Tutorial on predictive coding

Time
June 30, 2017
Location

Telluride Neuromorphic Workshop Workgroup on Compound Eyes and Event-based Vision

Slides

http://blog.invibe.net/files/2017-06-30_Telluride.html

reference

  • Laurent U. Perrinet. Tutorial on predictive coding, URL . In Telluride Neuromorphic Workshop, Workgroup on Compound Eyes and Event-based Vision, 2017 abstract.

Back to the present: dealing with delays in biological and neuromorphic systems

Time
June 28th, 2017
Location

Workshop on Computational Neuroscience entitled Neuromorphic Event-based Compound Eyes and Vision"

Slides

http://blog.invibe.net/files/2017-06-28_Telluride.html

reference

  • Laurent U. Perrinet. Back to the present: dealing with delays in biological and neuromorphic systems, URL . In Workshop on Computational Neuroscience entitled "New trends and challenges for 2030", 2017 abstract.

PhD Program: course in Computational Neuroscience

Objective
The course aims at introducing students with the major tools that will be necessary during their thesis to model or analyze their neuroscientific results. While it will start by a short, generic introduction, we will then explore different systems at different scales. On the first day, we will study the different possible regimes in which a single neuron can behave, while progressively introducing the theory of dynamical systems to understand these more globally. Then, during the second day, we will introduce methods to analyze neuroscientific data in general, such as Bayesian methods and information theory. This will be implemented by simple practical examples.

program

day 1 : 2017-03-06 : an introduction to Computational Neuroscience

day 2 : 2017-03-13 : Information theory / bayesian models

day 1 - morning : the single neuron

day 1 - afternoon : neural mass models

day 2 - morning : information theory

day 2 - afternoon : bayesian models

contacts

Tutorial: Active inference for eye movements: Bayesian methods, neural inference, dynamics

Time
January 20th, 2017 - 10:45 - 11:45
Location

LACONEU2017: 4th Latin-American Summer School in Computational Neuroscience: http://www.laconeu.cl

Slides

http://blog.invibe.net/files/2017-01-20_LACONEU.html

reference

  • Laurent U. Perrinet. Tutorial: Active inference for eye movements: Bayesian methods, neural inference, dynamics, URL . In LACONEU2017: 4th Latin-American Summer School in Computational Neuroscience, 2017 abstract.

Tutorial: Sparse optimization in neural computations

Time
January 19th, 2017 - 10:45 - 11:45
Location

LACONEU2017: 4th Latin-American Summer School in Computational Neuroscience: http://www.laconeu.cl

Slides

http://blog.invibe.net/files/2017-01-19_LACONEU.html

reference

  • Laurent U. Perrinet. Tutorial: Sparse optimization in neural computations, URL . In LACONEU2017: 4th Latin-American Summer School in Computational Neuroscience, 2017 abstract.

Back to the present: how neurons deal with delays

reference

  • Laurent U. Perrinet. Back to the present: how neurons deal with delays, URL . In Workshop on Computational Neuroscience entitled "New trends and challenges for 2030", 2017 abstract.

RENCONTRES INTERNATIONALES SCIENCES & CINÉMAS

cinéma les Variétés

http://pollymaggoo.org/wp-content/uploads/2016/10/RISC2016_A3-724x1024.jpg L'Association Polly Maggoo http://www.pollymaggoo.org/ programme la 10e édition des RENCONTRES INTERNATIONALES SCIENCES & CINÉMAS (RISC) à Marseille, au cours desquelles l'association programme des films à caractère scientifique. Les projections se déroulent en présence de chercheurs et/ou de cinéastes dans la perspective d’un développement de la culture cinématographique et scientifique en direction des publics scolaires.

Ce dimanche 20 novembre, je suis venu échanger au côté de Serge Dentin et Caroline Renard (Maître de conférences en études cinématographiques à Aix-Marseille Université), autour de films traitant du rapport fiction/réel, de la mémoire, et du temps. Une occasion aussi de parler du métier de chercheur.

Date
25 Avril 2016
Location
cinéma les Variétés
Programmation
"addendum" court métrage de Jérôme Lefdup et "Poétique du cerveau" long métrage de Nurith Aviv

The flash-lag effect as a motion-based predictive shift

Time
Thursday, November 3rd, 2016
Location

Signal, Image, Geometry, Modelling, Approximation (SIGMA) https://www.ceremade.dauphine.fr/~peyre/sigma2016/

Slides

http://blog.invibe.net/files/2016-11-03_SIGMA.html (takes a few seconds to load)

reference

  • Laurent U. Perrinet. The flash-lag effect as a motion-based predictive shift, URL . In Workshop SIGMA'2016: Signal, Image, Geometry, Modelling, Approximation, 2016 abstract.

Reinforcement contingencies modulate anticipatory smooth eye movements

Time
Thursday, November 3rd, 2016
Location

GDR-Vision 2016 - https://gdrvision2016.sciencesconf.org/

Abstract
Natural environments potentially contain several interesting targets for goal-directed be- havior. Thus sensorimotor systems need to operate a competitive selection based on behav- iorally meaningful parameters. Recently, it has been observed that voluntary eye movements such as saccades and smooth pursuit can be considered as operant behaviors (Madelain et al, 2011). Indeed, parameters of saccades such as peak-velocity or latency (Montagnini et al, 2005) as well as smooth pursuit behavior during transient blanking (Madelain et al, 2003) or visually-guided pursuit of ambiguous stimuli (Sch ́’utz et al, 2015) can be modified by reinforcement contingencies. Here we address the question of whether expectancy-based anticipatory smooth pursuit can be modulated by reinforcement contingencies. When pre- dictive information is available, anticipatory smooth pursuit eye movements (aSPEM) is frequently observed before target appearance. Actions that occur at some distance in time from the reinforcement outcome, such as aSPEM -which occurs without any concurrent sen- sory feedback- suffer of the well-known credit assignment problem (Kaelbling et al, 1996). We designed a direction-bias task as a baseline and modified it by setting an implicit eye velocity criterion during anticipation. The nature of the following trial-outcome (reward or punishment) was contingent to the online criterion matching. We observed a dominant graded effect of motion-direction bias and a small modulational effect of reinforcement on aSPEM velocity. A yoked-control paradigm corroborated this result showing a strong reduc- tion in anticipatory behavior when the reward/punishment schedule was not contingent to behavior. An additional classical conditioning paradigm confirmed that reinforcement con- tingencies have to be operant to be effective and that they have a role in solving the credit assignment problem during aSPEM.

reference

  • Jean-Bernard Damasse, Laurent Perrinet, Jeremie Jozefowiez, Laurent Madelain, Anna Montagnini. Reinforcement contingencies modulate anticipatory smooth eye movements. In GDR Vision, Toulouse, Nov 3rd, 2016, 2016 abstract.

Biologically-inspired characterization of sparseness in natural images

EUVIP Session 7: Biologically Inspired Computer Vision (Special Session)

Compound eyes''

Code

https://github.com/bicv/Perrinet16EUVIP

Date
October 26th, 2016
Location
Ecole Centrale Marseille
Slides

see online

Special Session

Biologically Inspired Computer Vision

Conference paper

Perrinet16EUVIP

Reprint

HAL

reference

  • Laurent U. Perrinet. Biologically-inspired characterization of sparseness in natural images, URL URL2 . In EUVIP (Special Session): Biologically Inspired Computer Vision - October 16th, 2016, 2016 abstract.

Categorization of microscopy images using a biologically inspired edge co-occurrences descriptor

EUVIP Session 7: Biologically Inspired Computer Vision (Special Session)

Compound eyes''

Date
October 26th, 2016
Location
Ecole Centrale Marseille
Slides

see online

Special Session

Biologically Inspired Computer Vision

reference

  • Lionel Fillatre, Michel Barlaud, Laurent U. Perrinet. Categorization of microscopy images using a biologically inspired edge co-occurrences descriptor, URL . In EUVIP (Special Session): Biologically Inspired Computer Vision - October 16th, 2016, 2016 abstract.

EUVIP Session 7: Biologically Inspired Computer Vision (Special Session)

Compound eyes''

description of the session

Recent advances in imaging technologies have yielded scientific data at unprecedented detail and volume, leading to the need of a shift of paradigm in image processing and computer vision. Beyond the usual classical von Neumann architecture, one strategy that is emerging in order to process and interpret this amount of data follows from the architecture of biological organisms and shows for instance computational paradigms implementing asynchronous communication with a high degree of local connectivity in sensors or brain tissues. This session aims at bringing together researchers from different fields of Biologically Inspired Computer Vision to present latest results in the field, from fundamental to more specialized topics, including visual analysis based on a computational level, hardware implementation, and the design of new more advanced vision sensors. It is expected to provide a comprehensive overview in the computer area of biologically motivated vision. On the one hand, biological organisms can provide a source of inspiration for new computationally efficient and robust vision models and on the other hand machine vision approaches can provide new insights for understanding biological visual systems. This session covers a wide range of topics from fundamental to more specialized topics, including visual analysis based on a computational level, hardware implementation, and the design of new more advanced vision sensors. In particular, we expect to provide an overview of a few representative applications and current state of the art of the research in this area.

