{"id":1736,"date":"2022-11-23T10:22:49","date_gmt":"2022-11-23T09:22:49","guid":{"rendered":"https:\/\/enpc.ergeais.com\/?p=1736"},"modified":"2025-07-29T14:45:19","modified_gmt":"2025-07-29T12:45:19","slug":"unraveling-the-mystery-of-deep-neural-networks","status":"publish","type":"post","link":"https:\/\/ingenius.ecoledesponts.fr\/en\/articles\/unraveling-the-mystery-of-deep-neural-networks\/","title":{"rendered":"Unraveling the mystery of deep neural networks"},"content":{"rendered":"\n\n\n
\"\"
\u00a9 Jakub Jirs\u00e1k (source : Adobe Stock)<\/figcaption><\/figure>\n\n\n\n

Intelligence and artificial vision are at the heart of the research work of Mathieu Aubry, a researcher in the IMAGINE<\/a> team (a component of the Gaspard Monge<\/a> Computer Science Laboratory).<\/p>\n\n\n\n

Currently, as part of projects funded by the French National Research Agency <\/a>(ANR), he is working with several PhD students to develop tools for recognizing non-photo-realistic images in old paintings, engravings and drawings. Why? In particular, to enable historians to automatically identify similarities and borrowings on a large number of works in order, for example to quantitatively analyze the recurrences of an artistic movement or a school of painting. \u201cThis type of system could also be used to identify paper watermarks on old documents and classify them. This could change the way historians work by automating tedious tasks and allowing them to concentrate on analysis,\u201d explains Mathieu Aubry, who is collaborating on these projects with PSL University, the Sorbonne University, the University of Cambridge and the University of California at Berkeley, where he worked for a year. But the team’s field of exploration, which includes some 20 researchers and doctoral students, is much broader. In fact, Mathieu Aubry, his colleagues and doctoral students are working on a host of topics related to artificial vision and machine learning.<\/p>\n\n\n\n

Deep learning<\/em> at the service of Adobe, Valeo or Facebook<\/strong><\/h2>\n\n\n\n

Among them, projects related to the representation and generation of 3D models with Adobe, Valeo, Facebook, etc. The common point of these projects is learning through deep learning<\/em> or deep neural networks, a method that has gained considerable ground since a famous publication in 2012 by the Canadian Geoffrey Hinton’s team1<\/a><\/sup>.\u00a0<\/p>\n\n\n\n

A better understanding and improvement of the algorithms underlying deep neural networks, or the development of new algorithms, is thus one of the foundations of the team’s work. These neural networks work well today for certain tasks, such as recognizing an object when you have many annotated images. \u201cHowever, there is still a great deal of empiricism and an experimental nature to this subject, hence the term ‘black box’ sometimes used\u201d explains Mathieu Aubry. We do not fully understand how this works in theory. And we don’t really understand the influence of the training data used or how the ‘weight’ given to a parameter can affect the results.<\/em> One of the major research axes of the team today is therefore to develop more interpretable models, i.e. whose results and their reasoning are directly visualizable and understandable to the user.<\/p>\n\n\n\n

AI – an interdisciplinary field<\/strong><\/h2>\n\n\n\n

IMAGINE is also interested in the interaction between artificial vision and robotics in the building industry A theme that brought the team to collaborate with Willow, an Inria project team specializing in the recognition and modeling of three-dimensional objects and scenes.<\/p>\n\n\n\n

But the interest in artificial intelligence goes beyond research alone. Indeed, \u201cthe current success of neural networks and deep learning is due to the fact that they can be implemented in a large number of applications. Scientists and companies are constantly coming up with these applications,\u201d says Mathieu Aubry, who even believes that \u201cevery idea can generate start-ups\u201d. In terms of collaboration with start-ups, the team works with the Parisian company Quantcube, which uses AI to automatically analyze large quantities of satellite images and produce economic and environmental indicators.<\/p>\n\n\n\n

