement ouvert reste difficile : conduire dans toutes les conditions comme un humain ou atteindre, avec un humano\u00efde, la dext\u00e9rit\u00e9 d\u2019un enfant demeure un d\u00e9fi majeur et co\u00fbteux.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"autonomous\" data-step=\"0\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Concept cl\u00e9<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>L\u2019objectif est limit\u00e9 par le domaine d\u2019utilisation : une autonomie sans limites explicites n\u2019est pas une conception, c\u2019est un danger.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Pourquoi c\u2019est important<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Se d\u00e9placer dans un espace ouvert, sur une route ou un terrain, reste l\u2019un des probl\u00e8mes les plus difficiles de la robotique et exige beaucoup de capteurs et de puissance de calcul.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"autonomous\" data-step=\"1\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Comment \u00e7a marche<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Les capteurs recueillent des donn\u00e9es sur la distance, la position, la temp\u00e9rature ou le contenu visuel ; le syst\u00e8me de commande fusionne ces indices pour construire une image de l\u2019environnement.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Attention<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Les donn\u00e9es des capteurs sont utiles mais jamais parfaites. Les syst\u00e8mes robustes partent du principe que certaines entr\u00e9es seront d\u00e9grad\u00e9es et recoupent les capteurs entre eux.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"autonomous\" data-step=\"2\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Comment \u00e7a marche<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>GPS, odom\u00e9trie, capteurs inertiels, rep\u00e8res, lidar et indices visuels sont combin\u00e9s \u2014 et peuvent se contredire. L\u2019estimation porte une incertitude, pas une certitude.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Pourquoi c\u2019est important<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>L\u2019incertitude doit entrer dans la d\u00e9cision : un robot qui ignore \u00e0 quel point son estimation est incertaine agit avec une fausse assurance.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"autonomous\" data-step=\"3\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Concept cl\u00e9<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Le mod\u00e8le de sc\u00e8ne est une estimation qui \u00e9volue et sert \u00e0 d\u00e9cider quels mouvements restent s\u00fbrs \u2014 pas une compr\u00e9hension parfaite du monde.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>En pratique<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Un robot d\u2019entrep\u00f4t utilise ce mod\u00e8le pour \u00e9viter les obstacles tout en se dirigeant vers une destination.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"autonomous\" data-step=\"4\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Concept cl\u00e9<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Les algorithmes de planification permettent \u00e0 un robot de naviguer en s\u00e9curit\u00e9, souvent en combinant vision par ordinateur pour reconna\u00eetre les objets et Machine Learning pour s\u2019am\u00e9liorer avec le temps.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Pourquoi c\u2019est important<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Les trajectoires dangereuses doivent \u00eatre \u00e9limin\u00e9es avant le d\u00e9but du mouvement : c\u2019est au stade de la planification que le risque est filtr\u00e9.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"autonomous\" data-step=\"5\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Comment \u00e7a marche<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Le contr\u00f4le fonctionne bien plus vite que la planification : une nouvelle route peut \u00eatre calcul\u00e9e quelques fois par seconde, tandis que l\u2019\u00e9quilibre et le contr\u00f4le des roues ou des articulations corrigent le mouvement des centaines de fois par seconde.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Attention<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Les robots humano\u00efdes capables de marcher et d\u2019utiliser leurs mains avec dext\u00e9rit\u00e9 restent encore loin des capacit\u00e9s d\u2019un jeune enfant.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"autonomous\" data-step=\"6\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Concept cl\u00e9<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Les \u00e9v\u00e9nements inattendus doivent d\u00e9clencher un comportement prudent : passer dans un \u00e9tat s\u00fbr est une fonction pr\u00e9vue, pas un \u00e9chec.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>En pratique<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Un robot de livraison qui s\u2019arr\u00eate et demande une assistance \u00e0 distance se comporte correctement \u2014 ce n\u2019est pas un \u00e9chec.