neurosciences?

Mon projet scientifique s'intéresse aux mécanismes computationnels qui sous-tendent la cognition. C'est-à-dire que l'on sait où se produisent ces mécanismes définissant la système nerveux central en un réseau de neurones connectés par des synapses et qu'ils sont supportés par des signaux électro-chimiques entre ces noeuds, mais on ne connaît pas encore totalement comment l'information qui semble être portée ces signaux peut être interprété. Ce décodage, qui est le fond de notre travail en neurosciences, à un "Graal" qui est la découverte d'un hypothétique "code neural", c'est-à-dire du langage qui est utilisé dans notre cerveau. On ne sait si cette découverte est possible; la question se pose: peut-il exister une connaissance globale du cerveau à la manière d'autres disciplines scientifiques (par exemple, la trajectoire d'une planète avec les lois de Newton?). Il est clair que le cerveau de chaque humain n'est pas assez complexe pour en délimiter la complexité, même les cerveaux mis en réseau avec toutes les communautés neuro-scientifiques, artistiques vont nous permettre dans le futur de mieux comprendre cet objet...

Nous sommes encore au Moyen-Âge d'un compréhension globale de la cognition. Il n'y a pas de brique élémentaire ou de principe universel comme cela peut l'être dans d'autres disciplines comme la mécanique, la chimie ou la logique mathématique classique. Nous sommes ici dans le domaine des sciences du complexe: on parle alors de concepts encore très jeunes par rapport à l'âge de l'humanité comme l'utilisation de mesures d'information, l'auto-organisation ou l'émergence.

des axes

Il y a de nombreuses perspectives pour le découvrir progressivement et je suis particulièrement interessé par ces axes:

• La découverte d'algorithmes neuraux permet de construire de nouveaux paradigmes de calcul. En effet une chose que l'on sait du cerveau est qu'il n'est pas un ordinateur! au moins il n'est pas un ordinateur classique (de von Neumann) où toute l'information passe par un (ou un petit nombre) de processeurs à très grande vitesse. À le place de cela, l'architecture du système nerveux est massivement parallèle, asynchrone et adaptative. Ces nouveaux algorithmes pourront être implantées sur des puces de nouvelle génération qui sont actuellement en développement.

• Une meilleure connaissance des mécanismes ouvre bien sûr la voie à de nombreuses applications thérapeutiques et sur un large spectre depuis le contrôle des épilepsie jusqu'à la compréhension des dégénérescences neurales. Nous appliquons au laboratoire cette démarche scientifique en nous concentrant sur les bases de la vision et en particulier sur la capacité à détecter le mouvement.

• On le voit notre démarche scientifique est relativement large est si elle est appliquée à un cas particulier (la détection de mouvement), nous faisons en sorte qu'elle puisse toujours être approchée de façon générique à d'autres problèmes: d'autres modalités sensorielles ou cognitives mais surtout d'autres échelles d'analyses, du très petit (l'interaction de sous-parties d'un neurone) au très grand (interactions sociales).

Voilà pour une brève présentation.

Tropique

Il y a donc une grande proximité du champ d'action avec la démarche artistique d'Etienne Rey qui a conduit à l'émergence de ce projet. J'étais au début surpris de l'utilisation de mots clés (diffraction, particule, résonance, émergence, ...) et pensais qu'il étaient plus utilisés pour le pouvoir poétique de leur évocation. En fait, au cours des discussions nous nous sommes rendus compte que nous parlions le même langage et qu'une voie s'ouvre si nous confrontions nos perspectives en redéfinissant ce qui ne l'est pas encore précisément: c'est l'intérêt de Tropique en tant que chercheur en neuroscience: un espace de création dans la mise en oeuvre du projet dans la définition du "cerveau artificiel" qui va le contrôler, un espace de création imprévisible qui va naître de l'interaction avec le public.

• la phase de gestion de l'information du mouvement de plusieurs acteurs est une prouesse technologique qui sera une épreuve du feu pour les algorithmes neuro-mimétiques que nous développons. En particulier, le concept de particule élémentaire d'information de mouvement pourra montrer son utilité à un niveau pratique,

• explorer en pratique la résonance entre Perception et Action. Ces deux facettes de la cognition qui sont gravées dans l'anatomie du cerveau sont indissociables. Instinct Paradise donne un espace d'expérimentation qui nous permet de manipuler directement la perception d'espace d'un personne (son "aura") ainsi que ses interactions. A la manière d'une fractale nous envisageons de transposer ce niveau d'interactions sociales inter-personnes (10mx10m) sur un modèles d'interactions neurales (1cmx1cm) sur des règles similaires élémentaires de diffusion/agrégation, on envisage en particulier utiliser les données enregstrées pour les interpréter / voir les différences entre lieux, temps, configurations

• c'est une aventure humaine, une série d'échanges, un projet que nous voulons donner à partager. À mon niveau, c'est aussi pour la reconnaissance qu'il soit porté par les institutions . À l'heure ou le seul espace public pour la science sont le mysticisme de jumeaux lipo-chirurgés ou le scpeticisme industriuex d'un ex-minstre géologiquement mamouthé, c'est un réel bonheur qu'on puisse monter un projet qui me permette de présenter quelques avancées sur notre connaissance du cerveau. Finalement, mon intérêt est aussi de pouvoir partager une bière au bar de la Friche pour discuter à bâtons rompus de concepts métaphysiques puis de plonger sur un détail très spécifique de la construction d'un détecteur ou d'imaginer les scénarios possibles d'interaction.

the poll

• On april 1st, 2011, I sent the following message to the comp-neuro and connectionists lists: (see http://www.neuroinf.org/pipermail/comp-neuro/2011-April/002613.html

• Dear list
  
  A recent paper in PLoS Computational Biology
  
  > The Roots of Bioinformatics in Theoretical Biology
  > Paulien Hogeweg
  > Volume 7(3) March 2011 http://www.ploscompbiol.org/article/info%3Adoi%2F10.1371%2Fjournal.pcbi.1002021
  
  makes a point in the evolution of the meaning of the field of bioinformatics with the advent of data-driven modeling.
  
  The same seems to have appeared in computational neuroscience. The sense slowly drifted from the original papers (such as Science, Vol. 241, No. 4871, 1988, pp. 1299-1306. by T. J. Sejnowski, C. Koch, P. S. Churchland) which I believed is perfectly captured in the sentence: "The ultimate aim of computational neuroscience is to explain how electrical and chemical signals are used in the brain to represent and process information."  (this does not exclude using computers of course).
  
  It seems to be solely a semantical problem, but this may generate some confusion (realpolitik translation: "and this may hinder the efficiency of your grant proposal"). Recently an (anonymous) colleague told me they called their group "computational AND theoretical neuroscience" (just as these two fields where separated) out of the lack of consensus on the meaning of words and to not exclude anyone. Nowadays, even in the university, there is a continuum of fields combining biology,  mathematics or computer science and all computational neuroscientists reflect this as individuals. so what's the situation in 2011?
  
  I often asked to fellow colleagues this question, "what is computational neuroscience?" and often got one of these answers (I try to be unbiased - please correct me):
  
   [ ] it is a field of neuroscience involving the use of computers (von Neumann machines, Dell boxes, macbooks, ...) to simulate and analyze data obtained from experimental neuroscience and advance our knowledge from this dialogue. Theoretical neuroscience is different in the sense that it proposes mathematical models of how it works.
  
   [ ] it is the field of neuroscience studying how information is represented and processed in neural activity. This involves a dialogue with experimental neuroscience to analyze and propose experiments. It proposes models, that is theories for the relation between function and structure. Theoretical neuroscience is a subset of computational neuroscience that tries to express these models in standard mathematical language.
  