As we can see, artificial intelligence concerns various fields of science, but also at various scales: research laboratories, classrooms and companies. This is proof of the great potential of the subject, which concerns the future of everyone. \u201cDeveloping skills and research on these subjects takes time, but it is a real challenge for the future of the the IMAGINE team and our country,\u201d concludes Mathieu Aubry.<\/p>\n\n\n\n

Link established by an algorithm between a painting and two studies from a collection of 195 works of art by Dante Gabriel Rossetti<\/h3>\n\n\n\n
\"\"
Left, The Bower Meadow Study<\/em> (Study of Dancing Girls<\/em>) chalk study, 1872, held at the Birmingham Museum & Art Gallery. In the center, The Bower Meadow<\/em>, oil on canvas, 1872, kept at the Manchester Art Gallery. Right, The Bower Meadow<\/em>, pastel, 1871-1872, in the Fitzwilliam Museum in Cambridge (source: www.wikiart.org).<\/figcaption><\/figure>\n\n\n\n

Text adapted from an article published in Le Cahier des Ponts n\u00b01<\/a>, L’intelligence artificielle, March 2019<\/strong>.<\/p>\n\n\n

  1. \u21a9\ufe0e<\/a><\/li><\/ol>","protected":false},"excerpt":{"rendered":"

    Intelligence and artificial vision are at the heart of the research work of Mathieu Aubry, a researcher in the IMAGINE […]<\/p>\n","protected":false},"author":6,"featured_media":871,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_related_content_post":[],"_related_content_subject":[690],"_related_content_author":[932],"_related_content_category":[1716],"_related_content_folder":[],"_excerpt":"The unprecedented appetite for artificial intelligence in recent decades has led to rapid technological and economic developments in this area. At the origin of this boom are deep neural networks, a mathematical model inspired by the human brain and more specifically by the transmission of information between neurons.<\/strong>","_duration":3,"_manual_duration":false,"footnotes":"[{\"id\":\"358c88a5-df9e-4221-b2ab-6bd7eaa0074f\",\"content\":\"\"}]"},"article-types":[13],"class_list":["post-1736","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","article-types-article"],"has_blocks":true,"block_data":[{"blockName":"enpc\/excerpt","attrs":{"lock":[],"metadata":[],"className":"","style":""},"innerBlocks":[],"innerHTML":"","innerContent":[],"rendered":""},{"blockName":"core\/image","attrs":{"id":1571,"sizeSlug":"large","linkDestination":"none","align":"wide","blob":"","url":"https:\/\/ingenius.ecoledesponts.fr\/wp-content\/uploads\/2022\/07\/CdP_Numero_1_00_Couverture-1024x683.jpeg","alt":"","caption":null,"lightbox":[],"title":"","href":"","rel":"","linkClass":"","width":"","height":"","aspectRatio":"","scale":"","linkTarget":"","lock":[],"metadata":[],"className":"wp-block-image alignwide size-large","style":"","borderColor":"","anchor":""},"innerBlocks":[],"innerHTML":"\n

    \"\"
    \u00a9 Jakub Jirs\u00e1k (source : Adobe Stock)<\/figcaption><\/figure>\n","innerContent":["\n
    \"\"
    \u00a9 Jakub Jirs\u00e1k (source : Adobe Stock)<\/figcaption><\/figure>\n"],"rendered":"\n
    \"\"
    \u00a9 Jakub Jirs\u00e1k (source : Adobe Stock)<\/figcaption><\/figure>\n"},{"blockName":"core\/paragraph","attrs":{"align":"","content":null,"dropCap":false,"placeholder":"","direction":"","lock":[],"metadata":[],"className":"","style":"","backgroundColor":"","textColor":"","gradient":"","fontSize":"","fontFamily":"","borderColor":"","anchor":""},"innerBlocks":[],"innerHTML":"\n

    Intelligence and artificial vision are at the heart of the research work of Mathieu Aubry, a researcher in the IMAGINE<\/a> team (a component of the Gaspard Monge<\/a> Computer Science Laboratory).<\/p>\n","innerContent":["\n