URL

http://www-l2ti.univ-paris13.fr/euvip2016/index.php/86-euvip2016/129-tentative-technical-program-in-detail

date
October 26th, 2016
Location
Ecole Centrale Marseille
Address

38 rue Frédéric Joliot-Curie 13013 Marseille, France Phone : +33 (0)4 91 05 45 45

Programme
  • 13.30 '''Visual System Inspired Algorithm For Contours, Corner And T Junction Detection''', Antoni Buades, Rafael Grompone Von Gioi

  • 13.50 '''Biologically-inspired characterization of sparseness in natural images''', Laurent Perrinet

  • 14.10 Color filter array imitating the random nature of color arrangement in the human cone mosaic Prakhar Amba, David Alleysson

  • 14.30 '''An Illuminant-Independent Analysis Of Reflectance As Sensed By Humans, And Its Applicability To Computer Vision''', Alban Flachot, Phelma, J.Kevin O'Regan, Edoardo Provenzi

  • 14.50 '''Categorization of microscopy images using a biologically inspired edge co-occurrences descriptor''', Lionel Fillatre, Michel Barlaud, Laurent Perrinet

  • biography of organizer and contact information

    Laurent Perrinet is researcher in Computational Neuroscience at the "Institut de Neurosciences de la Timone" (France), a joint research unit (CNRS / Aix-Marseille Université). He graduated from the aeronautics engineering school SUPAERO, in Toulouse (France) with a signal processing and stochastic modelization degree. His research program is focusing in bridging the complex dynamics of realistic models of large-scale models of spiking neurons with functional models of low-level vision. His current challenge is to be able to translate, or compile in computer terminology, such functional models into neural architectures that would exhibit similar dynamics. He was co-editor of the book "Biologically inspired computer vision: fundamentals and applications", Wiley VCH, 2016. Contact Information

    Eye movements as a model for active inference

    Time
    October 13th, 2016
    Location

    The Lyon Active Inference Workshop (LAW) https://law2016.sciencesconf.org/

    Slides

    http://blog.invibe.net/files/2016-10-13_LAW.html

    reference

    • Laurent U. Perrinet. Eye movements as a model for active inference, URL . In Lyon Active Inference Workshop (LAW) https://law2016.sciencesconf.org/ - October 13th, 2016, 2016 abstract.

    Modelling the dynamics of cognitive processes: from the Bayesian brain to particles

    Time
    July 7th, 2016
    Location

    Summer School: PDE and Probability for Life Sciences @ CIRM, Marseille - http://scientific-events.weebly.com/prog-1426.html

    Slides

    http://blog.invibe.net/files/2016-07-07_EDP-proba

    reference

    • Laurent U. Perrinet. Modelling the dynamics of cognitive processes: from the Bayesian brain to particles, URL . In Summer School: PDE and Probability for Life Sciences @ CIRM, Marseille - http://scientific-events.weebly.com/prog-1426.html - CIRM, July 7th, 2016, 2016 abstract.

    Les illusions visuelles, un révélateur du fonctionnement de notre cerveau

    Cycle de conférences "Tous connectés", Bibliothèque de Méjanes

    conférence tout public à la Bibliothèque de Méjanes (Aix-en-Provence, Avril 2016)

    Date
    28 Avril 2016
    Location
    Bibliothèque de Méjanes
    Visuels

    HTML

    Résumé
    Les illusions visuelles sont des créations d'artistes, de scientifiques et plus récemment, grâce aux réseaux sociaux, du grand public qui proposent des situations souvent incongrues, dans lesquelles l'eau remonte une cascade, les personnes volent dans les airs ou des serpents se mettent à tourner. Au-delà de leur indéniable coté ludique, ces illusions nous apprennent beaucoup sur le fonctionnement du cerveau. En tant que chercheur en Neurosciences à l'Institut de Neurosciences de la Timone à Marseille, je vous dévoilerai des aspects du fonctionnement du cerveau qui sont souvent méconnus. En particulier, nous verrons pourquoi un magicien peut tromper nos sens ou comment des objets peuvent voyager dans le temps. Surtout nous essaierons de comprendre le fonctionnement de notre perception visuelle sur les bases d'une théorie de la vision non pas comme une simple caméra qui enregistre des images mais comme un processus actif en relation avec le monde qui nous entoure.

    Les illusions visuelles, un révélateur du fonctionnement de notre cerveau

    Cinésciences, collège Clair Soleil

    L'Association Polly Maggoo http://www.pollymaggoo.org/ met en place tout le long de l’année, des actions de culture scientifique et artistique en direction des collèges et des lycées, les Cinésciences, au cours desquelles l'association programme des films à caractère scientifique, au sein d’établissements scolaires. Les projections se déroulent en présence de chercheurs et/ou de cinéastes dans la perspective d’un développement de la culture cinématographique et scientifique en direction des publics scolaires.

    Ce lundi 25 avril de 9h à 12h, je suis venu échanger au côté de Serge Dentin autour de films traitant du rapport fiction/réel, des illusion visuelles (" Qu’est ce qu’une image? "), des rapports d’échelles, de la perception, ... et qui sont projetés lors de la séance, avec des élèves de 4e lors d’une séance Cinésciences au collège Clair Soleil, 53 Boulevard Charles Moretti, 13014 Marseille. Une occasion aussi de parler du métier de chercheur.

    Date
    25 Avril 2016
    Location
    collège Clair Soleil, Marseille
    Visuels

    HTML

    Programmation
  • LAZARUS MIRAGES : TÉLÉPATHIE À L'UNIVERSITÉ DE SHANGAI de Patric JEAN et Henry BROCH (France, 2012, documentaire, 3'21): Découvrez une surprenante expérience de transmission de pensée réalisée sous le contrôle de scientifiques de l'université de Shanghai...

  • TONDO de Jérémie VAN QUYNH (France, expérimental, 2015, 3'57): Expérience visuelle et sonore où chaque spectateur se laisse emporter dans un voyage hypnotique.

  • RELIEF DE L'INVISIBLE : PAPILLON de Pierre Oscar LÉVY, Gabriel TURKIEH et Jean-Michel SANCHEZ (France, 2000, série documentaire, 3'): Une plongée dans la matière de l'échelle un à l'échelle atomique, à partir d'images réelles de microscopie électronique, avec un traitement numérique qui permet une parfaite fluidité du mouvement : http://www.universcience.tv/video-papillon-4728.html

  • RHIZOME de Boris LABBÉ (France, animation, 2015, 11'25): De l'infiniment petit à l'infiniment grand, une combinaison de mouvements en perpétuelle métamorphose.

  • CARLITOPOLIS de Luis NIETO (France, 2005, performance, 3'): Un étudiant présente son projet de fin d'études devant un jury. Un acte banal qui peu à peu devient une performance absurde et trompeuse...
  • CORRESPONDANCE(S) : L'ARAIGNÉE ET LE NEURONE d'Hervé NISIC (France, 2010, série documentaire, 1'30) : L'araignée souriante d'Odilon Redon et une cellule nerveuse du cerveau racontées par le biologiste Jean Claude Ameisen.

  • PLANET A de Momoko SETO (France, 2008, art vidéo/animation, 7'40") : Le monde est devenu une vaste planète desséchée, où la culture du coton exercée à outrance pour des raisons économiques, est la cause principale de la désertification. Un désert salin recouvre des hectares de terrain asséché où apparaissent de curieux arbres de sel. Ce phénomène fait écho à une plus grande catastrophe écologique, la désertification de la mer d'Aral, et toujours l'homme comme responsable...
  • Objective
    The course aims at introducing students with the major tools that will be necessary during their thesis to model or analyze their neuroscientific results. While it will start by a short, generic introduction, we will then explore different systems at different scales. On the first day, we will study the different possible regimes in which a single neuron can behave, while progressively introducing the theory of dynamical systems to understand these more globally. Then, during the second day, we will introduce methods to analyze neuroscientific data in general, such as Bayesian methods and information theory. This will be implemented by simple practical examples.

    Program

    day 1 : 2015-12-07 : Computational Neuroscience

    day 2 : 2015-12-08 : Information theory / statistics of decoding

    more material

    contacts

    Motion-based prediction with neuromorphic hardware

    Universidad Técnica Federico Santa María, Valparaíso, Chile

    Time
    November 5th, 2015
    Location

    http://www.eventos.usm.cl/evento/charla-motion-based-prediction-with-neuromorphic-hardware/

    Slides

    slides

    reference

    • Laurent U. Perrinet. Motion-based prediction with neuromorphic hardware, URL . In Universidad Técnica Federico Santa María, Valparaíso, Chile, November 5th, 2015, 2015 abstract.