   [ ] it is some field of neuroscience and why would you care to give an exact definition? its frontiers are moving and it has many facets, theoretical neuroscience being just one example. it cares about being less ignorant on relation between function and structure in neuroscience.
  
  If you want to express your opinion, you can so in one click:
  https://spreadsheets.google.com/viewform?formkey=dDc5X2dJRS1zMHRiSndSNERWelBkQlE6MQ
  results :
  https://spreadsheets.google.com/lv?key=0AueMPskll6yrdDc5X2dJRS1zMHRiSndSNERWelBkQlE&hl=fr&f=0&rm=full#gid=0
  
  cheers,
  Laurent

results

• I have received 36 responses with the following results (see https://spreadsheets.google.com/lv?key=0AueMPskll6yrdDc5X2dJRS1zMHRiSndSNERWelBkQlE&hl=fr&f=0&rm=full#gid=0 ):

  1. [ 5 ] it is a field of neuroscience involving the use of computers (von Neumann machines, Dell boxes, macbooks, ...) to simulate and analyze data obtained from experimental neuroscience and advance our knowledge from this dialogue. Theoretical neuroscience is different in the sense that it proposes mathematical models of how it works.

  2. [ 21 ] it is the field of neuroscience studying how information is represented and processed in neural activity. This involves a dialogue with experimental neuroscience to analyze and propose experiments. It proposes models, that is theories for the relation between function and structure. Theoretical neuroscience is a subset of computational neuroscience that tries to express these models in standard mathematical language.

     • one answer is amended by "is a sub-field of theoretical neuroscience that seeks to understand how information is represented and processed in the nervous system by implementing and testing theories in the form of computer simulations."

     • another answer comments "Not a definition, but a comment. Since Comp. Neurosci. (or whatever it should be called -- some people now call it Neurodynamics) is already such a small field and barely represented on the map, I think it is foolish to further subdivide it. That is why I like the first definition a bit more, such that both are collected under one roof."

  3. [ 6 ] it is some field of neuroscience and why would you care to give an exact definition? its frontiers are moving and it has many facets, theoretical neuroscience being just one example. it cares about being less ignorant on relation between function and structure in neuroscience.

  4. [ 4 ] other free-form answers were given:

     1. with a slight modification to the first definition

        I agree with definition (1) except the "Theoretical neuroscience is a subset of ..." as I might argue that "Computational neuroscience is a subset of theoretical neuroscience".

     2. with a new definition close to the aims of theoretical neuroscience

        it is field of neuroscience that use mathematical models to analyze the data obtained from experimental neuroscience. Therefore, it gives a logical result to it and it can be explained instead to be as a magic box

     3. to a strict "computational" view

        It is the subset of theoretical neuroscience that hypothesises that the brain is a computer. This relates to the first definition to the extent that 'computation' is identified with 'information processing'. Theoretical neuroscience is simply the development of models (in any form, including mathematics or computer simulations) of neural processes. It is possible for a process to be simulated or analyzed using a computer - see the second definition - without claiming the process itself is an example of computation. 'Computational neuroscience' usually implies this stronger claim, though it is now often used more loosely (definition three).

     4. or to the interesting view that these different approaches overlap but correspond to different approaches:

        In my view, theoretical neuroscience is the non-experimenting version of neuroscience, much like theoretical physics is the non-experimenting version of physics.
        
        I would argue that theoretical neuroscience and computational neuroscience are different in their approaches.
        
        Computational neuroscience has a strong focus on simulation. It is the "virtual" extension of electrophysiology. The modeling philosophies of GENESIS and Neuron clearly reflect this. So called "biologically realistic" simulations are the gold standard in computational neuroscience.
        
        Theoretical neuroscience, by contrast, has its focus on mathematical descriptions and properties of nervous structures. Theoretical neuroscience starts, when the experiments, real or simulated, are done. The excellent books of Henry Tuckwell illustrate this. Here, simulation is not the method of choice, but the last resort after all pencils are broken and all paper is used up ;-)
        
        At best, I would say that there is overlap between the two rather than that one comprises the other.
        
        And while we are at definitions. Why not add the re-incarnated term of "Neuroinformatics" to the contest?

analysis

• While there is a strong majority of participants that express an agreement with the "standard", original definition of computational neuroscience from the original papers, discussions raised by this poll show that there is no consensus. A good sign of health for the community!

• some more pointers: http://en.wikipedia.org/wiki/Computational_neuroscience - http://www.citeulike.org/tag/computational_neuroscience

using SubVersion (SVN)

Version Control is an everyday tool to handle your important source files. It is useful:

• to grab the latest source code from open source projects and to always keep up-to-date,

• to share a bunch of files (a set of latex source code, python scripts, ...) allowing to work on different computers with different persons,

• to keep track of revisions from your project.

an alternative : Git

• The place to begin.

• From the SVN world: http://git-scm.com/course/svn.html

• Everyday's git.

• to set-up

  git config --global user.name "Your Name Comes Here"
  git config --global user.email you@yourdomain.example.com
  git config --global color.diff auto
  git config --global color.status auto
  git config --global color.branch auto

using Git with SVN

• install git-svn and use

  git svn fetch

articles

1. Publications/Barthelemy07

2. Publications/Cessac07

3. Publications/Daucé10

4. Publications/Davison08

5. Publications/Fischer07

6. Publications/Fischer07cv

7. Publications/Kremkow10jcns

8. Publications/Montagnini07

9. Publications/Perrinet06

10. Publications/Perrinet07neurocomp

11. Publications/Perrinet10shl

12. Publications/Voges10neurocomp

13. SciBlog/2011-07-05

references of full articles

• Laurent Perrinet. Dynamical Neural Networks: modeling low-level vision at short latencies, URL . pages 163--225.

• Frédéric Barthélemy, Laurent Perrinet, Éric Castet, Guillaume S. Masson. Dynamics of distributed 1D and 2D motion representations for short-latency ocular following, URL URL2 URL3 . Vision Research, 48(4):501--22, 2008

• Anna Montagnini, Pascal Mamassian, Laurent U. Perrinet, Eric Castet, Guillaume S. Masson. Bayesian modeling of dynamic motion integration, URL . Journal of Physiology (Paris), 101(1-3):64-77, 2007

  The quality of the representation of an object's motion is limited by the noise in the sensory input as well as by an intrinsic ambiguity due to the spatial limi- tation of the visual motion analyzers (aperture prob- lem). Perceptual and oculomotor data demonstrate that motion processing of extended ob jects is initially dominated by the local 1D motion cues orthogonal to the ob ject's edges, whereas 2D information takes pro- gressively over and leads to the final correct represen- tation of global motion. A Bayesian framework ac- counting for the sensory noise and general expectancies for ob ject velocities has proven successful in explaining several experimental findings concerning early motion processing [1, 2, 3]. However, a complete functional model, encompassing the dynamical evolution of ob- ject motion perception is still lacking. Here we outline several experimental observations concerning human smooth pursuit of moving ob jects and more particu- larly the time course of its initiation phase. In addi- tion, we propose a recursive extension of the Bayesian model, motivated and constrained by our oculomotor data, to describe the dynamical integration of 1D and 2D motion information.