    Intelligence and artificial vision are at the heart of the research work of Mathieu Aubry, a researcher in the IMAGINE<\/a> team (a component of the Gaspard Monge<\/a> Computer Science Laboratory).<\/p>\n"],"rendered":"\n

    Intelligence and artificial vision are at the heart of the research work of Mathieu Aubry, a researcher in the IMAGINE<\/a> team (a component of the Gaspard Monge<\/a> Computer Science Laboratory).<\/p>\n"},{"blockName":"core\/paragraph","attrs":{"align":"","content":null,"dropCap":false,"placeholder":"","direction":"","lock":[],"metadata":[],"className":"","style":"","backgroundColor":"","textColor":"","gradient":"","fontSize":"","fontFamily":"","borderColor":"","anchor":""},"innerBlocks":[],"innerHTML":"\n

    Currently, as part of projects funded by the French National Research Agency <\/a>(ANR), he is working with several PhD students to develop tools for recognizing non-photo-realistic images in old paintings, engravings and drawings. Why? In particular, to enable historians to automatically identify similarities and borrowings on a large number of works in order, for example to quantitatively analyze the recurrences of an artistic movement or a school of painting. \u201cThis type of system could also be used to identify paper watermarks on old documents and classify them. This could change the way historians work by automating tedious tasks and allowing them to concentrate on analysis,\u201d explains Mathieu Aubry, who is collaborating on these projects with PSL University, the Sorbonne University, the University of Cambridge and the University of California at Berkeley, where he worked for a year. But the team's field of exploration, which includes some 20 researchers and doctoral students, is much broader. In fact, Mathieu Aubry, his colleagues and doctoral students are working on a host of topics related to artificial vision and machine learning.<\/p>\n","innerContent":["\n

    Currently, as part of projects funded by the French National Research Agency <\/a>(ANR), he is working with several PhD students to develop tools for recognizing non-photo-realistic images in old paintings, engravings and drawings. Why? In particular, to enable historians to automatically identify similarities and borrowings on a large number of works in order, for example to quantitatively analyze the recurrences of an artistic movement or a school of painting. \u201cThis type of system could also be used to identify paper watermarks on old documents and classify them. This could change the way historians work by automating tedious tasks and allowing them to concentrate on analysis,\u201d explains Mathieu Aubry, who is collaborating on these projects with PSL University, the Sorbonne University, the University of Cambridge and the University of California at Berkeley, where he worked for a year. But the team's field of exploration, which includes some 20 researchers and doctoral students, is much broader. In fact, Mathieu Aubry, his colleagues and doctoral students are working on a host of topics related to artificial vision and machine learning.<\/p>\n"],"rendered":"\n

    Currently, as part of projects funded by the French National Research Agency <\/a>(ANR), he is working with several PhD students to develop tools for recognizing non-photo-realistic images in old paintings, engravings and drawings. Why? In particular, to enable historians to automatically identify similarities and borrowings on a large number of works in order, for example to quantitatively analyze the recurrences of an artistic movement or a school of painting. \u201cThis type of system could also be used to identify paper watermarks on old documents and classify them. This could change the way historians work by automating tedious tasks and allowing them to concentrate on analysis,\u201d explains Mathieu Aubry, who is collaborating on these projects with PSL University, the Sorbonne University, the University of Cambridge and the University of California at Berkeley, where he worked for a year. But the team's field of exploration, which includes some 20 researchers and doctoral students, is much broader. In fact, Mathieu Aubry, his colleagues and doctoral students are working on a host of topics related to artificial vision and machine learning.<\/p>\n"},{"blockName":"core\/heading","attrs":{"textColor":"red","textAlign":"","content":null,"level":2,"levelOptions":[],"placeholder":"","lock":[],"metadata":[],"align":"","className":"wp-block-heading has-red-color has-text-color","style":"","backgroundColor":"","gradient":"","fontSize":"","fontFamily":"","borderColor":"","anchor":""},"innerBlocks":[],"innerHTML":"\n

    Deep learning<\/em> at the service of Adobe, Valeo or Facebook<\/strong><\/h2>\n","innerContent":["\n