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspectset\" data-flow=\"autonomous\" data-step=\"7\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Pourquoi c\u2019est important<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>L\u2019autonomie ma\u00eetris\u00e9e exige des preuves : sans logs ni dossiers de s\u00e9curit\u00e9, on ne peut pas justifier la confiance accord\u00e9e \u00e0 un syst\u00e8me d\u2019IA physique.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<div class=\"ns-aix-aspect\">\n\t\t\t\t\t\t\t\t\t\t<h5>Attention<\/h5>\n\t\t\t\t\t\t\t\t\t\t<p>Les robots demandent beaucoup d\u2019ing\u00e9nierie, de maintenance et de travail de s\u00e9curit\u00e9 ; leurs d\u00e9faillances peuvent causer des dommages physiques aux personnes et aux biens.<\/p>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t<\/div>\n\t<\/div><\/div>\n<\/div>\n\n<div class=\"et_pb_column_23 et_pb_column et-last-child et_block_column et_pb_css_mix_blend_mode_passthrough et_flex_column_24_24 et_flex_column_24_24_tablet et_flex_column_24_24_phone et_flex_column_24_24_tabletWide\">\n<div class=\"et_pb_text_14 et_pb_text et_pb_bg_layout_light et_pb_module et_flex_module\"><div class=\"et_pb_text_inner\"><h3>The Challenges<\/h3>\n<p>Robotics, particularly when it comes to robots navigating an open space like a road or terrain, is still a very difficult field, requiring\u00a0 significant sensing and computing power. Making an autonomous car drive under all conditions like a human driver is still a significant and expensive challenge. And creating humanoid robots that can walk and dextrously use their \"hands\" is still difficult. Today's models are still far away from reaching the overall ability even a child has when it comes to processing sensory input and turning it into smooth and seamless motion.<\/p>\n<p>Robots also require substantial engineering effort, maintenance, and safety considerations. Battery life, mechanical wear, and system failures can limit performance and create risk of physical damage to people and things.<\/p>\n<\/div><\/div>\n<\/div>\n<\/div>\n<\/div>\n\n<div class=\"et_pb_section_9 et_pb_section et_section_regular et_block_section\">\n<div class=\"et_pb_row_15 et_pb_row et_flex_row ns-cta\">\n<div class=\"et_pb_column_24 et_pb_column et_block_column et_pb_css_mix_blend_mode_passthrough et_flex_column_24_24 et_flex_column_24_24_tabletWide\">\n<div class=\"et_pb_text_15 et_pb_text et_pb_bg_layout_light et_pb_module et_block_module\"><div class=\"et_pb_text_inner\"><p>Understanding these different AI approaches is essential when designing real-world AI systems. Each method has distinct strengths and limitations, which must be carefully considered when selecting the right approach for a given problem.<\/p>\n<p>If you want to see how we put these approaches to work, <a href=\"\/transformation\/\">move to what we can do for you<\/a>.<\/p>\n<\/div><\/div>\n<\/div>\n\n<div class=\"et_pb_column_25 et_pb_column et-last-child et_block_column et_pb_css_mix_blend_mode_passthrough et_flex_column_24_24 et_flex_column_24_24_tabletWide\">\n<div class=\"et_pb_module et_pb_button_module_wrapper et_pb_button_1_wrapper\"><a class=\"et_pb_button_1 et_pb_button et_pb_bg_layout_light et_pb_module et_block_module\" href=\"\/transformation\/\" style=\"text-wrap:balance\">9senses AI Transformation<\/a><\/div>\n<\/div>\n<\/div>\n<\/div>","protected":false},"excerpt":{"rendered":"<p>9senses view on Artificial Intelligence is driven by a realistic view of the opportunities, limits and risks that AI comes with. It is grounded in a view that even though it very well simulates intelligence, it isn&#8217;t consciously intelligent and thus needs oversight.<\/p>","protected":false},"author":15,"featured_media":0,"parent":0,"menu_order":0,"comment_status":"closed","ping_status":"closed","template":"","meta":{"_acf_changed":false,"_ns_ir":"","_ns_ir_pending":"","_ns_structural_eyebrow":"","_ns_ir_live":"","_ns_structural_enabled":"","_ns_structural_body_v1":"","n9tr_seo_title_de_DE":"Was KI wirklich ist - 9senses.ai","n9tr_seo_description_de_DE":"9senses sieht KI realistisch: Chancen, Grenzen und Risiken. KI simuliert Intelligenz sehr gut, ist aber nicht bewusst intelligent und braucht deshalb Aufsicht.","n9tr_seo_title_fr_FR":"","n9tr_seo_description_fr_FR":"","footnotes":""},"class_list":["post-227133","page","type-page","status-publish","hentry"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.9senses.ai\/fr\/wp-json\/wp\/v2\/pages\/227133","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.9senses.ai\/fr\/wp-json\/wp\/v2\/pages"}],"about":[{"href":"https:\/\/www.9senses.ai\/fr\/wp-json\/wp\/v2\/types\/page"}],"author":[{"embeddable":true,"href":"https:\/\/www.9senses.ai\/fr\/wp-json\/wp\/v2\/users\/15"}],"replies":[{"embeddable":true,"href":"https:\/\/www.9senses.ai\/fr\/wp-json\/wp\/v2\/comments?post=227133"}],"version-history":[{"count":345,"href":"https:\/\/www.9senses.ai\/fr\/wp-json\/wp\/v2\/pages\/227133\/revisions"}],"predecessor-version":[{"id":230176,"href":"https:\/\/www.9senses.ai\/fr\/wp-json\/wp\/v2\/pages\/227133\/revisions\/230176"}],"wp:attachment":[{"href":"https:\/\/www.9senses.ai\/fr\/wp-json\/wp\/v2\/media?parent=227133"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}