    Motion-based prediction with neuromorphic hardware

    First GDR BioComp workshop, Saint-Paul de Vence

    First GDR BioComp workshop, Saint-Paul de Vence

    Time
    October 7th, 2015
    Location

    http://gdr-biocomp.fr/colloque/

    Slides

    slides

    reference

    • Laurent U. Perrinet. Motion-based prediction with neuromorphic hardware, URL . In First GDR BioComp workshop, Saint-Paul de Vence, October 7th, 2015, 2015 abstract.

    Edge co-occurrences can account for rapid categorization of natural versus animal images

    Architecture of the model
    Figure 1: Edge co-occurrences (A) An example image with the list of extracted edges overlaid. Each edge is represented by a red line segment which represents its position (center of segment), orientation, and scale (length of segment). . (B) The relationship between a reference edge "A" and another edge "B" can be quantified in terms of the difference between their orientations, ratio of scale, distance between their centers, and difference of azimuth. This is used to compute the chevron map in Figure 2. Go back to manuscript page.

    Un séminaire de l'équipe SIS aura lieu le lundi 22 juin 2015 à 11h00 dans la salle de conférence de l'I3S.

    Laurent Perrinet, chargé de recherche à l'Institut de Neurosciences de la Timone (Marseille, France), nous présentera des résultats récents en catégorisation d'images publiés dans Nature Scientific Reports.

    Titre
    Edge co-occurrences can account for rapid categorization of natural versus animal images.
    Résumé
    Making a judgment about the semantic category of a visual scene, such as whether it contains an animal, is typically assumed to involve high-level associative brain areas. Previous explanations require progressively analyzing the scene hierarchically at increasing levels of abstraction, from edge extraction to mid-level object recognition and then object categorization. Here we show that the statistics of edge co-occurrences alone are sufficient to perform a rough yet robust (translation, scale, and rotation invariant) scene categorization. We first extracted the edges from images using a scale-space analysis coupled with a sparse coding algorithm. We then computed the "association field" for different categories (natural, man-made, or containing an animal) by computing the statistics of edge co-occurrences. These differed strongly, with animal images having more curved configurations. We show that this geometry alone is sufficient for categorization, and that the pattern of errors made by humans is consistent with this procedure. Because these statistics could be measured as early as the primary visual cortex, the results challenge widely held assumptions about the flow of computations in the visual system. The results also suggest new algorithms for image classification and signal processing that exploit correlations between low-level structure and the underlying semantic category.
    Slides

    slides, slides with notes

    Code

    For more information on Matching Pursuit on naturalimages, follow The hitchhiker guide to Matching Pursuit.

    HDR: “Codage prédictif dans les transformations visuo-motrices”

    Madame, Monsieur,

    J'ai soutenu mon habilitation à diriger des recherches (HDR) :

    Quand
    le 17 avril 2014 de 14 H30 à 16 H 30,
    Quoi
    “Codage prédictif dans les transformations visuo-motrices”.

    Le manuscript est disponible sur la page http://invibe.net/LaurentPerrinet/Publications/Perrinet14hdr.

    Salle Henri Gastaut , Faculté de Médecine de Marseille
    • Institut de Neurosciences de la Timone (UMR 7289) 
      Aix Marseille Université, CNRS 
      Faculté de Médecine - Bâtiment Neurosciences 
      27, Bd Jean Moulin 
      13385 Marseille Cedex 05 
      France
    • Plan d'accès: http://www.int.univ-amu.fr/-contact-

    • Ligne 1 (bleu) direction "La Fourragère", stop at "La Timone" and take exit "Hopital de la Timone".
    • From metro station "La Timone" to INT http://goo.gl/maps/jwT04

    La soutenance est ouverte à tous, merci d’annoncer votre présence à laurent.perrinet@univ-amu.fr

    Le jury est composé par
  • Prof. Laurent Madelain, Université Lille III
  • Dr. Alain Destexhe, Université Paris XI (Rapporteur)
  • Prof. Gustavo Deco, Universitat Pompeu Fabra, Barcelona (Rapporteur)
  • Dr. Guillaume Masson, Aix-Marseille Université
  • Dr. Viktor Jirsa, Aix-Marseille Université (Rapporteur)
  • Prof. J.-L. Mege, Aix-Marseille Université
  • WP5 - Demo 1.3 : Spiking model of motion-based prediction

    4th BrainScaleS Plenary meeting

    Time
    From Thu, 20 March 2014 until Fri, 21 March 2014
    Location

    in Manchester (UK), get the program (internal link)

    Slides

    slides, slides with notes

    reference

    • Laurent U. Perrinet, Bernhard A. Kaplan, Mina A. Khoei, Anders Lansner, Guillaume Masson. WP5 - Demo 1.3 : Spiking model of motion-based prediction, URL . In 4th BrainScaleS Plenary meeting - March 20th, 2014, 2014 abstract.

    Axonal delays and on-time control of eye movements

    Marseille INT Fest

    Problem statement: optimal motor control under axonal delays.
    Problem statement: optimal motor control under axonal delays. The central nervous system has to contend with axonal delays, both at the sensory and the motor levels. For instance, in the human visuo-oculomotor system, it takes approximately $\tau_s=50~\ms$ for the retinal image to reach the visual areas implicated in motion detection, and a further $\tau_m=40~\ms $ to reach the oculomotor muscles. As a consequence, for a tennis player trying to intercept a ball at a speed of $20~\m\cdot \s^{-1}$, the sensed physical position is $1~\m$ behind the true position (as represented here by $\tau_s \cdot \vec{V}$), while the position at the moment of emitting the motor command will be $.8~\m$ ahead of its execution ($\tau_m \cdot \vec{V}$). Note that while the actual position of the ball when its image hits photoreceptors on the retina is approximately at $45$ degrees of eccentricity (red dotted line), the player's gaze is directed to the ball at its \emph{present} position (red line), in anticipatory fashion. Optimal control directs action (future motion of the eye) to the expected position (red dashed line) of the ball in the future --- and the racket (black dashed line) to the expected position of the ball when motor commands reach the periphery (muscles).

    Time
    January 10th, 2014, 11:30am
    Location
    CAV/LPP - 45 rue des Saints Pères - salle H432

    reference

    • Laurent U. Perrinet. Axonal delays and on-time control of eye movements, URL . In Marseille INT Fest, January 10th, 2014, 2014 abstract.

    Demo 1-3: Apparent Motion in V1/ MT/MST: Neural Implementation of Probabilistic Approaches

    • Together with Bernhard Kaplan, we talked about how we aim at "compiling" a predictive motion-based approach as a spiking neural networks and then as a parallel wafer systems in the BrainscaleS project (Demo 1, Task4).
    • slides

    • (private to the consortium: meeting info &Agenda, including copies of the slides)

    reference

    • Bernhard Kaplan, Laurent Perrinet. Demo 1, Task4: Implementation of models showing emergence of cortical fields and maps, URL . In Demo 1-3: Apparent Motion in V1/ MT/MST: Neural Implementation of Probabilistic Approaches, 2013 abstract.

    Edge co-occurrences and categorizing natural images

    a seminar for the 20 years of the CerCo

    Time
    July 5th, 2013
    Location

    CerCo, Toulouse

    URL

    http://20anscerco.ups-tlse.fr/Programme%2020%20ans.pdf

    See also

    Talk on edge statistics in natural images at ANC (Edinburgh) on Thursday, January 24th, 2012.

    Slides

    slides - slides with notes

    reference

    • Laurent Perrinet, David Fitzpatrick, James A. Bednar. Edge co-occurrences and categorizing natural images, URL . In A seminar at the CerCo, Toulouse, France, 2013 abstract.

    Why methods and tools are the key to artificial brain-like systems

    3rd BrainScaleS Plenary Meeting

    Time
    March 21st, 2013
    Location
    INT, Marseille.

    reference

    • Laurent U. Perrinet. Why methods and tools are the key to artificial brain-like systems, URL . In 3rd BrainScaleS Plenary Meeting - Friday, March 21st, 2013, 2013 abstract.

    Edge statistics in natural images versus laboratory animal environments: implications for understanding lateral connectivity in V1

    See also:: Talk at Sfn (Washington) in November 2011 or at ANC (Edinburgh) on Thursday, January 24th, 2012.

    Slides:: slides

    Natural

    Laboratory

    edgestats_vanilla_proba-angle_natural.png

    edgestats_vanilla_proba-angle_laboratory.png

    Probability distribution function of "chevrons" in natural and laboratory images. By computing measures of the independence of the different variables, we found that the probability density function of the second-order statistics of edges factorizes with on one side distance and scale and on the other side the 2 angles. The first component proved to be quite similar across both classes and the greater difference is seen for different angle configuration. As it can be reduced to 2 dimensions, we can plot the full probability as shown here by different contrast values assigned to all possible chevrons configurations, for all possible "azimuth" values $\phi$ on the horizontal axis and difference of orientation $\theta$ on the vertical axis. Such a plot most strikingly shows the difference between these 2 classes.

    reference

    • Laurent Perrinet, David Fitzpatrick, James A. Bednar. Edge statistics in natural images versus laboratory animal environments: implications for understanding lateral connectivity in V1, URL . In iTWIST '12 workshop, 2012 abstract.