    Deep learning<\/em> at the service of Adobe, Valeo or Facebook<\/strong><\/h2>\n"],"rendered":"\n

    Deep learning<\/em> at the service of Adobe, Valeo or Facebook<\/strong><\/h2>\n"},{"blockName":"core\/paragraph","attrs":{"align":"","content":null,"dropCap":false,"placeholder":"","direction":"","lock":[],"metadata":[],"className":"","style":"","backgroundColor":"","textColor":"","gradient":"","fontSize":"","fontFamily":"","borderColor":"","anchor":""},"innerBlocks":[],"innerHTML":"\n

    Among them, projects related to the representation and generation of 3D models with Adobe, Valeo, Facebook, etc. The common point of these projects is learning through deep learning<\/em> or deep neural networks, a method that has gained considerable ground since a famous publication in 2012 by the Canadian Geoffrey Hinton's team1<\/a><\/sup>.\u00a0<\/p>\n","innerContent":["\n

    Among them, projects related to the representation and generation of 3D models with Adobe, Valeo, Facebook, etc. The common point of these projects is learning through deep learning<\/em> or deep neural networks, a method that has gained considerable ground since a famous publication in 2012 by the Canadian Geoffrey Hinton's team1<\/a><\/sup>.\u00a0<\/p>\n"],"rendered":"\n

    Among them, projects related to the representation and generation of 3D models with Adobe, Valeo, Facebook, etc. The common point of these projects is learning through deep learning<\/em> or deep neural networks, a method that has gained considerable ground since a famous publication in 2012 by the Canadian Geoffrey Hinton's team1<\/a><\/sup>.\u00a0<\/p>\n"},{"blockName":"core\/paragraph","attrs":{"align":"","content":null,"dropCap":false,"placeholder":"","direction":"","lock":[],"metadata":[],"className":"","style":"","backgroundColor":"","textColor":"","gradient":"","fontSize":"","fontFamily":"","borderColor":"","anchor":""},"innerBlocks":[],"innerHTML":"\n

    A better understanding and improvement of the algorithms underlying deep neural networks, or the development of new algorithms, is thus one of the foundations of the team's work. These neural networks work well today for certain tasks, such as recognizing an object when you have many annotated images. \u201cHowever, there is still a great deal of empiricism and an experimental nature to this subject, hence the term 'black box' sometimes used\u201d explains Mathieu Aubry. We do not fully understand how this works in theory. And we don't really understand the influence of the training data used or how the 'weight' given to a parameter can affect the results.<\/em> One of the major research axes of the team today is therefore to develop more interpretable models, i.e. whose results and their reasoning are directly visualizable and understandable to the user.<\/p>\n","innerContent":["\n

    A better understanding and improvement of the algorithms underlying deep neural networks, or the development of new algorithms, is thus one of the foundations of the team's work. These neural networks work well today for certain tasks, such as recognizing an object when you have many annotated images. \u201cHowever, there is still a great deal of empiricism and an experimental nature to this subject, hence the term 'black box' sometimes used\u201d explains Mathieu Aubry. We do not fully understand how this works in theory. And we don't really understand the influence of the training data used or how the 'weight' given to a parameter can affect the results.<\/em> One of the major research axes of the team today is therefore to develop more interpretable models, i.e. whose results and their reasoning are directly visualizable and understandable to the user.<\/p>\n"],"rendered":"\n

    A better understanding and improvement of the algorithms underlying deep neural networks, or the development of new algorithms, is thus one of the foundations of the team's work. These neural networks work well today for certain tasks, such as recognizing an object when you have many annotated images. \u201cHowever, there is still a great deal of empiricism and an experimental nature to this subject, hence the term 'black box' sometimes used\u201d explains Mathieu Aubry. We do not fully understand how this works in theory. And we don't really understand the influence of the training data used or how the 'weight' given to a parameter can affect the results.<\/em> One of the major research axes of the team today is therefore to develop more interpretable models, i.e. whose results and their reasoning are directly visualizable and understandable to the user.<\/p>\n"},{"blockName":"core\/heading","attrs":{"textColor":"red","textAlign":"","content":null,"level":2,"levelOptions":[],"placeholder":"","lock":[],"metadata":[],"align":"","className":"wp-block-heading has-red-color has-text-color","style":"","backgroundColor":"","gradient":"","fontSize":"","fontFamily":"","borderColor":"","anchor":""},"innerBlocks":[],"innerHTML":"\n