• Laurent Perrinet, Guillaume S. Masson. Modeling spatial integration in the ocular following response using a probabilistic framework, URL . Journal of Physiology (Paris), 2007

  The machinery behind the visual perception of motion and the subsequent sensori-motor transformation, such as in Ocular Following Response (OFR), is confronted to uncertainties which are efficiently resolved in the primate's visual system. We may understand this response as an ideal observer in a probabilis- tic framework by using Bayesian theory (Weiss et al., 2002) which we previously proved to be successfully adapted to model the OFR for different levels of noise with full field gratings (Perrinet et al., 2005). More recent experiments of OFR have used disk gratings and bipartite stimuli which are optimized to study the dy- namics of center-surround integration. We quantified two main characteristics of the spatial integration of motion : (i) a finite optimal stimulus size for driving OFR, surrounded by an antagonistic modulation and (ii) a direction selective sup- pressive effect of the surround on the contrast gain control of the central stim- uli (Barth\'elemy et al., 2006). Herein, we extended the ideal observer model to simulate the spatial integration of the different local motion cues within a proba- bilistic representation. We present analytical results which show that the hypoth- esis of independence of local measures can describe the integration of the spatial motion signal. Within this framework, we successfully accounted for the con- trast gain control mechanisms observed in the behavioral data for center-surround stimuli. However, another inhibitory mechanism had to be added to account for suppressive effects of the surround.

• B. Cessac, E. Daucé, Laurent U. Perrinet, M. Samuelides. Topics in Dynamical Neural Networks: From Large Scale Neural Networks to Motor Control and Vision, `URL <https://laurentperrinet.github.io/publication/cessac-07>`__ `URL2 <http://www.springerlink.com/content/q00921n9886h/?p=03c19c7c204d4fa78b850f88b97da2f7π=0>`__ . Springer Berlin / Heidelberg, 2007.

• Andrew P Davison, Daniel Bruderle, Jochen Eppler, Jens Kremkow, Eilif Muller, Dejan Pecevski, Laurent Perrinet, Pierre Yger. PyNN: A Common Interface for Neuronal Network Simulators., URL . Frontiers in Neuroinformatics, 2:11, 2008

  Computational neuroscience has produced a diversity of software for simulations of networks of spiking neurons, with both negative and positive consequences. On the one hand, each simulator uses its own programming or configuration language, leading to considerable difficulty in porting models from one simulator to another. This impedes communication between investigators and makes it harder to reproduce and build on the work of others. On the other hand, simulation results can be cross-checked between different simulators, giving greater confidence in their correctness, and each simulator has different optimizations, so the most appropriate simulator can be chosen for a given modelling task. A common programming interface to multiple simulators would reduce or eliminate the problems of simulator diversity while retaining the benefits. PyNN is such an interface, making it possible to write a simulation script once, using the Python programming language, and run it without modification on any supported simulator (currently NEURON, NEST, PCSIM, Brian and the Heidelberg VLSI neuromorphic hardware). PyNN increases the productivity of neuronal network modelling by providing high-level abstraction, by promoting code sharing and reuse, and by providing a foundation for simulator-agnostic analysis, visualization and data-management tools. PyNN increases the reliability of modelling studies by making it much easier to check results on multiple simulators. PyNN is open-source software and is available from http://neuralensemble.org/PyNN.

• Sylvain Fischer, Rafael Redondo, Laurent Perrinet, Gabriel Crist\'obal. Sparse approximation of images inspired from the functional architecture of the primary visual areas, URL . EURASIP Journal on Advances in Signal Processing, special issue on Image Perception, :Article ID 90727, 16 pages, 2007

• Sylvain Fischer, Filip Sroubek, Laurent U. Perrinet, Rafael Redondo, Gabriel Crist\'obal. Self-invertible 2D log-Gabor wavelets, URL . Int. Journal of Computional Vision, 2007

• Nicole Voges, Laurent Perrinet. Phase space analysis of networks based on biologically realistic parameters, URL . Journal of Physiology (Paris), 104(1-2):51--60, 2010

  We study cortical network dynamics for a more realistic network model. It represents, in terms of spatial scale, a large piece of cortex allowing for long-range connections, resulting in a rather sparse connectivity. We use two different types of conductance-based I&F neurons as excitatory and in- hibitory units, as well as specific connection probabilities. In order to re- main computationally tractable, we reduce neuron density, modelling part of the missing internal input via external poissonian spike trains. Compared to previous studies, we observe significant changes in the dynamical phase space: Altered activity patterns require another regularity measure than the coefficient of variation. We identify two types of mixed states, where differ- ent phases coexist in certain regions of the phase space. More notably, our boundary between high and low activity states depends predominantly on the relation between excitatory and inhibitory synaptic strength instead of the input rate.Key words: Artificial neural networks, Data analysis, Simulation, Spiking neurons. This work is supported by EC IP project FP6-015879 (FACETS).

• Emmanuel Daucé, Laurent Perrinet. Computational Neuroscience, from Multiple Levels to Multi-level, URL . Journal of Physiology (Paris), 104(1--2):1--4, 2010

  Despite the long and fruitful history of neuroscience, a global, multi-level description of cardinal brain functions is still far from reach. Using analytical or numerical approaches, \emphComputational Neuroscience aims at the emergence of such common principles by using concepts from Dynamical Systems and Information Theory. The aim of this Special Issue of the Journal of Physiology (Paris) is to reflect the latest advances in this field which has been presented during the NeuroComp08 conference that took place in October 2008 in Marseille (France). By highlighting a selection of works presented at the conference, we wish to illustrate the intrinsic diversity of this field of research but also the need of an unification effort that is becoming more and more necessary to understand the brain in its full complexity, from multiple levels of description to a multi-level understanding.

• Laurent U. Perrinet. Role of homeostasis in learning sparse representations, URL . Neural Computation, 22(7):1812--36, 2010

  Neurons in the input layer of primary visual cortex in primates develop edge-like receptive fields. One approach to understanding the emergence of this response is to state that neural activity has to efficiently represent sensory data with respect to the statistics of natural scenes. Furthermore, it is believed that such an efficient coding is achieved using a competition across neurons so as to generate a sparse representation, that is, where a relatively small number of neurons are simultaneously active. Indeed, different models of sparse coding coupled with Hebbian learning and homeostasis have been proposed that successfully match the observed emergent response. However, the specific role of homeostasis in learning such sparse representations is still largely unknown. By quantitatively assessing the efficiency of the neural representation during learning, we derive a cooperative homeostasis mechanism which optimally tunes the competition between neurons within the sparse coding algorithm. We apply this homeostasis while learning small patches taken from natural images and compare its efficiency with state-of-the-art algorithms. Results show that while different sparse coding algorithms give similar coding results, the homeostasis provides an optimal balance for the representation of natural images within the population of neurons. Competition in sparse coding is optimized when it is fair: By contributing to optimize statistical competition across neurons, homeostasis is crucial in providing a more efficient solution to the emergence of independent components.

references of articles and proceedings

• Pierre Yger, Daniel Bruderle, Jochen Eppler, Jens Kremkow, Dejan Pecevski, Laurent Perrinet, Michael Schmuker, Eilif Muller, Andrew P Davison. NeuralEnsemble: Towards a meta-environment for network modeling and data analysis, URL . In Eighth Göttingen Meeting of the German Neuroscience Society, pages T26-4C. 2009 NeuralEnsemble (http://neuralensemble.org) is a multilateral effort to coordinate and organise neuroscience software development efforts based around the Python programming language into a larger, meta-simulator software system. To this end, NeuralEnsemble hosts services for source code management and bug tracking (Subversion/Trac) for a number of open-source neuroscience tools, organizes an annual workshop devoted to collaborative software development in neuroscience, and manages a google-group discussion forum. Here, we present two NeuralEnsemble hosted projects:PyNN (http://neuralensemble.org/PyNN) is a package for simulator-independent specification of neuronal network models. You can write the code for a model once, using the PyNN API, and then run it without modification on any simulator that PyNN supports. Currently NEURON, NEST, PCSIM and a VLSI hardware implementation are fully supported.NeuroTools (http://neuralensemble.org/NeuroTools) is a set of tools to manage, store and analyse computational neuroscience simulations. It has been designed around PyNN, but can also be used for data from other simulation environments or even electrophysiological measurements.We will illustrate how the use of PyNN and NeuroTools ease the developmental process of models in computational neuroscience, enhancing collaboration between different groups and increasing the confidence in correctness of results. NeuralEnsemble efforts are supported by the European FACETS project (EU-IST-2005-15879)

