UE448 - Robotic modelling approaches to Cognitive Sciences [COGSCI 314]


Lieu et planning


  • Autre lieu Paris
    Université Paris-Descartes, 12 rue de l'École de Médecine 75006 Paris
    1er semestre / hebdomadaire, vendredi 09:30-12:30
    du 24 septembre 2021 au 14 janvier 2022


Description


Dernière modification : 26 mai 2021 20:28

Type d'UE
Enseignements fondamentaux de master
Domaine
-
Disciplines
Psychologie et sciences cognitives
Page web
https://docs.google.com/document/d/1hb74o9eIbff-iOw3cEPknjkHNJWEtF0BhEOZQePa1Zo/edit 
Langues
anglais
Mots-clés
Psychologie Sciences cognitives
Aires culturelles
-
Intervenant·e·s

The purpose of this course is to present recent work in cognitive robotics and how it can contribute to fundamental questions in cognitive sciences. Historically tightly related to AI, cognitive robotics puts a strong emphasis on how high-level reasoning and cognition are anchored on low-level sensorimotor processes and on embodiment (the role of the body in cognition). Experimenting with robots also forces one to make hypotheses about how different cognitive functions interact (perception, decision, action, control motor, exploration / curiosity, learning, social interaction) and should be integrated within a cognitive architecture in order to make things work in the real world. The course will present advances in robot learning (incl. deep learning), Bayesian approaches, evolutionary robotics, navigation, human-robot interaction, optimal control, and developmental learning. Links with computational neuroscience & developmental psychology will be shown.

Le but de ce cours est de présenter les travaux récents en robotique cognitive et comment ils peuvent contribuer à des questions fondamentales en sciences cognitives. Historiquement étroitement liée à l'IA, la robotique cognitive insiste beaucoup sur la manière dont le raisonnement et la cognition de haut niveau sont ancrés dans les processus sensorimoteurs de bas niveau et dans un corps qui contribue à la cognition (embodiment). Expérimenter avec des robots oblige également à formuler des hypothèses sur la manière dont différentes fonctions cognitives interagissent (perception, décision, action, contrôle moteur, exploration / curiosité, apprentissage, interaction sociale) et devrait être intégré dans une architecture cognitive afin de faire fonctionner les choses dans le monde réel. Le cours présentera les avancées en apprentissage robotique (y compris par apprentissage profond), les approches Bayésiennes, la robotique évolutionnaire, la navigation, les interactions homme-robot, le contrôle optimal et l'apprentissage développemental. Des liens avec les neurosciences computationnelles et la psychologie du développement seront présentés.

On successful completion of this course, students should be able to:

  • Have a clear overview of current research in cognitive robotics

  • Understand the complementarity with which cognitive robotics approaches can make contributions to cognitive sciences compared to computational neuroscience, experimental psychology and developmental psychology

  • Understand what are the conventions, methods and requirements for publishing a research article in cognitive robotics

  • Choose the appropriate robotic method (reinforcement learning, Bayesian approach, evolutionary approach, neural network approach, etc.) for their own problem

  • Devise a reinforcement learning algorithm for simple robotic tasks (e.g., navigation)

  • Devise a simple human-robot interaction experiment involving emotion expression and other non-verbal signals

  • Have pointers to access relevant material/articles in other subfields of cognitive robotics


Master


  • Séminaires de tronc commun – Sciences cognitives – M1/S1-M2/S3
    Suivi et validation – semestriel hebdomadaire = 6 ECTS
    MCC – Two practical labs + Critical presentation

Renseignements


Contacts additionnels
benoit.girard@isir.upmc.fr
Informations pratiques
-
Direction de travaux des étudiants

The course relies on lectures and interventions by specialists in different subfields of cognitive robotics, whose course is shared every week on Schoology. The course is supplemented by 2 practical labs of 3h00, which permit to have a concrete handover of a subset of methods. Students are encouraged to ask questions in class.

Two practical labs (50%)

Critical presentation with model replication/simulation of a computational article with available code (50%)

Réception des candidats
-
Pré-requis

Good knowledge in maths (probability theory, differential equations, linear algebra) & programming. It is recommended, but not mandatory, to have previously attended MOD 101 (Modelling brain, mind, and behavior), and NEURO 101 (Introduction to Cognitive Neurosciences). This course is complementary to other computational courses at Cogmaster, such as MOD 202 (Computational neuroscience methods). Some of its notions can be useful for COGSCI 310 (Action, decision and volition), for COGSCI 313 (Neuro-economics), for COGSCI 309 (New Approaches to Human Memory), and for the DEC’s course ‘Human voluntary action’.