    AI - an interdisciplinary field<\/strong><\/h2>\n","innerContent":["\n

    AI - an interdisciplinary field<\/strong><\/h2>\n"],"rendered":"\n

    AI - an interdisciplinary field<\/strong><\/h2>\n"},{"blockName":"core\/paragraph","attrs":{"align":"","content":null,"dropCap":false,"placeholder":"","direction":"","lock":[],"metadata":[],"className":"","style":"","backgroundColor":"","textColor":"","gradient":"","fontSize":"","fontFamily":"","borderColor":"","anchor":""},"innerBlocks":[],"innerHTML":"\n

    IMAGINE is also interested in the interaction between artificial vision and robotics in the building industry A theme that brought the team to collaborate with Willow, an Inria project team specializing in the recognition and modeling of three-dimensional objects and scenes.<\/p>\n","innerContent":["\n

    IMAGINE is also interested in the interaction between artificial vision and robotics in the building industry A theme that brought the team to collaborate with Willow, an Inria project team specializing in the recognition and modeling of three-dimensional objects and scenes.<\/p>\n"],"rendered":"\n

    IMAGINE is also interested in the interaction between artificial vision and robotics in the building industry A theme that brought the team to collaborate with Willow, an Inria project team specializing in the recognition and modeling of three-dimensional objects and scenes.<\/p>\n"},{"blockName":"core\/paragraph","attrs":{"align":"","content":null,"dropCap":false,"placeholder":"","direction":"","lock":[],"metadata":[],"className":"","style":"","backgroundColor":"","textColor":"","gradient":"","fontSize":"","fontFamily":"","borderColor":"","anchor":""},"innerBlocks":[],"innerHTML":"\n

    But the interest in artificial intelligence goes beyond research alone. Indeed, \u201cthe current success of neural networks and deep learning is due to the fact that they can be implemented in a large number of applications. Scientists and companies are constantly coming up with these applications,\u201d says Mathieu Aubry, who even believes that \u201cevery idea can generate start-ups\u201d. In terms of collaboration with start-ups, the team works with the Parisian company Quantcube, which uses AI to automatically analyze large quantities of satellite images and produce economic and environmental indicators.<\/p>\n","innerContent":["\n

    But the interest in artificial intelligence goes beyond research alone. Indeed, \u201cthe current success of neural networks and deep learning is due to the fact that they can be implemented in a large number of applications. Scientists and companies are constantly coming up with these applications,\u201d says Mathieu Aubry, who even believes that \u201cevery idea can generate start-ups\u201d. In terms of collaboration with start-ups, the team works with the Parisian company Quantcube, which uses AI to automatically analyze large quantities of satellite images and produce economic and environmental indicators.<\/p>\n"],"rendered":"\n

    But the interest in artificial intelligence goes beyond research alone. Indeed, \u201cthe current success of neural networks and deep learning is due to the fact that they can be implemented in a large number of applications. Scientists and companies are constantly coming up with these applications,\u201d says Mathieu Aubry, who even believes that \u201cevery idea can generate start-ups\u201d. In terms of collaboration with start-ups, the team works with the Parisian company Quantcube, which uses AI to automatically analyze large quantities of satellite images and produce economic and environmental indicators.<\/p>\n"},{"blockName":"core\/paragraph","attrs":{"align":"","content":null,"dropCap":false,"placeholder":"","direction":"","lock":[],"metadata":[],"className":"","style":"","backgroundColor":"","textColor":"","gradient":"","fontSize":"","fontFamily":"","borderColor":"","anchor":""},"innerBlocks":[],"innerHTML":"\n