    Motion-based prediction is sufficient to solve the aperture problem

    a INT-INRIA meeting session

    Time
    Monday, April 16th
    Location
    INRIA.
  • Relevant documents:
  • read more :
  • sequence_DCBA.gif

    sequence_ABCD.gif

    A predictive sequence is essential in resolving the aperture problem. The sequence in which a set of local motion is shown is essential for the detection of global motion. we replicate here the experiments by Scott Watamaniuk and colleagues. They have shown behaviourally that a dot in noise is much more detectable when it follows a coherent trajectory, up to an order of magnitude of 10 times what would be predicted by the local components of the trajectory. (Left) In this first movie we observe white noise and at first sight, no information is detectable. In fact, there is a dot moving along some smooth linear trajectory, but we broke this trajectory into eight equal parts and shuffled their order in the movie. (Right) if we re-arrange these local motions to be compatible with a predictive sequence, it is much easier to see the dot (from left to right in the top of the image, a smooth pursuit helps to catch it). This simple experiment shows that, even if local motion is similar in both movies, a coherent trajectory is more easy to track. Obviously, we may thus conclude that the whole trajectory is more that its individual parts, and that the independence hypothesis does not hold if we want to account for the predictive information in input sequences such as seems to be crucial for the AP.

    Apparent motion in V1 - Probabilistic approaches

    2nd BrainScaleS Plenary Meeting

    Time
    March 23rd, 2012, from 14:00 am to 14:15 pm
    Location
    Forschungszentrum Jülich.
    Slides

    slides, slides with notes

    reference

    • Laurent U. Perrinet. Apparent motion in V1 - Probabilistic approaches, URL . In 2nd BrainScaleS Plenary Meeting - Friday, March 23rd, 2012, 2012 abstract.

    Motion Clouds: Model-based stimulus synthesis of natural-like random textures for the study of motion perception

    2nd BrainScaleS Plenary Meeting

    Time
    March 22nd, 2012, from 14:00 am to 14:15 pm
    Location
    Forschungszentrum Jülich.
    Slides

    slides, slides with notes

    reference

    • Laurent U. Perrinet. Motion Clouds: Model-based stimulus synthesis of natural-like random textures for the study of motion perception, URL . In 2nd BrainScaleS Plenary Meeting - Friday, March 22nd, 2012, 2012 abstract.

    Grabbing, tracking and sniffing as models for motion detection and eye movements

    Brain meeting @ FIL, London

    Time
    Jan 27, 2012, from 16:15 am to 17:00 pm
    Location
    Seminar room @ FIL, Queen's square, 4th floor.

    reference

    • Laurent U. Perrinet. Grabbing, tracking and sniffing as models for motion detection and eye movements, URL . In Brain meeting @ FIL, London - Friday, January 27th, 2012, 2012 abstract.

    Edge statistics in natural images versus laboratory animal environments: implications for understanding lateral connectivity in V1.

    a seminar from the Institute for Adaptive and Neural Computation (ANC)

    Time
    Jan 24, 2012, from 11:00 am to 12:00 pm
    Location
    IF 4.31/4.33
    URL

    http://www.anc.ed.ac.uk/events/anc-dtc-seminar-laurent-perrinet

    reference

    • Laurent Perrinet, David Fitzpatrick, James A. Bednar. Edge statistics in natural images versus laboratory animal environments: implications for understanding lateral connectivity in V1, URL . In A seminar from the Institute for Adaptive and Neural Computation (ANC), 2012 abstract.

    Motion-based prediction is sufficient to solve the aperture problem

    a vision@UCL seminar

    Time
    Thursday, 12th January, 5pm
    Location

    Malet Place Eng Bldg 1.03 (first floor). Route from Russel square tube station.

    Slides

    slides, slides with notes

  • read more :
  • sequence_DCBA.gif

    sequence_ABCD.gif

    A predictive sequence is essential in resolving the aperture problem. The sequence in which a set of local motion is shown is essential for the detection of global motion. we replicate here the experiments by Scott Watamaniuk and colleagues. They have shown behaviourally that a dot in noise is much more detectable when it follows a coherent trajectory, up to an order of magnitude of 10 times what would be predicted by the local components of the trajectory. (Left) In this first movie we observe white noise and at first sight, no information is detectable. In fact, there is a dot moving along some smooth linear trajectory, but we broke this trajectory into eight equal parts and shuffled their order in the movie. (Right) if we re-arrange these local motions to be compatible with a predictive sequence, it is much easier to see the dot (from left to right in the top 25% line of the image, a smooth, slow pursuit helps to catch it). This simple experiment shows that, even if local motion is similar in both movies, a coherent trajectory is more easy to track. Obviously, we may thus conclude that the whole trajectory is more that its individual parts, and that the independence hypothesis does not hold if we want to account for the predictive information in input sequences such as seems to be crucial for the AP.

    reference

    • Laurent Perrinet. Motion-based prediction is sufficient to solve the aperture problem, URL . In Vision@UCL seminar - Thursday, 12th January, 5pm, 2012 abstract.

    Implementation of models showing emergence of cortical fields and maps

    • Here, I talk about how we aim at "compiling" such models as neural networks and as parallel wafer systems in the BrainscaleS project (Demo 1, Task4). (1) I give 3 examples of models (Bayesian or neural) and how we may translate them in a pyNN + ESS compatible implementation (2) we will show how these can be validated by behavioral data collected at the lab but also do some predictions.
    • slides

    • (private to the consortium: meeting info &Agenda, including copies of the slides)

    reference

    • Laurent Perrinet. Demo 1, Task4: Implementation of models showing emergence of cortical fields and maps, URL . In Using the ESS + Neuromorphic hardware Workshop,5th Oktober, 2011 at TU Dresden, Germany, 2011 abstract.

    Edge statistics in natural images versus laboratory animal environments: implications for understanding lateral connectivity in V1

    reference

    • Laurent Perrinet, David Fitzpatrick, James A. Bednar. Edge statistics in natural images versus laboratory animal environments: implications for understanding lateral connectivity in V1, URL . In Proceedings of SfN, 2011, 2011 abstract.

    Atelier Neurosciences Computationnelles, 2-3 Juillet 2011 Khemisset, Maroc

    Propriétés émergentes d'un modèle de prédiction probabiliste utilisant un champ neural

    • La finalité de cette manifestation est de permettre à nos chercheurs de se réunir en groupes de travail et en ateliers afin de découvrir la thématique des neurosciences et son interdisciplinarité. La manifestation se tient dans le cadre des activités du laboratoire LAMS, de ABC MATHINFO, du GDRI NeurO et du réseau méditerranéen NeuroMed.

    reference

    • Laurent Perrinet. Propriétés émergentes d'un modèle de prédiction probabiliste utilisant un champ neural, URL . In Atelier Neurosciences Computationnelles, 2-3 Juillet 2011 Khemisset, Maroc, 2011 abstract.

    NeuroMed

    From Mathematical Image Analysis to Neurogeometry of the Brain

    LADISLAV TAUC CONFERENCES & GDR MSPC NEUROSCIENCE Joint Meeting

    sequence_DCBAline.gif

    sequence_ABCDline.gif

    A predictive sequence is essential in resolving the aperture problem. (Top) In order to demonstrate that, I show in this sequence different apertures; for the aperture problem, motions are segregated (appear to be part of different objects) and still appear to be diagonal. (Right) However, if the same apertures are arranged in an order which is compatible with a predictive sequence, the horizontal motion appears as the most probable: prediction rotates the perception of the direction...

    reference

    • Laurent Perrinet. Probabilistic models of the low-level visual system: the role of prediction in detecting motion, URL . In LADISLAV TAUC & GDR MSPC NEUROSCIENCES CONFERENCE, From Mathematical Image Analysis to Neurogeometry of the Brain, 2010 abstract.

    Include: Nothing found for "----"!

    Let's try it

    • you'll now just need the simple receptive field script and issue

      python SimpleReceptiveField.py 
    • the demo consists of three panels: the input image, the input to the neuron, the output. the receptive field is defined by the first frame that is grabbed. Therefore after first use, I recommend to quit the program, draw a simple stimulus (a line on a paper, a grating, ...) and restart the program by holding the drawing close to the camera.

    [ATTACH]

    Conclusion

    • (!) It's easy to do! A simple installation setup, a short script is enough to demonstrate an essential neuroscience experiment. The simplicity of python is essential in leveraging the complexity of hooking together different pieces of hardware.

    • /!\ There's yet no API for real-time neural simulations to my knowledge. Non von-Neumann architectures (that is algorithms using a different substrate than sequential computers) are becoming a possibility in the near future. Building such applications with real-time demo will enable to study neural models in a different way that the present 1) set-up 2) run 3) analyze workflow that is used in most neural simulators. Yet, this demo is over-simplistic and do not use the computational power of parallelism yet.