• Adrien Wohrer, Guillaume Masson, Laurent Perrinet, Pierre Kornprobst, Thierry Vieville. Contrast sensitivity adaptation in a virtual spiking retina and its adequation with mammalians retinas. In Perception, pages 67. 2009

• Nicole Voges, Laurent Perrinet. Phase space analysis of networks based on biologically realistic parameters, URL . Journal of Physiology (Paris), 104(1-2):51--60, 2010

  We study cortical network dynamics for a more realistic network model. It represents, in terms of spatial scale, a large piece of cortex allowing for long-range connections, resulting in a rather sparse connectivity. We use two different types of conductance-based I&F neurons as excitatory and in- hibitory units, as well as specific connection probabilities. In order to re- main computationally tractable, we reduce neuron density, modelling part of the missing internal input via external poissonian spike trains. Compared to previous studies, we observe significant changes in the dynamical phase space: Altered activity patterns require another regularity measure than the coefficient of variation. We identify two types of mixed states, where differ- ent phases coexist in certain regions of the phase space. More notably, our boundary between high and low activity states depends predominantly on the relation between excitatory and inhibitory synaptic strength instead of the input rate.Key words: Artificial neural networks, Data analysis, Simulation, Spiking neurons. This work is supported by EC IP project FP6-015879 (FACETS).

• Nicole Voges, Laurent Perrinet. Dynamics of cortical networks including long-range patchy connections, URL . In Eighth Göttingen Meeting of the German Neuroscience Society, pages T26-3C. 2009 Most studies of cortical network dynamics are either based on purely random wiring or neighborhood couplings [1], focussing on a rather local scale. Neuronal connections in the cortex, however, show a more complex spatial pattern composed of local and long-range patchy connections [2,3] as shown in the figure: It represents a tracer injection (gray areas) in the GM of a flattened cortex (top view): Black dots indicate neuron positions, blue lines their patchy axonal ramifications, and red lines represent the local connections. Moreover, to include distant synapses, one has to enlarge the spatial scale from the typically assumed 1mm to 5mm side length.As it is our aim to analyze more realistic network models of the cortex we assume a distance dependent connectivity that reflects the geometry of dendritesand axons [3]. Here, we ask to what extent the assumption of specific geometric traits influences the resulting dynamical behavior of these networks. Analyzing various characteristic measures that describe spiking neurons (e.g., coefficient of variation, correlation coefficient), we compare the dynamical state spaces of different connectivity types: purely random or purely local couplings, a combination of local and distant synapses, and connectivity structures with patchy projections.On top of biologically realistic background states, a stimulus is applied in order to analyze their stabilities. As previous studies [1], we also find different dynamical states depending on the external input rate and the numerical relation between excitatory and inhibitory synaptic weights. Preliminary results indicate, however, that transitions between these states are much sharper in case of local or patchy couplings.This work is supported by EU Grant 15879 (FACETS). Thanks to Stefan Rotter who supervised the PhD project [3] this work is based on. Network dynamics are simulated with NEST/PyNN [4].[1] A. Kumar, S. Schrader, A. Aertsen and S. Rotter, Neural Computation 20, 2008, 1-43. [2] T. Binzegger, R.J. Douglas and K.A.C. Martin, J. of Neurosci., 27(45), 2007, 12242-12254. [3] Voges N, Fakultaet fuer Biologie, Albert-Ludwigs-Universitaet Freiburg, 2007. [4] NEST. M.O. Gewaltig and M. Diesmann, Scholarpedia 2(4):1430.

• Nicole Voges, Laurent U. Perrinet. Dynamical state spaces of cortical networks representing various horizontal connectivities, URL . In Proceedings of COSYNE, 2009

  Most studies of cortical network dynamics are either based on purely random wiring or neighborhood couplings, e.g., [Kumar, Schrader, Aer tsen, Rotter, 2008, Neural Computation 20, 1--43]. Neuronal connections in the cortex, however, show a complex spatial pattern composed of local and long-range connections, the latter featuring a so-called patchy projection pattern, i.e., spatially clustered synapses [Binzegger, Douglas, Martin, 2007, J. Neurosci. 27(45), 12242--12254]. The idea of our project is to provide and to analyze probabilistic network models that more adequately represent horizontal connectivity in the cortex. In particular, we investigate the effect of specific projection patterns on the dynamical state space of cortical networks. Assuming an enlarged spatial scale we employ a distance dependent connectivity that reflects the geometr y of dendrites and axons. We simulate the network dynamics using a neuronal network simulator NEST/PyNN. Our models are composed of conductance based integrate-and-fire neurons, representing fast spiking inhibitor y and regular spiking excitator y cells. In order to compare the dynamical state spaces of previous studies with our network models we consider the following connectivity assumptions: purely random or purely local couplings, a combination of local and distant synapses, and connectivity structures with patchy projections. Similar to previous studies, we also find different dynamical states depending on the input parameters: the external input rate and the numerical relation between excitatory and inhibitory synaptic weights. These states, e.g., synchronous regular (SR) or asynchronous irregular (AI) firing, are characterized by measures like the mean firing rate, the correlation coefficient, the coefficient of variation and so forth. On top of identified biologically realistic background states (AI), stimuli are applied in order to analyze their stability. Comparing the results of our different network models we find that the parameter space necessary to describe all possible dynamical states of a network is much more concentrated if local couplings are involved. The transition between different states is shifted (with respect to both input parameters) and sharpened in dependence of the relative amount of local couplings. Local couplings strongly enhance the mean firing rate, and lead to smaller values of the correlation coefficient. In terms of emergence of synchronous states, however, networks with local versus non-local or patchy versus random remote connections exhibit a higher probability of synchronized spiking. Concerning stability, preliminary results indicate that again networks with local or patchy connections show a higher probability of changing from the AI to the SR state. We conclude that the combination of local and remote projections bears important consequences on the activity of network: The apparent differences we found for distinct connectivity assumptions in the dynamical state spaces suggest that network dynamics strongly depend on the connectivity structure. This effect might be even stronger with respect to the spatio-temporal spread of signal propagation. This work is suppor ted by EC IP project FP6-015879 (FACETS).