    As we can see, artificial intelligence concerns various fields of science, but also at various scales: research laboratories, classrooms and companies. This is proof of the great potential of the subject, which concerns the future of everyone. \u201cDeveloping skills and research on these subjects takes time, but it is a real challenge for the future of the the IMAGINE team and our country,\u201d concludes Mathieu Aubry.<\/p>\n","innerContent":["\n

    As we can see, artificial intelligence concerns various fields of science, but also at various scales: research laboratories, classrooms and companies. This is proof of the great potential of the subject, which concerns the future of everyone. \u201cDeveloping skills and research on these subjects takes time, but it is a real challenge for the future of the the IMAGINE team and our country,\u201d concludes Mathieu Aubry.<\/p>\n"],"rendered":"\n

    As we can see, artificial intelligence concerns various fields of science, but also at various scales: research laboratories, classrooms and companies. This is proof of the great potential of the subject, which concerns the future of everyone. \u201cDeveloping skills and research on these subjects takes time, but it is a real challenge for the future of the the IMAGINE team and our country,\u201d concludes Mathieu Aubry.<\/p>\n"},{"blockName":"core\/heading","attrs":{"level":3,"textColor":"medium-grey","textAlign":"","content":null,"levelOptions":[],"placeholder":"","lock":[],"metadata":[],"align":"","className":"wp-block-heading has-medium-grey-color has-text-color","style":"","backgroundColor":"","gradient":"","fontSize":"","fontFamily":"","borderColor":"","anchor":""},"innerBlocks":[],"innerHTML":"\n

    Link established by an algorithm between a painting and two studies from a collection of 195 works of art by Dante Gabriel Rossetti<\/h3>\n","innerContent":["\n

    Link established by an algorithm between a painting and two studies from a collection of 195 works of art by Dante Gabriel Rossetti<\/h3>\n"],"rendered":"\n

    Link established by an algorithm between a painting and two studies from a collection of 195 works of art by Dante Gabriel Rossetti<\/h3>\n"},{"blockName":"core\/image","attrs":{"id":1577,"sizeSlug":"large","linkDestination":"none","align":"center","blob":"","url":"https:\/\/ingenius.ecoledesponts.fr\/wp-content\/uploads\/2022\/07\/Cdp_Num1_Mathieu_Aubry-1024x456.jpg","alt":"","caption":null,"lightbox":[],"title":"","href":"","rel":"","linkClass":"","width":"","height":"","aspectRatio":"","scale":"","linkTarget":"","lock":[],"metadata":[],"className":"wp-block-image aligncenter size-large","style":"","borderColor":"","anchor":""},"innerBlocks":[],"innerHTML":"\n
    \"\"
    Left, The Bower Meadow Study<\/em> (Study of Dancing Girls<\/em>) chalk study, 1872, held at the Birmingham Museum & Art Gallery. In the center, The Bower Meadow<\/em>, oil on canvas, 1872, kept at the Manchester Art Gallery. Right, The Bower Meadow<\/em>, pastel, 1871-1872, in the Fitzwilliam Museum in Cambridge (source: www.wikiart.org).<\/figcaption><\/figure>\n","innerContent":["\n
    \"\"
    Left, The Bower Meadow Study<\/em> (Study of Dancing Girls<\/em>) chalk study, 1872, held at the Birmingham Museum & Art Gallery. In the center, The Bower Meadow<\/em>, oil on canvas, 1872, kept at the Manchester Art Gallery. Right, The Bower Meadow<\/em>, pastel, 1871-1872, in the Fitzwilliam Museum in Cambridge (source: www.wikiart.org).<\/figcaption><\/figure>\n"],"rendered":"\n
    \"\"
    Left, The Bower Meadow Study<\/em> (Study of Dancing Girls<\/em>) chalk study, 1872, held at the Birmingham Museum & Art Gallery. In the center, The Bower Meadow<\/em>, oil on canvas, 1872, kept at the Manchester Art Gallery. Right, The Bower Meadow<\/em>, pastel, 1871-1872, in the Fitzwilliam Museum in Cambridge (source: www.wikiart.org).<\/figcaption><\/figure>\n"},{"blockName":"core\/paragraph","attrs":{"textColor":"medium-grey","align":"","content":null,"dropCap":false,"placeholder":"","direction":"","lock":[],"metadata":[],"className":"has-medium-grey-color has-text-color","style":"","backgroundColor":"","gradient":"","fontSize":"","fontFamily":"","borderColor":"","anchor":""},"innerBlocks":[],"innerHTML":"\n