Dernière modification : 26 mai 2021 20:28

Type d'UE
Enseignements fondamentaux de master
Domaine
-
Disciplines
Psychologie et sciences cognitives
Page web
https://docs.google.com/document/d/1hb74o9eIbff-iOw3cEPknjkHNJWEtF0BhEOZQePa1Zo/edit 
Langues
anglais
Mots-clés
Psychologie Sciences cognitives
Aires culturelles
-
Intervenant·e·s

The purpose of this course is to present recent work in cognitive robotics and how it can contribute to fundamental questions in cognitive sciences. Historically tightly related to AI, cognitive robotics puts a strong emphasis on how high-level reasoning and cognition are anchored on low-level sensorimotor processes and on embodiment (the role of the body in cognition). Experimenting with robots also forces one to make hypotheses about how different cognitive functions interact (perception, decision, action, control motor, exploration / curiosity, learning, social interaction) and should be integrated within a cognitive architecture in order to make things work in the real world. The course will present advances in robot learning (incl. deep learning), Bayesian approaches, evolutionary robotics, navigation, human-robot interaction, optimal control, and developmental learning. Links with computational neuroscience & developmental psychology will be shown.

Le but de ce cours est de présenter les travaux récents en robotique cognitive et comment ils peuvent contribuer à des questions fondamentales en sciences cognitives. Historiquement étroitement liée à l'IA, la robotique cognitive insiste beaucoup sur la manière dont le raisonnement et la cognition de haut niveau sont ancrés dans les processus sensorimoteurs de bas niveau et dans un corps qui contribue à la cognition (embodiment). Expérimenter avec des robots oblige également à formuler des hypothèses sur la manière dont différentes fonctions cognitives interagissent (perception, décision, action, contrôle moteur, exploration / curiosité, apprentissage, interaction sociale) et devrait être intégré dans une architecture cognitive afin de faire fonctionner les choses dans le monde réel. Le cours présentera les avancées en apprentissage robotique (y compris par apprentissage profond), les approches Bayésiennes, la robotique évolutionnaire, la navigation, les interactions homme-robot, le contrôle optimal et l'apprentissage développemental. Des liens avec les neurosciences computationnelles et la psychologie du développement seront présentés.

On successful completion of this course, students should be able to:

  • Have a clear overview of current research in cognitive robotics

  • Understand the complementarity with which cognitive robotics approaches can make contributions to cognitive sciences compared to computational neuroscience, experimental psychology and developmental psychology

  • Understand what are the conventions, methods and requirements for publishing a research article in cognitive robotics

  • Choose the appropriate robotic method (reinforcement learning, Bayesian approach, evolutionary approach, neural network approach, etc.) for their own problem

  • Devise a reinforcement learning algorithm for simple robotic tasks (e.g., navigation)

  • Devise a simple human-robot interaction experiment involving emotion expression and other non-verbal signals

  • Have pointers to access relevant material/articles in other subfields of cognitive robotics

  • Séminaires de tronc commun – Sciences cognitives – M1/S1-M2/S3
    Suivi et validation – semestriel hebdomadaire = 6 ECTS
    MCC – Two practical labs + Critical presentation
Contacts additionnels
benoit.girard@isir.upmc.fr
Informations pratiques
-
Direction de travaux des étudiants

The course relies on lectures and interventions by specialists in different subfields of cognitive robotics, whose course is shared every week on Schoology. The course is supplemented by 2 practical labs of 3h00, which permit to have a concrete handover of a subset of methods. Students are encouraged to ask questions in class.

Two practical labs (50%)

Critical presentation with model replication/simulation of a computational article with available code (50%)

Réception des candidats
-
Pré-requis

Good knowledge in maths (probability theory, differential equations, linear algebra) & programming. It is recommended, but not mandatory, to have previously attended MOD 101 (Modelling brain, mind, and behavior), and NEURO 101 (Introduction to Cognitive Neurosciences). This course is complementary to other computational courses at Cogmaster, such as MOD 202 (Computational neuroscience methods). Some of its notions can be useful for COGSCI 310 (Action, decision and volition), for COGSCI 313 (Neuro-economics), for COGSCI 309 (New Approaches to Human Memory), and for the DEC’s course ‘Human voluntary action’.

  • Autre lieu Paris
    Université Paris-Descartes, 12 rue de l'École de Médecine 75006 Paris
    1er semestre / hebdomadaire, vendredi 09:30-12:30
    du 24 septembre 2021 au 14 janvier 2022