Le cerveau apprend de mani\u00e8re constante pour nous permettre d\u2019am\u00e9liorer nos actions en fonction de nos exp\u00e9riences. Plusieurs th\u00e9ories visent \u00e0 rendre compte de cette propri\u00e9t\u00e9 fondamentale, l\u2019une des plus populaires \u00e9tant l\u2019apprentissage par renforcement, utilis\u00e9e aussi en intelligence artificielle. Cette th\u00e9orie postule que l\u2019apprentissage \u00e9merge gr\u00e2ce \u00e0 un renforcement sp\u00e9cifique des connections entre les neurones qui sont actifs durant un \u00e9v\u00e9nement, une action ou une suite d\u2019\u00e9v\u00e9nements et d\u2019actions menant \u00e0 une r\u00e9compense. Un des point cl\u00e9 de cette th\u00e9orie est que plus les neurones impliqu\u00e9s sont actifs, plus le renforcement des connections est rapide et solide. Ainsi les \u00e9v\u00e9nements qui activent le plus fortement notre cerveau devraient \u00eatre appris de mani\u00e8re prioritaire par rapport \u00e0 d\u2019autres \u00e9v\u00e9nements.<\/p>\n
Afin de d\u00e9montrer que ce principes guide effectivement l\u2019apprentissage biologique, l\u2019\u00e9quipe de Brice Bathellier (Institut des Neurosciences Paris Saclay) a utilis\u00e9 deux m\u00e9thodes optiques permettant de suivre et de modifier l\u2019activit\u00e9 de larges ensembles de neurones d\u00e9finis g\u00e9n\u00e9tiquement (optog\u00e9n\u00e9tique). Ils ont pu ainsi montrer que la quantit\u00e9 d\u2019activit\u00e9 g\u00e9n\u00e9r\u00e9e dans le syst\u00e8me auditif corr\u00e8le avec la vitesse d\u2019apprentissage lorsque des souris apprennent \u00e0 associer des sons avec une r\u00e9compense. Dans un deuxi\u00e8me temps ils ont pu faire apprendre aux souris \u00e0 associer, non plus des sons, mais des activations pr\u00e9cises, artificielles du syst\u00e8me auditif. Gr\u00e2ce \u00e0 cette manipulation, ils ont pu montrer un lien causal direct entre la quantit\u00e9 d\u2019activit\u00e9 g\u00e9n\u00e9r\u00e9e et la force du renforcement. Ainsi les neurones les plus actifs lors d\u2019un \u00e9v\u00e9nement sont bien ceux s\u00e9lectionn\u00e9s en priorit\u00e9 par les m\u00e9canismes d\u2019apprentissage.<\/p>\n
<\/p>\n
R\u00e9f\u00e9rence<\/strong><\/p>\n Ceballo et al., Cortical recruitment determines learning dynamics and strategy. Nature Communications, 10: 1479 (2019)<\/p>\n https:\/\/rdcu.be\/budts<\/p>\n <\/p>\n Contact Chercheur<\/strong><\/p>\n Brice Bathellier A key point of learning theory demonstrated through optogenetics<\/strong><\/p>\n The brain learns constantly to improve our actions according to our experiences. Several theories aim to account for this fundamental property, one of the most popular being reinforcement learning, also used in artificial intelligence. This theory postulates that learning emerges through a specific reinforcement of connections between neurons that are active during an event, an action, or a sequence of events and actions leading to a reward. One of the key points of this theory is that the more active involved neurons are, the faster and more solid the reinforcement of connection is. Thus the events that most strongly activate our brain should be learned in priority with respect to other events.<\/p>\n In order to demonstrate that this principle actually guides biological learning, the team of\u00a0 Brice Bathellier (Paris Saclay Institute of Neuroscience) used two optical methods to read and modify the activity of large sets of genetically defined neurons (optogenetics). They were able to show that the amount of activity generated in the auditory system correlates with the speed of learning when mice learn to associate sounds with a reward. In a second experiment, they were able to teach mice to associate, no longer sounds, but precise, artificial activations of the auditory system. Through this manipulation, they were able to show a direct causal link between the amount of activity generated and the strength of reinforcement. Thus, the most active neurons during an event are those selected in priority by the learning mechanisms.<\/p>\n <\/p>\n","protected":false},"excerpt":{"rendered":" Le cerveau apprend de mani\u00e8re constante pour nous permettre d\u2019am\u00e9liorer nos actions en fonction de nos exp\u00e9riences. Plusieurs th\u00e9ories visent \u00e0 rendre compte de cette propri\u00e9t\u00e9 fondamentale, l\u2019une des plus populaires \u00e9tant l\u2019apprentissage par renforcement, utilis\u00e9e aussi en intelligence artificielle. Cette th\u00e9orie postule que l\u2019apprentissage \u00e9merge gr\u00e2ce \u00e0 un renforcement sp\u00e9cifique des connections entre les […]<\/p>\n","protected":false},"author":4,"featured_media":11025,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[25],"tags":[31],"class_list":["post-11017","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-uncategorized","tag-actualite-en"],"publishpress_future_action":{"enabled":false,"date":"2026-05-15 06:51:40","action":"change-status","newStatus":"draft","terms":[],"taxonomy":"category"},"publishpress_future_workflow_manual_trigger":{"enabledWorkflows":[]},"_links":{"self":[{"href":"https:\/\/www.neurosciences.asso.fr\/en\/wp-json\/wp\/v2\/posts\/11017","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.neurosciences.asso.fr\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.neurosciences.asso.fr\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.neurosciences.asso.fr\/en\/wp-json\/wp\/v2\/users\/4"}],"replies":[{"embeddable":true,"href":"https:\/\/www.neurosciences.asso.fr\/en\/wp-json\/wp\/v2\/comments?post=11017"}],"version-history":[{"count":1,"href":"https:\/\/www.neurosciences.asso.fr\/en\/wp-json\/wp\/v2\/posts\/11017\/revisions"}],"predecessor-version":[{"id":11019,"href":"https:\/\/www.neurosciences.asso.fr\/en\/wp-json\/wp\/v2\/posts\/11017\/revisions\/11019"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.neurosciences.asso.fr\/en\/wp-json\/wp\/v2\/media\/11025"}],"wp:attachment":[{"href":"https:\/\/www.neurosciences.asso.fr\/en\/wp-json\/wp\/v2\/media?parent=11017"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.neurosciences.asso.fr\/en\/wp-json\/wp\/v2\/categories?post=11017"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.neurosciences.asso.fr\/en\/wp-json\/wp\/v2\/tags?post=11017"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
\nParis-Saclay Institute of Neuroscience (NeuroPSI)
\nDepartment for Integrative and Computational Neuroscience (ICN)
\nUMR9197 CNRS\/University Paris Sud
\nCNRS, Bldg. 32\/33
\n1 Av. de la Terrasse, 91190 Gif-sur-Yvette, France
\nwww.bathellier-lab.org
\nphone: +33 1 69823408
\nmail :\u00a0bathellier@unic.cnrs-gif.fr<\/p>\n