    • more info @ SimpleCellDemo page

    Include: Nothing found for "----"!

    Putting the things together...

    the "eye"

    http://code.astraw.com/projects/motmot/_images/homepage.png

    http://code.astraw.com/projects/motmot/_images/plugin-flytrax-linux-639x437.png

    • but it's good to install required libraries from motmot:

      sudo easy_install motmot.utils motmot.wxvalidatedtext  motmot.FlyMovieFormat  motmot.wxglvideo  motmot.wxvideo 

      so that you can run fview to test your isight (check Camera... Initialize camera).

    http://code.astraw.com/projects/motmot/_images/ctrax-screenshot-tiny.jpg

    the loudspeaker

    • for the demo script, you will need the pyaudio module:

      svn co https://www.portaudio.com/repos/portaudio/trunk portaudio
      cd portaudio/
      ./configure
      make
      sudo make install
      sudo /usr/bin/install -c -m 644 -m 644 ./include/pa_mac_core.h /usr/local/include/pa_mac_core.h
      sudo easy_install pyaudio

    glueing using Traits

    • controller

         1 # HasTraits class that supplies the callable for the timer event.
         2 class TimerController(HasTraits):
         3     def onTimer(self, *args):
         4         global hist
         5         im = get_webcam_data()
         6         corr, Vm = neuron(im)
         7         corr_data = self.corr_data.get_data('corr')
         8         corr_data = hstack((corr_data[1:], transpose([corr])))
         9         self.corr_data.set_data('corr', corr_data)
        10 
        11         Vm_data = self.Vm_data.get_data('Vm')
        12         Vm_data = hstack((Vm_data[1:], transpose([Vm])))
        13         self.Vm_data.set_data('Vm', Vm_data)
        14 
        15         self.webcam_plotdata.set_data('imagedata', im)
        16         self.corr_plot.request_redraw()
        17         return
      
    • plotting

         1 #============================================================================
         2 # Create the Chaco plot.
         3 #============================================================================
         4 
         5 def _create_plot_component(obj):
         6     # corr plot
         7     times = linspace(0., 1, num=NUM_SAMPLES)
         8     obj.corr_data = ArrayPlotData(time=times)
         9     empty_corr = zeros(NUM_SAMPLES)
        10     obj.corr_data.set_data('corr', empty_corr)
        11 
        12     obj.corr_plot = Plot(obj.corr_data)
        13     obj.corr_plot.plot(("time", "corr"), name="corr", color="red")
        14     obj.corr_plot.padding = 50
        15     obj.corr_plot.title = "Linear Correlation"
        16     corr_range = obj.corr_plot.plots.values()[0][0].value_mapper.range
        17     corr_range.low = -1.0
        18     corr_range.high = 1.0
        19     obj.corr_plot.index_axis.title = 'Time (seconds)'
        20     obj.corr_plot.value_axis.title = 'Correlation'
        21 
        22     # Vm plot
        23     obj.Vm_data = ArrayPlotData(time=times)
        24     empty_Vm = zeros(NUM_SAMPLES)
        25     obj.Vm_data.set_data('Vm', empty_Vm)
        26 
        27     obj.Vm_plot = Plot(obj.Vm_data)
        28     obj.Vm_plot.plot(("time", "Vm"), name="Time", color="blue")
        29     obj.Vm_plot.padding = 50
        30     obj.Vm_plot.title = "Neuron's potential"
        31     obj.Vm_plot.index_axis.title = 'Time (seconds)'
        32     obj.Vm_plot.value_axis.title = 'Amplitude'
        33     Vm_range = obj.Vm_plot.plots.values()[0][0].value_mapper.range
        34     Vm_range.low = 0.
        35     Vm_range.high = 1.
        36 
        37     # Webcam
        38     webcam_data = zeros((N_X, N_Y))
        39     obj.webcam_plotdata = ArrayPlotData()
        40     obj.webcam_plotdata.set_data('imagedata', webcam_data)
        41     webcam_plot = Plot(obj.webcam_plotdata)
        42     webcam_x = linspace(0.0, 1, num=N_X+1)
        43     webcam_y = linspace(0.0, 1, num=N_Y+1)
        44     webcam_plot.img_plot('imagedata',
        45                               name='webcam',
        46                               colormap=jet,
        47                               )
        48     range_obj = webcam_plot.plots['webcam'][0].value_mapper.range
        49     range_obj.high = 1.0
        50     range_obj.low = 0.0
        51     webcam_plot.title = 'webcam'
        52     obj.webcam_plot = webcam_plot
        53 
        54     container = HPlotContainer()
        55     container.add(obj.webcam_plot)
        56     container.add(obj.corr_plot)
        57     container.add(obj.Vm_plot)
        58 
        59     return container
      

    Include: Nothing found for "----"!

    Hooking together the pieces to build a (simplistic) neuroscience demo

    • During this talk, I wanted to make 2 points:
      • Science is fun: we need to expose interesting scientific results to students, kids and more generally to a wide audience. Here, I show how to build a simplistic demo. Few lines of code, relatively little effort to create a cross-platform, interactive demo.
      • It is still not wide-spread in the computational neuroscience community to design real-time simulations. is there some insight we may have from such an example?

    http://www.yorku.ca/eye/

    • In this demo project I hook together a neuron, a webcam and a loudspeaker interact in (approx) real-time. The goal is to conduct in computo what was done in vivo by Hubel & Wiesel. This was chosen since it is a well-known scientific that is fundamental in the sense that it links the response of a neuron to a stimulus in visual space (flashing a bar on a screen). It raises the question of what is represented by the neural activity: In the original experiment, ON and OFF subfields of the receptive field are directly marked on the screen by symbols. But how can we be sure that the spiking that we hear physically really corresponds to a neural representation, especially when this response is just one rumor in a whole, intricately connected recurrent network?

    • The principle is that when you launch the script, the input flow coming from your iSight gets converted through a dummy retina. The membrane of the neuron is directly excited by the instantaneous correlation coefficient between the RF and the present image, but modulated by a response curve "à la" Laughlin (1981).

    Hooking together the pieces to build a neuroscience demo

    Hooking together the pieces to build a (simplistic) neuroscience demo

    • During this talk, I wanted to make 2 points:
      • Science is fun: we need to expose interesting scientific results to students, kids and more generally to a wide audience. Here, I show how to build a simplistic demo. Few lines of code, relatively little effort to create a cross-platform, interactive demo.
      • It is still not wide-spread in the computational neuroscience community to design real-time simulations. is there some insight we may have from such an example?

    http://www.yorku.ca/eye/

    • In this demo project I hook together a neuron, a webcam and a loudspeaker interact in (approx) real-time. The goal is to conduct in computo what was done in vivo by Hubel & Wiesel. This was chosen since it is a well-known scientific that is fundamental in the sense that it links the response of a neuron to a stimulus in visual space (flashing a bar on a screen). It raises the question of what is represented by the neural activity: In the original experiment, ON and OFF subfields of the receptive field are directly marked on the screen by symbols. But how can we be sure that the spiking that we hear physically really corresponds to a neural representation, especially when this response is just one rumor in a whole, intricately connected recurrent network?

    • The principle is that when you launch the script, the input flow coming from your iSight gets converted through a dummy retina. The membrane of the neuron is directly excited by the instantaneous correlation coefficient between the RF and the present image, but modulated by a response curve "à la" Laughlin (1981).

    Computational Neuroscience: From Representations to Behavior

    Second NeuroComp Marseille Workshop

    Date
    27-28 May 2010
    Location

    Amphithéâtre Charve at the Saint-Charles' University campus - Métro : Line 1 et 2 (St Charles), a 5 minute walk from the railway station.
    Map (Amphithéâtre Charve, University Main Entrance, etc.)
    Metro, Bus and Tramway
    Getting to Marseille from Airport

    Registration

    Registration was free but mandatory, participation limited to 80 persons.

  • This workshop is organized by NeuroComp Marseille which is a local node of the NeuroComp group. First workshop was organized April 6-7 2009.

  • check out the following-up special issue rassembling some contributions to this workshop.

  • Computational Neuroscience emerges now as a major breakthrough in exploring cognitive functions. It brings together theoretical tools that elucidate fundamental mechanisms responsible for experimentally observed behaviour in the applied neurosciences. This is the second Computational Neuroscience Workshop organized by the "NeuroComp Marseille" network.

    It will focus on latest advances on the understanding of how information may be represented in neural activity (1st day) and on computational models of learning, decision-making and motor control (2nd day). The workshop will bring together leading researchers in these areas of theoretical neuroscience. The meeting will consist of invited speakers with sufficient time to discuss and share ideas and data. All conferences will be in English.