    Text adapted from an article published in Le Cahier des Ponts n\u00b01<\/a>, L'intelligence artificielle, March 2019<\/strong>.<\/p>\n","innerContent":["\n

    Text adapted from an article published in Le Cahier des Ponts n\u00b01<\/a>, L'intelligence artificielle, March 2019<\/strong>.<\/p>\n"],"rendered":"\n

    Text adapted from an article published in Le Cahier des Ponts n\u00b01<\/a>, L'intelligence artificielle, March 2019<\/strong>.<\/p>\n"},{"blockName":"core\/footnotes","attrs":{"lock":[],"metadata":[],"className":"","style":"","backgroundColor":"","textColor":"","fontSize":"","fontFamily":"","borderColor":""},"innerBlocks":[],"innerHTML":"","innerContent":[],"rendered":""}],"seo":{"title":"Unraveling the mystery of deep neural networks"},"media":{"img":"\"\"","src":"https:\/\/ingenius.ecoledesponts.fr\/wp-content\/uploads\/2021\/10\/CdP_Numero_1_00_Couverture-scaled.jpeg"},"url":"\/en\/articles\/unraveling-the-mystery-of-deep-neural-networks\/","related":{"post":[],"author":[{"title":"Mathieu Aubry","url":"\/en\/authors\/mathieu-aubry\/","id":"932","media":"\"\"","slug":"mathieu-aubry"}],"subject":[{"title":"Digital Technology, Modeling & Artificial Intelligence","url":"\/en\/subjects\/digital-technology-modeling-artificial-intelligence\/","id":"690","media":"\"\"","slug":"digital-technology-modeling-artificial-intelligence"}],"category":[{"title":"Articles","url":"\/en\/articles\/category\/articles\/","id":"1716","media":"","slug":"articles","_related_post_type":""}],"folder":[]},"translated":"https:\/\/ingenius.ecoledesponts.fr\/articles\/percer-le-mystere-des-reseaux-de-neurones-profonds\/","icon":"icon-article","duration":"3","custom_excerpt":"The unprecedented appetite for artificial intelligence in recent decades has led to rapid technological and economic developments in this area. At the origin of this boom are deep neural networks, a mathematical model inspired by the human brain and more specifically by the transmission of information between neurons.<\/strong>","duration_type":"","_links":{"self":[{"href":"https:\/\/ingenius.ecoledesponts.fr\/en\/wp-json\/wp\/v2\/posts\/1736","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/ingenius.ecoledesponts.fr\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/ingenius.ecoledesponts.fr\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/ingenius.ecoledesponts.fr\/en\/wp-json\/wp\/v2\/users\/6"}],"replies":[{"embeddable":true,"href":"https:\/\/ingenius.ecoledesponts.fr\/en\/wp-json\/wp\/v2\/comments?post=1736"}],"version-history":[{"count":5,"href":"https:\/\/ingenius.ecoledesponts.fr\/en\/wp-json\/wp\/v2\/posts\/1736\/revisions"}],"predecessor-version":[{"id":8939,"href":"https:\/\/ingenius.ecoledesponts.fr\/en\/wp-json\/wp\/v2\/posts\/1736\/revisions\/8939"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/ingenius.ecoledesponts.fr\/en\/wp-json\/wp\/v2\/media\/871"}],"wp:attachment":[{"href":"https:\/\/ingenius.ecoledesponts.fr\/en\/wp-json\/wp\/v2\/media?parent=1736"}],"wp:term":[{"taxonomy":"article-types","embeddable":true,"href":"https:\/\/ingenius.ecoledesponts.fr\/en\/wp-json\/wp\/v2\/article-types?post=1736"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}