• Nicole Voges, Laurent Perrinet. Recurrent cortical networks with realistic horizontal connectivities show complex dynamics, URL . In Eighteenth Annual Computational Neuroscience Meeting: CNS*2009 Berlin, Germany. 18–23 July 2009, pages T26-3C + 10(Suppl 1):P176. 2009 Most studies on the dynamics of recurrent cortical networks are either based on purely randomwiring or neighborhood couplings. They deal with a local spatial scale, where approx.10% of all possible connections are realized. Neuronal wiring in the cortex, however, shows acomplex spatial pattern composed of local and long-range patchy connections, i.e. spatiallyclustered synapses.We ask to what extent such geometric traits influence the ’idle’ dynamics of cortical networkmodels. Assuming an enlarged spatial scale we consider distinct network architectures, rang-ing from purely random to distance dependent connectivities with patchy projections. Thelatter are tuned to reflect the axonal arborizations present in layer 2/3 of cat V1. We con-sider different types of conductance based integrate-and-fire neurons with distance-dependentsynaptic delays.Analyzing the characteristic measures describing spiking neuronal networks (e.g. correlations,regularity), we explore and compare the phase spaces and activity patterns of different typesof network models. To examine stability and signal propagation properties we additionallyapplied local activity injections.Similar to previous studies we observe synchronous regular firing (SR state) for large νext andlow inhibition, while small νext combined with large g results in asynchronous irregular firing(AI). Our SRslow and SI state, the occurrence of ’mixed’ states, and the more vertical phasespace border significantly differ from previous findings.

• Nicole Voges, Laurent U. Perrinet. Analyzing cortical network dynamics with respect to different connectivity assumptions, URL . In Proceedings of the second french conference on Computational Neuroscience, Marseille, 2008

• Nicole Voges, Jens Kremkow, Laurent U. Perrinet. Dynamics of cortical networks based on patchy connectivity patterns. In FENS Abstract, 2008

• Claudio Simoncini, Laurent U. Perrinet, Anna Montagnini, Pascal Mamassian, Guillaume S. Masson. Different pooling of motion information for perceptual speed discrimination and behavioral speed estimation. In Vision Science Society, 2010

• Laurent U. Perrinet. Role of homeostasis in learning sparse representations, URL . Neural Computation, 22(7):1812--36, 2010

  Neurons in the input layer of primary visual cortex in primates develop edge-like receptive fields. One approach to understanding the emergence of this response is to state that neural activity has to efficiently represent sensory data with respect to the statistics of natural scenes. Furthermore, it is believed that such an efficient coding is achieved using a competition across neurons so as to generate a sparse representation, that is, where a relatively small number of neurons are simultaneously active. Indeed, different models of sparse coding coupled with Hebbian learning and homeostasis have been proposed that successfully match the observed emergent response. However, the specific role of homeostasis in learning such sparse representations is still largely unknown. By quantitatively assessing the efficiency of the neural representation during learning, we derive a cooperative homeostasis mechanism which optimally tunes the competition between neurons within the sparse coding algorithm. We apply this homeostasis while learning small patches taken from natural images and compare its efficiency with state-of-the-art algorithms. Results show that while different sparse coding algorithms give similar coding results, the homeostasis provides an optimal balance for the representation of natural images within the population of neurons. Competition in sparse coding is optimized when it is fair: By contributing to optimize statistical competition across neurons, homeostasis is crucial in providing a more efficient solution to the emergence of independent components.

• Laurent Perrinet, Guillaume S. Masson. Dynamical emergence of a neural solution for motion integration, URL . In Proceedings of AREADNE, 2010

• Laurent Perrinet. Qui créera le premier calculateur intelligent?, URL . DocSciences, (13), 2010

• Laurent Perrinet, Alexandre Reynaud, Frédéric Chavane, Guillaume S. Masson. Inferring monkey ocular following responses from V1 population dynamics using a probabilistic model of motion integration, URL . In Vision Science Society, 2009

  Short presentation of a large moving pattern elicits an ocular following response that exhibits many of the properties attributed to low-level motion processing such as spatial and temporal integration, contrast gain control and divisive interaction between competing motions. Similar mechanisms have been demonstrated in V1 cortical activity in response to center-surround gratings patterns measured with real-time optical imaging in awake monkeys (see poster of Reynaud et al., VSS09). Based on a previously developed Bayesian framework, we have developed an optimal statistical decoder of such an observed cortical population activity as recorded by optical imaging. This model aims at characterizing the statistical dependence between early neuronal activity and ocular responses and its performance was analyzed by comparing this neuronal read-out and the actual motor responses on a trial-by-trial basis. First, we show that relative performance of the behavioral contrast response function is similar to the best estimate obtained from the neural activity. In particular, we show that the latency of ocular response increases with low contrast conditions as well as with noisier instances of the behavioral task as decoded by the model. Then, we investigate the temporal dynamics of both neuronal and motor responses and show how motion information as represented by the model is integrated in space to improve population decoding over time. Lastly, we explore how a surrounding velocity non congruous with the central excitation information shunts the ocular response and how it is topographically represented in the cortical activity. Acknowledgment: European integrated project FACETS IST-15879.

• Laurent Perrinet, Nicole Voges, Jens Kremkow, Guillaume S. Masson. Decoding center-surround interactions in population of neurons for the ocular following response , URL . In Proceedings of COSYNE, 2009

  Short presentation of a large moving pattern elicits an Ocular Following Response (OFR) that exhibits many of the properties attributed to low-level motion processing such as spatial and temporal integration, contrast gain control and divisive interaction between competing motions. Similar mechanisms have been demonstrated in V1 cortical activity in response to center-surround gratings patterns measured with real-time optical imaging in awake monkeys. More recent experiments of OFR have used disk gratings and bipartite stimuli which are optimized to study the dynamics of center-surround integration. We quantified two main characteristics of the global spatial integration of motion from an intermediate map of possible local translation velocities: (i) a finite optimal stimulus size for driving OFR, surrounded by an antagonistic modulation and (ii) a direction selective suppressive effect of the surround on the contrast gain control of the central stimuli [Barthelemy06,Barthelemy07].In fact, the machinery behind the visual perception of motion and the subsequent sensorimotor transformation is confronted to uncertainties which are efficiently resolved in the primate's visual system. We may understand this response as an ideal observer in a probabilistic framework by using Bayesian theory [Weiss02] and we extended in the dynamical domain the ideal observer model to simulate the spatial integration of the different local motion cues within a probabilistic representation. We proved that this model is successfully adapted to model the OFR for the different experiments [Perrinet07neurocomp], that is for different levels of noise with full field gratings, with disks of various sizes and also for the effect of a flickering surround. However, another \emphad hoc inhibitory mechanism has to be added in this model to account for suppressive effects of the surround.We explore here an hypothesis where this could be understood as the effect of a recurrent prediction of information in the velocity map. In fact, in previous models, the integration step assumes independence of the local information while natural scenes are very predictable: Due to the rigidity and inertia of physical objects in visual space, neighboring local spatiotemporal information is redundant and one may introduce this \empha priori knowledge of the statistics of the input in the ideal observer model. We implement this in a realistic model of a layer representing velocities in a map of cortical columns, where predictions are implemented by lateral interactions within the cortical area. First, raw velocities are estimated locally from images and are propagated to this area in a feed-forward manner. Using this velocity map, we progressively learn the dependence of local velocities in a second layer of the model. This algorithm is cyclic since the prediction is using the local velocities which are themselves using both the feed-forward input and the prediction: We control the convergence of this process by measuring results for different learning rate. Results show that this simple model is sufficient to disambiguate characteristic patterns such as the Barber-Pole illusion. Due to the recursive network which is modulating the velocity map, it also explains that the representation may exhibit some memory, such as when an object suddenly disappears or when presenting a dot followed by a line (line-motion illusion).Finally, we applied this model that was tuned over a set of natural scenes to gratings of increasing sizes. We observed first that the feed-forward response as tuned to neurophysiological data gave lower responses at higher eccentricities, and that this effect was greater for higher grating frequencies. Then, we observed that depending on the size of the disk and on its spatial frequency, the recurrent network of lateral interactions Lastly, we explore how a surrounding velocity non congruous with the central excitation information shunts the ocular response and how it is topographically represented in the cortical activity. ,