    Program

    27 May 2010 Neural representations for sensory information & the structure-function relation

    9h00-9h30

    Reception and coffee

    9h30-10h00

    Laurent Perrinet
    Institut de Neurosciences Cognitives de la Méditerranée, CNRS and Université de la Méditerranée - Marseille
    «Presentation of the Workshop and Topic»

    10h00-11h00

    Gabriel Peyré
    CNRS and Université Paris-Dauphine
    '''«Sparse Geometric Processing of Natural Images»'''
    In this talk, I will review recent works on the sparse representations of natural images. I will in particular focus on both the application of these emerging models to image processing problems, and their potential implication for the modeling of visual processing.
    Natural images exhibit a wide range of geometric regularities, such as curvilinear edges and oscillating textures. Adaptive image representations select bases from a dictionary of orthogonal or redundant frames that are parameterized by the geometry of the image. If the geometry is well estimated, the image is sparsely represented by only a few atoms in this dictionary.
    On an ingeniering level, these methods can be used to enhance the resolution of super-resolution inverse problems, and can also be used to perform texture synthesis. On a biological level, these mathematical representations share similarities with low level grouping processes that operate in areas V1 and V2 of the visual brain. We believe both processing and biological application of geometrical methods work hand in hand to design and analyze new cortical imaging methods.

    11h00-12h00

    Jean Petitot
    Centre d'Analyse et de Mathématique Sociales, Ecole des Hautes Etudes en Sciences Sociales - Paris
    «Neurogeometry of visual perception»

    In relation with experimental data, we propose a geometric model of the functional architecture of the primary visual cortex (V1) explaining contour integration. The aim is to better understand the type of geometry algorithms implemented by this functional architecture. The contact structure of the 1-jet space of the curves in the plane, with its generalization to the roto-translation group, symplectifications, and sub-Riemannian geometry, are all neurophysiologically realized by long-range horizontal connections. Virtual structures, such as illusory contours of the Kanizsa type, can then be explained by this model.

    12h00

    Lunch

    14h00-14h45

    Peggy Series
    Institute for Adaptive and Neural Computation, Edinburgh
    «Bayesian Priors in Perception and Decision Making»
    We'll present two recent projects:
    - The first project (with M. Chalk and A. R. Seitz) is an experimental investigation of the influence of expectations on the perception of simple stimuli. Using a simple task involving estimation and detection of motion random dots displays, we examined whether expectations can be developed quickly and implicitly and how they affect perception. We find that expectations lead to attractive biases such that stimuli appear as being more similar to the expected one than they really are, as well as visual hallucinations in the absence of a stimulus. We discuss our findings in terms of Bayesian Inference.
    - In the second project (with A. Kalra and Q. Huys), we explore the concepts of optimism and pessimism in decision making. Optimism is usually assessed using questionnaires, such as the LOT-R. Here, using a very simple behavioral task, we show that optimism can be described in terms of a prior on expected future rewards. We examine the correlation between the shape of this prior for individual subjects and their scores on questionnaires, as well as with other measures of personality traits.

    14h45-15h45

    Heiko Neumann(in collaboration with Florian Raudies)
    Inst. of Neural Information Processing, Ulm University Germany
    «Cortical mechanisms of transparent motion perception – a neural model»
    Transparent motion is perceived when multiple motions different in directions and/or speeds are presented in the same part of visual space. In perceptual experiments the conditions have been studied under which motion transparency occurs. An upper limit in the number of perceived transparent layers has been investigated psychophysically. Attentional signals can improve the perception of a single motion amongst several motions. While criteria for the occurrence of transparent motion have been identified only few potential neural mechanisms have been discussed so far to explain the conditions and mechanisms for segregating multiple motions.
    A neurodynamical model is presented which builds upon a previously developed neural architecture emphasizing the role of feedforward cascade processing and feedback from higher to earlier stages for selective feature enhancement and tuning. Results of computational experiments are consistent with findings from physiology and psychophysics. Finally, the model is demonstrated to cope with realistic data from computer vision benchmark databases.
    Work supported by European Union (project SEARISE), BMBF, and CELEST

    15h45-15h00

    Coffee break

    16h00-17h00

    /!\ CANCELED /!\ Rudolf Friedrich
    Institute für Theoretische Physik Westfälische Wilhelms Universität Münster
    «Windows to Complexity: Disentangling Trends and Fluctuations in Complex Systems»
    In the present talk, we discuss how to perform an analysis of experimental data of complex systems by disentangling the effects of dynamical noise (fluctuations) and deterministic dynamics (trends). We report on results obtained for various complex systems like turbulent fields, the motion of dissipative solitons in nonequilibrium systems, traffic flows, and biological data like human tremor data and brain signals. Special emphasis is put on methods to predict the occurrence of qualitative changes in systems far from equilibrium.
    [1] R. Friedrich, J. Peinke, M. Reza Rahimi Tabar: Importance of Fluctuations: Complexity in the View of stochastic Processes (in: Springer Encyclopedia on Complexity and System Science, (2009))

    17h00-17h45

    General Discussion

    28 May 2010 Computational models of learning and decision making

    9h30-10h00

    Andrea Brovelli
    Institut de Neurosciences Cognitives de la Méditerranée, CNRS and Université de la Méditerranée - Marseille
    «An introduction to Motor Learning, Decision-Making and Motor Control»

    10h00-11h00

    Emmanuel Daucé
    Mouvement & Perception, UMR 6152, Faculté des sciences du sport
    «Adapting the noise to the problem : a Policy-gradient approach of receptive fields formation»
    In machine learning, Kernel methods are give a consistent framework for applying the perceptron algorithm to non-linear problems. In reinforcement learning, the analog of the perceptron delta-rule is called the "policy-gradient" approch proposed by Williams in 1992 in the framework of stochastic neural networks. Despite its generality and straighforward applicability to continuous command problems, quite few developments of the method have been proposed since. Here we present an account of the use of a kernel transformation of the perception space for learning a motor command, in the case of eye orientation and multi-joint arm control. We show that such transformation allows the system to learn non-linear transformation, like the log-like resolution of a foveated retina, or the transformation from a cartesian perception space to a log-polar command, by shaping appropriate receptive fields from the perception to the command space. We also present a method for using multivariate correlated noise for learning high-DOF control problems, and propose some interpretations on the putative role of correlated noise for learning in biological systems.

    11h00-12h00

    Máté Lengyel
    Computational & Biological Learning Lab, Department of Engineering, University of Cambridge
    «Why remember? Episodic versus semantic memories for optimal decision making»
    Memories are only useful inasmuch as they allow us to act adaptively in the world. Previous studies on the use of memories for decision making have almost exclusively focussed on implicit rather than declarative memories, and even when they did address declarative memories they dealt only with semantic but not episodic memories. In fact, from a purely computational point of view, it seems wasteful to have memories that are episodic in nature: why should it be better to act on the basis of the recollection of single happenings (episodic memory), rather than the seemingly normative use of accumulated statistics from multiple events (semantic memory)? Using the framework of reinforcement learning, and Markov decision processes in particular, we analyze in depth the performance of episodic versus semantic memory-based control in a sequential decision task under risk and uncertainty in a class of simple environments. We show that episodic control should be useful in a range of cases characterized by complexity and inferential noise, and most particularly at the very early stages of learning, long before habitization (the use of implicit memories) has set in. We interpret data on the transfer of control from the hippocampus to the striatum in the light of this hypothesis.

    12h00-14h00

    Lunch

    14h00-15h00

    Rafal Bogacz
    Department of Computer Science, University of Bristol
    «Optimal decision making and reinforcement learning in the cortico-basal-ganglia circuit»
    During this talk I will present a computational model describing decision making process in the cortico-basal ganglia circuit. The model assumes that this circuit performs statistically optimal test that maximizes speed of decisions for any required accuracy. In the model, this circuit computes probabilities that considered alternatives are correct, according to Bayes’ theorem. This talk will show that the equation of Bayes’ theorem can be mapped onto the functional anatomy of a circuit involving the cortex, basal ganglia and thalamus. This theory provides many precise and counterintuitive experimental predictions, ranging from neurophysiology to behaviour. Some of these predictions have been already validated in existing data and others are a subject of ongoing experiments. During the talk I will also discuss the relationships between the above model and current theories of reinforcement learning in the cortico-basal-ganglia circuit.

    15h00-15h30

    Coffee break

    15h30-16h30

    Emmanuel Guigon
    Institut des Systèmes Intelligents et de Robotique, UPMC - CNRS / UMR 7222
    «Optimal feedback control as a principle for adaptive control of posture and movement»

    16h30-17h15

    General Discussion

    Sponsored by

    http://www.incm.cnrs-mrs.fr/ http://www.ism.univmed.fr/ http://sites.univ-provence.fr/ifrscc/ http://www.univmed.fr/ http://www.univ-provence.fr/ Pole 3c

    Diffraction monochromatique, spectre audiographique

    http://www.ondesparalleles.org/visuels/Etienne Rey DiffractionPL_Ph_C_Weiner.jpg

    • Diffraction est une sculpture en suspension composée d’une multitude de plaques de matière transparente et réfléchissante. L’installation met en jeu notre perception de l’espace par des phénomènes de résonance et de réflection de la lumière. Chaque lieu d’exposition donne à expérimenter et à élaborer, in situ, de nouvelles formes. A Seconde Nature, Etienne Rey abordera la relation entre le volume et le son en prenant comme base de construction un spectre audio, en collaboration avec l’artiste sonore Mathias Delplanque.