• Laurent Perrinet, Guillaume S. Masson. Decoding the population dynamics underlying ocular following response using a probabilistic framework. In Eighteenth Annual Computational Neuroscience Meeting: CNS*2009 Berlin, Germany. 18--23 July 2009, pages 10(Suppl 1):P359. 2009

• Laurent Perrinet. Adaptive Sparse Spike Coding : applications of Neuroscience to the compression of natural images, URL . In Optical and Digital Image Processing Conference 7000 - Proceedings of SPIE Volume 7000, 7 - 11 April 2008, pages 15 - S4. 2008 If modern computers are sometimes superior to cognition in some specialized tasks such as playing chess or browsing a large database, they can't beat the efficiency of biological vision for such simple tasks as recognizing a relative or following an object in a complex background. We present in this paper our attempt at outlining the dynamical, parallel and event-based representation for vision in the architecture of the central nervous system. We will illustrate this by showing that in a signal matching framework, a L/LN (linear/non-linear) cascade may efficiently transform a sensory signal into a neural spiking signal and we apply this framework to a model retina. However, this code gets redundant when using an over-complete basis as is necessary for modeling the primary visual cortex: we therefore optimize the efficiency cost by increasing the sparseness of the code. This is implemented by propagating and canceling redundant information using lateral interactions. We compare the efficiency of this representation in terms of compression as the reconstruction quality as a function of the coding length. This will correspond to a modification of the Matching Pursuit algorithm where the ArgMax function is optimized for competition, or Competition Optimized Matching Pursuit (COMP). We will particularly focus on bridging neuroscience and image processing and on the advantages of such an interdisciplinary approach.

• Laurent Perrinet, Guillaume S. Masson. Modeling spatial integration in the ocular following response to center-surround stimulation using a probabilistic framework, URL . In Proceedings of COSYNE, 2008, 2008,

• Laurent Perrinet. What adaptive code for efficient spiking representations? A model for the formation of receptive fields of simple cells., URL . In Proceedings of COSYNE, 2008

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

• Laurent Perrinet, Guillaume S. Masson. Modeling spatial integration in the ocular following response using a probabilistic framework, URL . Journal of Physiology (Paris), 2007

  The machinery behind the visual perception of motion and the subsequent sensori-motor transformation, such as in Ocular Following Response (OFR), is confronted to uncertainties which are efficiently resolved in the primate's visual system. We may understand this response as an ideal observer in a probabilis- tic framework by using Bayesian theory (Weiss et al., 2002) which we previously proved to be successfully adapted to model the OFR for different levels of noise with full field gratings (Perrinet et al., 2005). More recent experiments of OFR have used disk gratings and bipartite stimuli which are optimized to study the dy- namics of center-surround integration. We quantified two main characteristics of the spatial integration of motion : (i) a finite optimal stimulus size for driving OFR, surrounded by an antagonistic modulation and (ii) a direction selective sup- pressive effect of the surround on the contrast gain control of the central stim- uli (Barth\'elemy et al., 2006). Herein, we extended the ideal observer model to simulate the spatial integration of the different local motion cues within a proba- bilistic representation. We present analytical results which show that the hypoth- esis of independence of local measures can describe the integration of the spatial motion signal. Within this framework, we successfully accounted for the con- trast gain control mechanisms observed in the behavioral data for center-surround stimuli. However, another inhibitory mechanism had to be added to account for suppressive effects of the surround.

• Laurent Perrinet. On efficient sparse spike coding schemes for learning natural scenes in the primary visual cortex, URL . In Sixteenth Annual Computational Neuroscience Meeting: CNS*2007, Toronto, Canada. 7--12 July 2007, 2007

  We describe the theoretical formulation of a learning algorithm in a model of the primary visual cortex (V1) and present results of the efficiency of this algorithm by comparing it to the SparseNet algorithm [1]. As the SparseNet algorithm, it is based on a model of signal synthesis as a Linear Generative Model but differs in the efficiency criteria for the representation. This learning algorithm is in fact based on an efficiency criteria based on the Occam razor: for a similar quality, the shortest representation should be privileged. This inverse problem is NP-complete and we propose here a greedy solution which is based on the architecture and nature of neural computations [2]). It proposes that the supra-threshold neural activity progressively removes redundancies in the representation based on a correlation-based inhibition and provides a dynamical implementation close to the concept of neural assemblies from Hebb [3]). We present here results of simulation of this network with small natural images (available at https://laurentperrinet.github.io/publication/perrinet-19-hulk) and compare it to the Sparsenet solution. Extending it to realistic images and to the NEST simulator http://www.nest-initiative.org/, we show that this learning algorithm based on the properties of neural computations produces adaptive and efficient representations in V1. 1. Olshausen B, Field DJ: Sparse coding with an overcomplete basis set: A strategy employed by V1? Vision Res 1997, 37:3311-3325.2. Perrinet L: Feature detection using spikes: the greedy approach. J Physiol Paris 2004, 98(4–6):530-539.3. Hebb DO: The organization of behavior. Wiley, New York; 1949.

• Laurent Perrinet, Frédéric V. Barthélemy, Guillaume S. Masson. Input-output transformation in the visuo-oculomotor loop: modeling the ocular following response to center-surround stimulation in a probabilistic framework. In 1ère conférence francophone NEUROsciences COMPutationnelles - NeuroComp, 2006

  The quality of the representation of an object's motion is limited by the noise in the sensory input as well as by an intrinsic ambiguity due to the spatial limi- tation of the visual motion analyzers (aperture prob- lem). Perceptual and oculomotor data demonstrate that motion processing of extended ob jects is initially dominated by the local 1D motion cues orthogonal to the ob ject's edges, whereas 2D information takes pro- gressively over and leads to the final correct represen- tation of global motion. A Bayesian framework ac- counting for the sensory noise and general expectancies for ob ject velocities has proven successful in explaining several experimental findings concerning early motion processing [1, 2, 3]. However, a complete functional model, encompassing the dynamical evolution of ob- ject motion perception is still lacking. Here we outline several experimental observations concerning human smooth pursuit of moving ob jects and more particu- larly the time course of its initiation phase. In addi- tion, we propose a recursive extension of the Bayesian model, motivated and constrained by our oculomotor data, to describe the dynamical integration of 1D and 2D motion information.

• Laurent Perrinet, Jens Kremkow, Frédéric Barthélemy, Guillaume S. Masson, Frédéric Chavane. Input-output transformation in the visuo-oculomotor loop: modeling the ocular following response to center-surround stimulation in a probabilistic framework. In FENS, 2006

• Laurent Perrinet, Jens Kremkow. Dynamical contrast gain control mechanisms in a layer 2/3 model of the primary visual cortex. In The Functional Architecture of the Brain : from Dendrites to Networks. Symposium in honour of Dr Suzanne Tyc-Dumont. 4- 5 May 2006. GLM, Marseille, France, 2006 Computations in a cortical column are characterized by the dynamical, event-based nature of neuronal signals and are structured by the layered and parallel structure of cortical areas. But they are also characterized by their efficiency in terms of rapidity and robustness. We propose and study here a model of information integration in the primary visual cortex (V1) thanks to the parallel and interconnected network of similar cortical columns. In particular, we focus on the dynamics of contrast gain control mechanisms as a function of the distribution of information relevance in a small population of cortical columns. This cortical area is modeled as a collection of similar cortical columns which receive input and are linked according to a specific connectivity pattern which is relevant to this area. These columns are simulated using the \sc Nest simulator \citepMorrison04 using conductance-based Integrate-and-Fire neurons and consist vertically in 3 different layers. The architecture was inspired by neuro-physiological observations on the influence of neighboring activities on pyramidal cells activity and correlates with the lateral flow of information observed in the primary visual cortex, notably in optical imaging experiments \citepJancke04, and is similar in its final implementation to local micro-circuitry of the cortical column presented by \citetGrossberg05. %They show prototypical spontaneous dynamical behavior to different levels of noise which are relevant to the generic modeling of biological cortical columns \citepKremkow05. In the future, the connectivity will be derived from an algorithm that was used for modeling the transient spiking response of a layer of neurons to a flashed image and which was based on the Matching Pursuit algorithm \citepPerrinet04. %The visual input is first transmitted from the Lateral Geniculate Nucleus (LGN) using the model of \citetGazeres98. It transforms the image flow into a stream of spikes with contrast gain control mechanisms specific to the retina and the LGN. This spiking activity converges to the pyramidal cells of layer 2/3 thanks to the specification of receptive fields in layer 4 providing a preference for oriented local contrasts in the spatio-temporal visual flow. In particular, we use in these experiments visual input organized in a center-surround spatial pattern which was optimized in size to maximize the response of a column in the center and to the modulation of this response by the surround (bipartite stimulus). This class of stimuli provide different levels of input activation and of visual ambiguity in the visual space which were present in the spatio-temporal correlations in the input spike flow optimized to the resolution of cortical columns in the visual space. It thus provides a method to reveal the dynamics of information integration and particularly of contrast gain control which are characteristic to the function of V1.

• Laurent Perrinet. An efficiency razor for model selection and adaptation in the primary visual cortex. In Fifteenth Annual Computational Neuroscience Meeting, 2006

  We describe the theoretical formulation of a learning algorithm in a model of the primary visual cortex (V1) and present results of the efficiency of this algorithm by comparing it to the Sparsenet algorithm (Olshausen, 1996). As the Sparsenet algorithm, it is based on a model of signal synthesis as a Linear Generative Model but differs in the efficiency criteria for the representation. This learning algorithm is in fact based on an efficiency criteria based on the Occam razor: for a similar quality, the shortest representation should be privilegied. This inverse problem is NP-complete and we propose here a greedy solution which is based on the architecture and nature of neural computations (Perrinet, 2006). We present here results of a simulation of this network of small natural images (available at https://laurentperrinet.github.io/publication/perrinet-19-hulk ) and compare it to the Sparsenet solution. We show that this solution based on neural computations produces an adaptive algorithm for efficient representations in V1.

• Laurent Perrinet, Jens Kremkow. Dynamical contrast gain control mechanisms in a layer 2/3 model of the primary visual cortex. In Physiogenic and pathogenic oscillations: the beauty and the beast, 5th INMED/TINS CONFERENCE SEPTEMBER 9 - 12, 2006, La Ciotat, France, 2006

• Laurent Perrinet. Dynamical Neural Networks: modeling low-level vision at short latencies, URL . pages 163--225.

• Anna Montagnini, Pascal Mamassian, Laurent U. Perrinet, Eric Castet, Guillaume S. Masson. Bayesian modeling of dynamic motion integration, URL . Journal of Physiology (Paris), 101(1-3):64-77, 2007

  The quality of the representation of an object's motion is limited by the noise in the sensory input as well as by an intrinsic ambiguity due to the spatial limi- tation of the visual motion analyzers (aperture prob- lem). Perceptual and oculomotor data demonstrate that motion processing of extended ob jects is initially dominated by the local 1D motion cues orthogonal to the ob ject's edges, whereas 2D information takes pro- gressively over and leads to the final correct represen- tation of global motion. A Bayesian framework ac- counting for the sensory noise and general expectancies for ob ject velocities has proven successful in explaining several experimental findings concerning early motion processing [1, 2, 3]. However, a complete functional model, encompassing the dynamical evolution of ob- ject motion perception is still lacking. Here we outline several experimental observations concerning human smooth pursuit of moving ob jects and more particu- larly the time course of its initiation phase. In addi- tion, we propose a recursive extension of the Bayesian model, motivated and constrained by our oculomotor data, to describe the dynamical integration of 1D and 2D motion information.

• Jens Kremkow, Laurent U. Perrinet, Guillaume S. Masson, Ad Aertsen. Functional consequences of correlated excitatory and inhibitory conductances in cortical networks, URL . Journal of Computational Neuroscience, 28(3):579-94, 2010

  Neurons in the neocortex receive a large number of excitatory and inhibitory synaptic inputs. Excitation and inhibition dynamically balance each other, with inhibition lagging excitation by only few milliseconds. To characterize the functional consequences of such correlated excitation and inhibition, we studied models in which this correlation structure is induced by feedforward inhibition (FFI). Simple circuits show that an effective FFI changes the integrative behavior of neurons such that only synchronous inputs can elicit spikes, causing the responses to be sparse and precise. Further, effective FFI increases the selectivity for propagation of synchrony through a feedforward network, thereby increasing the stability to background activity. Last, we show that recurrent random networks with effective inhibition are more likely to exhibit dynamical network activity states as have been observed in vivo. Thus, when a feedforward signal path is embedded in such recurrent network, the stabilizing effect of effective inhibition creates an suitable substrate for signal propagation. In conclusion, correlated excitation and inhibition support the notion that synchronous spiking may be important for cortical processing.

• Jens Kremkow. Correlating Excitation and Inhibition in Visual Cortical Circuits: Functional Consequences and Biological Feasibility, PhD thesis. 2009 The primary visual cortex (V1) is one of the most studied cortical area in the brain. Together with the retina and the lateral geniculate nucleus (LGN) it forms the early visual system. Artificial stimuli (i.e. drifting gratings (DG)) have given insights into the neural basis of visual processing. However, recently researchers have started to use more complex natural visual stimuli (NI), arguing that the low dimensional artificial stimuli are not sufficient for a complete understanding of the visual system.For example, whereas the responses of V1 neurons to DG are dense but with variable spike timings, the neurons respond with only few but precise spikes to NI. Furthermore, linear receptive field models provide a good fit to responses during simple stimuli, however, they often fail during NI. To investigate the mechanisms behind the stimulus dependent responses of cortical neurons we have built a biophysical model of the early visual system.Our results show that during NI the LGN afferents show epochs of correlated activity, resulting in precise spike timings in V1. The sparseness of the responses to NI can be explained by correlated inhibitory conductance. We continue by investigating the origin of stimulus dependent nonlinear responses, by comparing models of different complexity. Our results suggest that adaptive processes shape the responses, depending on the temporal properties of the stimuli. Lastly we study the functional consequences of correlated excitatory and inhibitory condutances in more details in generic models.The presented work gives new perspectives on the processing of the early visual system, in particular on the importance of correlated conductances.