    • Live de Mathias Delplanque et rencontre autour de Diffraction, le Mercredi 14 avril 2010: A l’occasion de cette rencontre publique, quatre chercheurs spécialistes de l’architecture, de la perception, du son, et de la lumière exposeront depuis leurs domaines de recherches les processus engagés autour de Diffraction.`
      • Farid Ameziane, Ecole Nationale Supérieure d’Architecture de Marseille Luminy (EAML), Directeur de l’InsARTis, Marseille
      • Guillaume Bonello, Chargé de mission, POPsud, co/OAMP, Marseille
      • Fabrice Mortessagne, Directeur du laboratoire de Physique de la Matière Condensée (LPMC), Nice-Sophia Antipolis
      • Laurent Perrinet, Chercheur à l’Institut de Neurosciences Cognitives de Méditerranée, Equipe DyVA, Marseille
      • Modératrice : Colette Tron, Fondatrice d’Alphabetville, Marseille
    • Entrée libre & gratuite - 19h, durée 2h.

    • Renseignements pratiques :

      Espace Sextius investi par Seconde Nature  :
      27bis rue du 11 novembre, 
      13100 Aix-en-Provence
    • (!) visitez le site de Seconde Nature

    notes de l'intervention de Laurent Perrinet

    • Qu'est-ce que voir? En perception, les neurones « parlent » tous en même temps par de brèves impulsions électrochimiques, générant un mélange de signaux, un bruit. Pourtant c'est par eux que nous pensons, voyons, sentons. Les ordinateurs sont différents, plus rapides. Ils sont construits avec pour modèle la grammaire humaine autour d’une unité centrale, car on imaginait la cognition sous cet angle à leur invention. Le bit est le quantum d’un algorithme mécanique (thèse de Church-Turing). Une théorie tranche par rapport à la précédente, proposée par «von Neumann» : beaucoup d’unités sont présentes dans le cerveau. Comparée à la chaîne logique du langage, dans cet algorithme, beaucoup d’autres chaînes et logiques se mêlent. Comment vont-elles « parler » entre elles ? Existe-t-il des algorithmes biologiques ? Ouchi.jpeg

    Définir ce « langage », c'est comprendre comment une somme d’informations locales peut produire une perception globale. Comment en jouant avec les atomes du code, en les superposant, les « cassant » pour les mettre en résonance, les neurosciences et l'artiste questionnent le langage de notre pensée ? Quel est le code utilisé par les neurones pour communiquer (code neuronal ? existe-t-il un même vocabulaire au sens homomorphique ?). En pratique, on apprend par exemple la sélectivité à l'orientation. Les phénomènes d’orientation sont radicaux à la fin de l’expérience, « gelant » son évolution. Un lien évident avec l’installation Phytosphère d’Etienne Rey.

    L’information dans le cerveau se propage par diffusion, par diffraction (contamination des informations entre neurones pour occuper l’espace), en lien avec le travail sur la lumière d’Etienne Rey. L'image a besoin de 30 millisecondes pour se diffuser de l’œil vers l’arrière du crâne et 85 millisecondes pour produire un réflexe oculaire. Les neurosciences cherchent à savoir comment comprendre la globalité par l'émergence.

    Il y a donc une superposition d’états, comme dans la diffraction d’Etienne Rey.

    http://www.voss-web.com/DATAS/GALLERY_04.jpg En perception, le mécanisme neuronal cherche à sortir de l’ambiguïté première quand il connaît une image : il superpose des particules élémentaires d'information, les diffuse pour les prendre toutes. Ce qui émerge est non linéaire. Le cerveau interfère ces particules, donc les met en compétition, en coopération (voir expérience plus haut avec les neurones rouges et bleus), dans une dynamique où ces particules se réorientent elles-mêmes. Elles créent des phénomènes d’organisation, se collent, deviennent plus lumineuses. La perception n’est donc pas séquentielle mais fluide et la sortie de l'ambiguité depuis l'image pixel vient de l'introduction de ces contraintes. Ainsi quand nous voyons un objet, nous le « capturons ». Quand nous sommes vus, nous cherchons à nous séparer de cette capture.

    Un problème classique est l'ambiguité du monde sensible. Une couleur que l’on ne voit pas va apparaître visuellement. L’inpainting créé une œuvre qui correspond à un mécanisme neuronal, cherchant à reproduire toujours une même structure. La mémoire iconique du monde extérieur va imprégner le cerveau, s’y figer. Tout le problème de la perception pour les neurosciences repose sur deux dialectiques. La première présente une analogie avec les images informatiques par pixels : ce serait en neurosciences une métaphore de la sensation pure. La seconde rappelle l’image vectorisée : pour s’extraire de la sensation pure, le cerveau retiendra des règles proches des algorithmes. En cognition, il permet de mettre en lumière le symptome d'autisme. Dans un schéma montrant un bloc derrière un arbre, dépassant des deux côtés, sera découpé visuellement par l’autiste en plusieurs morceaux distincts. Il ne généralise pas l’information.

    diffractionFriche_0134.jpg Comment être sûr d’une perception globale en désignant les modules de l’installation d’Etienne Rey, ou signifiants des atomes, dans ce passage du local au global ? Les modules ne se voient pas forcément dans l’installation, mais d’autres aspects sont perçus. La relation à l’atome, même si elle n’est pas signifiante pour le public, n’est pas primordiale. Le public voit une accumulation de « choses », car par principe quand un phénomène est concentré « il se passe des choses » par jeu de contraste. Le fait de bouger face à l’installation rend unique à l'individu la perception et réalise la globalité de l’œuvre: on a alors passage de l’atome à la forme globale. Cette résolution rejoint Giotto et les débuts de la perspective en art pictural. Il a révélé la question du point de vue, par positionnement et déplacement. En effet, les personnes penchent la tête dans l’installation spirale en container, d’Etienne Rey, pour le festival Ozosphère à Strasbourg. Ce phénomène est à rattaché aux théories sur la perception.

    Biographie Laurent Perrinet, chercheur à l’Institut de Neurosciences Cognitives de la Méditerranée à Marseille, unité mixte du CNRS, aime citer « La vie de Brian » des Monty Python : (Brian:) "You have to work it out for yourselves!" (Crowd:) "Yes, we have to work it out for ourselves... (silence) Tell us more!". L’individualité et la perception du monde… Dans l’équipe DyVA (pour Dynamique de la perception visuelle et de l'action), Laurent Perrinet s'intéresse aux neurones impulsionnels et au codage neuronal, ainsi qu’à la perception des mouvements spatio-temporels. Ces processus définis comme des algorithmes, la représentation du flux vidéo modélise via l’informatique ces interactions au niveau cellulaire (colonnes corticales) et au niveau cognitif (aires corticales). Il cherche à comprendre le fonctionnement des calculs corticaux dans le système visuel. Cette recherche fournit des réponses aux problèmes cognitifs. Après un diplôme d'ingénieur de traitement du signal et de modélisation stochastique de l'école d’aéronautique Supaéro à Toulouse et des études à San Diego et à Pasadena (Californie) pour la Nasa, Laurent Perrinet obtient un doctorat de Sciences Cognitives. Répondant aux questions « Peut-on parler d’intelligence mécanique ? », « Pourquoi une grenouille gobe mieux une mouche qu’un robot ? » ou « Quelle est la différence entre intelligence et algorithme ? », il intervient en 2009 au colloque marseillais « Les chemins de l’intelligence ». Parmi ses publications : Role of homeostasis in learning sparse representations, et sa thèse Comment déchiffrer le code impulsionnel de la vision ? Étude du flux parallèle, asynchrone et épars dans le traitement visuel ultra-rapide

    FACETS plenary meeting, WP 5/9: Modeling and Databases at the Network Level - Friday, January 8th 2010

    reference

    • Laurent Perrinet, Guillaume S. Masson. Models of low-level vision: linking probabilistic models and neural masses, URL . 2010 abstract.

    Reading out the dynamics of lateral interactions in the primary visual cortex from VSD data

    Embedded application/pdf

    reference

    • Laurent U. Perrinet, Alexandre Reynaud, Frédéric Chavane, Guillaume S. Masson. Reading out the dynamics of lateral interactions in the primary visual cortex from VSD data, URL . In Macroscopic aspects of neuronal activity: ''Macroscopic models, LFP models and VSD models'' a FACETS workshop in Marseille, Nov. 30th /Dec. 1st, 2009 abstract.

    Peut-on parler d’intelligence mécanique?

    http://www.cerco.ups-tlse.fr/fr_vers/images/objet_espace_image1g.jpg

    • Intervenants: Laurent Perrinet et Thierry Viéville (Laurent Perrinet est Chercheur Permanent à l'Institut de Neurosciences Cognitives de la Méditerranée et Thierry Vieville est Directeur de Recherche à l'Institut National de Recherche en Informatique et en Automatique (INRIA), Equipe Projet Cortex.)
    • Nous parlerons de cette partie "mécanique" du cerveau animal ou humain qui permet de percevoir les mouvements et de ... survivre au sein de l'environnement. On verra, par exemple, que notre cerveau peut-être plus rapide que nous, qu'il y a des solutions "stupides" qui marchent remarquablement bien pour sortir d'un labyrinthe, et qui si la grenouille sait gober une mouche bien mieux qu'un robot ... elle n'est pas plus maligne ! Parce que ce qu'il ne faut pas confondre ici c'est la différence entre calculer et penser, entre intelligence et algorithmes. En comprenant cela, avec Alan Mathison Turing, le Gutenberg du XXème siècle, l'humanité a basculé des temps modernes à l'ère du numérique.

    • Cycle de conférences organisé par l’Association Science Technologie Société - PACA ayant pour thème cette année : "Biologie et civilisation : les chemins de l’intelligence".

      • 20/10/09 : État des lieux en neurobiologie
      • 24/11/09 : Peut-on parler d’intelligence mécanique ?
      • 15/12/09 : Naissance d’une civilisation…
      • 26/01/10 : Intelligence collective des insectes sociaux
      • 23/02/10 : Inné et acquis dans la construction du cerveau
      • 23/03/10 : Peut-on gouverner et prévoir par la science ?
      • 24/04/10 : Vers une cartographie des troubles du comportement ?
      • 25/05/10 : Histoire des neurosciences
    • Entrée libre & gratuite - 18h 30, durée 1h.

    • Renseignements pratiques :

      Espace Ecureuil, 26 rue Montgrand, 13006 Marseille
      Tél : 04 91 57 26 49 / 06 84 43 68 45
      mathieu.orban@asts.asso.fr 
    • (!) visitez le site d'interstices!

    "Brain dynamics, from information to behavior" workshop

    reference

    • Laurent Perrinet, Guillaume S. Masson. Decoding low-level neural information to track visual motion, URL . 2009 abstract.

    WORKSHOP: "Brain dynamics, from information to behavior"

    April 6th and 7th, 2009 @ Medecine Faculty «la Timône»

    This workshop is organized in order to promote theoretical and computational neuroscience at the future Neuroscience Institute @ la Timône.

    Monday, April 6th

    Session 1

    Modeling and Analysis of Behavior and Function Chair: Guillaume Masson

    9 :00-9 :40

    Raoul Huys: A phase flow perspective to motor control and timing mechanisms

    9 :40-10 :20

    Anna Montagnini: Dynamic inference for smooth pursuit eye movements

    10 :20-11 :00

    BREAK

    11 :00-11 :40

    Laurent Pezard: Symbolic dynamics of behavioral sequences

    11:40-12 :20

    Laurent Perrinet: Decoding low-level neural information to track visual motion

    12 :20-14 :00

    LUNCH

    Session 2

    Dynamics of neural and brain networks - Chair: Bruno Cessac

    14 :00-14 :40

    Viktor Jirsa: What does the brain do when it does nothing?

    14:40- 15 :20

    Alain Destexhe: Modeling network activity states based on conductance measurements

    15 :20-16 :00

    BREAK

    16 :00-16 :40

    Alexa Riehle: Modifications in motor cortical dynamics induced by practice

    16:40- 17 :20

    Jens Starke: Analysis of spatio-temporal pattern formation in the olfactory bulb

    Tuesday, April 7th

    Session 3

    Epilepsy - Chair: Patrick Chauvel

    9 :00-9 :40

    Christophe Bernard: Alterations in brain dynamics and behavior in epilepsy

    9 :40-10 :20

    John Terry: Applications of mean-field modelling of macroscale brain activity: Anticipation, Evolution and Suppression of Seizures

    10 :20-11 :00

    BREAK

    11 :00-11 :40

    Louis Lemieux: The study of epileptogenic networks using multimodal imaging in the resting state

    11:40-12 :20

    Fabrice Wendling: Lumped-parameter and detailed models of epileptogenic networks: insights into the interpretation of depth-EEG recordings in TLE

    12 :20-14 :00

    LUNCH

    Session 4

    Brain dynamics from brain imaging - Chair: Bruno Poucet

    14 :00-14 :40

    Stefan Kiebel: Recognizing sequences of sequences

    14:40- 15 :20

    Jean-Luc Blanc: Entropy estimation of symbolic sequences: application to synchronization detection

    15 :20-16 :00

    BREAK

    16 :00-16 :40

    Andrea Brovelli: The neural computations of instrumental learning

    16:40- 17 :20

    Petra Ritter: Spatiotemporal dynamics of spontaneous and evoked EEG-fMRI signals

    Deuxième conférence française de Neurosciences Computationnelles, "Neurocomp08"

    La deuxième conférence française de Neurosciences Computationnelles, "Neurocomp08", s'est déroulée à la Faculté de Médecine de Marseille du 8 au 11 octobre 2008. Cette conférence, organisée par le groupe de travail Neurocomp, a permis de réunir les principaux acteurs français du domaine (francophones ou non). Le champ des Neurosciences Computationnelles porte sur l'étude des mécanismes de calcul qui sont à l'origine de nos capacités cognitives. Cette approche nécessite l'intégration constructive de nombreux domaines disciplinaires, du neurone au comportement, des sciences du vivant à la modélisation numérique. Avec ce colloque, nous avons offert un lieu d'échanges afin de favoriser des collaborations interdisciplinaires entre des équipes relevant des neurosciences, des sciences de l'information, de la physique statistique, de la robotique. Cette édition a également été l'occasion d'ouvrir le cadre à de nouveaux domaines (modèles pour l'imagerie, interfaces cerveau-machine,...) notamment grâce à des ateliers thématiques (une nouveauté dans cette édition). Certains des principaux enjeux du domaine ont été présentés par quatre conférenciers invités : Ad Aertsen (Freiburg, Allemagne), Gustavo Deco (Barcelone, Espagne), Gregor Schöner (Bochum, Allemagne), Andrew B. Schwartz (Pittsburgh, USA). Des interventions orale plus courtes et plus spécifiques étaient également au programme, sur la base d'une sélection du comité de lecture. Une cinquantaine de posters ont également été présentés au cours de ces journées. Le premier jour était consacré aux modèles de la cellule neurale, aux modèles des traitements visuels et corticaux, ainsi qu'aux modèles de réseaux de neurones bio-mimétiques. La seconde journée était consacrée aux interfaces cerveau-machine, à la dynamique des grands ensembles de neurones, à la plasticité fonctionnelle et aux interfaces neurales. Enfin, la journée de samedi était consacrée à des ateliers thématiques, l'un sur les interfaces cerveau-machine, l'autre sur la vision computationnnelle. Cette conférence a connu un beau succès de par l'affluence (200 personnes environ) et la qualité des interventions. Ce succès tient également au fort soutien financier et organisationnel qu'elle a obtenu de ses partenaires. Les organisateurs remercient le CNRS, la Société des neurosciences, le conseil régional de la région Provence Alpes Côte d'Azur, le conseil général des Bouches de Rhône, la mairie de Marseille, l'université de Provence, l'IFR "Sciences du cerveau et de la cognition", l'INRIA, ainsi que la faculté de médecine de Marseille et l'université de la Méditerranée qui ont hébergé la conférence.

    Plus d'informations

    @ Universität Ulm - Heiko Neumann's lab

    • Laurent Perrinet, Guillaume S. Masson. Decoding the population dynamics underlying ocular following responseusing a probabilistic framework, URL . 2008 abstract.

    @ INCM (Marseille) April 11th, 2008

    • Laurent Perrinet. From neural activity to behavior: computational neuroscience as a synthetic approach for understanding the neural code., URL . 2008 abstract.

    Prisma workshop, Toledo (Spain), February 7, 2008

    • Laurent Perrinet. Modeling of spikes, sparseness and adaptation in the primary visual cortex: applications to imaging, URL . In Prisma workshop, Toledo (Spain), February 7, 2008, 2008 abstract.

    The Rank Prize Funds, Mini-Symposium on Representations of the Visual World in the Brain, 2007

    • Laurent Perrinet. What efficient code for adaptive spiking representations?, URL . In The Rank Prize Funds, Mini-Symposium on Representations of the Visual World in the Brain, 2007 abstract.

    Mathematical image processing meeting (Marseille, France) September 5, 2007

    • Laurent Perrinet. Neural Codes for Adaptive Sparse Representations of Natural Images, URL . In Mathematical image processing meeting (Marseille, France) September 5, 2007 abstract.
    welcome: please sign in