• Jens Kremkow, Laurent Perrinet, Guillaume S. Masson, Ad Aertsen. Functional consequences of correlated excitation and inhibition on single neuron integration and signal propagation through synfire chains, URL . In Eighth Göttingen Meeting of the German Neuroscience Society, pages T26-6B. 2009 Neurons receive a large number of excitatory and inhibitory synaptic inputs whose temporal interplay determines their spiking behavior. On average, excitation (Gexc) and inhibition (Ginh) balance each other, such that spikes are elicited by fluctuations [1]. In addition, it has been shown in vivo that Gexc and Ginh are correlated, with Ginh lagging Gexc only by few milliseconds (6ms), creating a small temporal integration window [2,3]. This correlation structure could be induced by feed-forward inhibition (FFI), which has been shown to be present at many sites in the central nervous system.To characterize the functional consequences of the FFI, we first modeled a simple circuit using spiking neurons with conductance based synapses and studied the effect on the single neuron integration. We then coupled many of such circuits to construct a feed-forward network (synfire chain [4,5]) and investigated the effect of FFI on signal propagation along such feed-forward network.We found that the small temporal integration window, induced by the FFI, changes the integrative properties of the neuron. Only transient stimuli could produce a response when the FFI was active whereas without FFI the neuron responded to both steady and transient stimuli. Due to the increase in selectivity to transient inputs, the conditions of signal propagation through the feed-forward network changed as well. Whereas synchronous inputs could reliable propagate, high asynchronous input rates, which are known to induce synfire activity [6], failed to do so. In summary, the FFI increased the stability of the synfire chain.Supported by DFG SFB 780, EU-15879-FACETS, BMBF 01GQ0420 to BCCN Freiburg[1] Kumar A., Schrader S., Aertsen A. and Rotter S. (2008). The high-conductance state of cortical networks. Neural Computation, 20(1):1--43. [2] Okun M. and Lampl I. (2008). Instantaneous correlation of excitation and inhibition during ongoing and sensory-evoked activities. Nat Neurosci, 11(5):535--7.[3] Baudot P., Levy M., Marre O., Monier C. and Fr\'egnac (2008). submitted. [4] Abeles M. (1991). Corticonics: Neural circuits of the cerebral cortex. Cambridge, UK [5] Diesmann M., Gewaltig M-O and Aertsen A. (1999). Stable propagation of synchronous spiking in cortical neural networks. Nature, 402(6761):529--33. [6] Kumar A., Rotter S. and Aertsen A. (2008), Conditions for propagating synchronous spiking and asynchronous firing rates in a cortical network model. J Neurosci 28 (20), 5268--80.,

• Jens Kremkow, Laurent Perrinet, Cyril Monier, Yves Fregnac, Guillaume S. Masson, Ad Aertsen. Control of the temporal interplay between excitation and inhibition by the statistics of visual input, URL . URL2 . In Eighteenth Annual Computational Neuroscience Meeting: CNS*2009 Berlin, Germany. 18–23 July 2009, pages Oral presentation, 10(Suppl 1):O21. 2009

• Jens Kremkow, Laurent Perrinet, Alexandre Reynaud, Ad Aertsen, Guillaume S. Masson, Frédéric Chavane. Dynamics of non-linear cortico-cortical interactions during motion integration in early visual cortex: A spiking neuron model of an optical imaging study in the awake monkey, URL . URL2 . In Eighteenth Annual Computational Neuroscience Meeting: CNS*2009 Berlin, Germany. 18–23 July 2009, pages 10(Suppl 1):P176. 2009

• Jens Kremkow, Laurent Perrinet, Pierre Baudot, Manu Levy, Olivier Marre, Cyril Monier, Yves Fregnac, Guillaume Masson, Ad Aertsen. Control of the temporal interplay between excitation and inhibition by the statistics of visual input: a V1 network modelling study, URL . In Proceedings of the Society for Neuroscience conference, 2008

  In the primary visual cortex (V1), single cell responses to simple visual stimuli (gratings) are usually dense but with a high trial-by-trial variability. In contrast, when exposed to full field natural scenes, the firing patterns of these neurons are sparse but highly reproducible over trials (Marre et al., 2005; Fr\'egnac et al., 2006). It is still not understood how these two classes of stimuli can elicit these two distinct firing behaviours. A common model for simple-cell computation in layer 4 is the ``push-pull'' circuitry (Troyer et al. 1998). It accounts for the observed anti-phase behaviour between excitatory and inhibitory conductances in response to a drifting grating (Anderson et al., 2000; Monier et al., 2008), creating a wide temporal integration window during which excitation is integrated without the shunting or opponent effect of inhibition and allowed to elicit multiple spikes. This is in contrast to recent results from intracellular recordings in vivo during presentation of natural scenes (Baudot et al., submitted). Here the excitatory and inhibitory conductances were highly correlated, with inhibition lagging excitation only by few milliseconds (~6 ms). This small lag creates a narrow temporal integration window such that only synchronized excitatory inputs can elicit a spike, similar to parallel observations in other cortical sensory areas (Wehr and Zador, 2003; Okun and Lampl, 2008). To investigate the cellular and network mechanisms underlying these two different correlation structures, we constructed a realistic model of the V1 network using spiking neurons with conductance based synapses. We calibrated our model to fit the irregular ongoing activity pattern as well as in vivo conductance measurements during drifting grating stimulation and then extracted predicted responses to natural scenes seen through eye-movements. Our simulations reproduced the above described experimental observation, together with anti-phase behaviour between excitation and inhibition during gratings and phase lagged activation during natural scenes. In conclusion, the same cortical network that shows dense and variable responses to gratings exhibits sparse and precise spiking to natural scenes. Work is under way to show to which extent this feature is specific for the feedforward vs recurrent nature of the modelled circuit. ,

• Jens Kremkow, Laurent U. Perrinet, Ad Aertsen, Guillaume S. Masson. Functional properties of feed-forward inhibition, URL . In Proceedings of the second french conference on Computational Neuroscience, Marseille, 2008

• Jens Kremkow, Laurent Perrinet, Arvind Kumar, Ad Aertsen, Guillaume Masson. Synchrony in thalamic inputs enhances propagation of activity through cortical layers, URL URL2 . In Sixteenth Annual Computational Neuroscience Meeting: CNS*2007, Toronto, Canada. 7--12 July 2007, 2007

  Sensory input enters the cortex via the thalamocortical (TC) projection, where it elicits large postsynaptic potentials in layer 4 neurons [1]. Interestingly, the TC connections account for only 15% of synapses onto these neurons. It has been therefore controversially discussed how thalamic input can drive the cortex. Strong TC synapses have been one suggestion to ensure the strength of the TC projection ("strong-synapse model"). Another possibility is that the excitation from single thalamic fibers are weak but get amplified by recurrent excitatory feedback in layer 4 ("amplifier model"). Bruno and Sakmann [2] recently provided new evidence that individual TC synapses in vivo are weak and only produce small excitatory postsynaptic potentials. However, they suggested that thalamic input can activate the cortex due to the synchronous firing and that cortical amplification is not required. This would support the "synchrony model" proposed by correlation analysis [3].Here, we studied the effect of correlation in the TC input, with weak synapses, to the responses of a layered cortical network model. The connectivity of the layered network was taken from Binzegger et al. 2004 [4]. The network was simulated using NEST [5] with the Python interface PyNN [6] to enable interoperability with different simulators. The sensory input to layer 4 was modelled by a simple retino-geniculate model of the transformation of light into spike trains [7], which was implemented by leaky integrate-and-fire model neurons.We found that introducing correlation into TC inputs enhanced the likelihood to produce responses in layer 4 and improved the activity propagation across layers. In addition, we compared the response of the cortical network to different noise conditions and obtained contrast response functions which were in accordance with neurophysiological observations. This Work is supported by the 6th RFP of the EU (grant no. 15879-FACETS) and by the BMBF grant 01GQ0420 to the BCCN Freiburg.1. Chung S, Ferster D: Strength and orientation tuning of the thalamic input to simple cells revealed by electrically evoked cortical suppression. Neuron 1998, 20:1177-1189. 2. Bruno M, Sakmann B: Cortex is driven by weak but synchronously active thalamocortical synpases. Science 2006, 312:1622-1627. 3. Alonso JM, Usrey WM, Reid RC: Precisely correlated firing in cells of the lateral geniculate nucleus. Nature 1996, 383:815-819. 4. Binzegger T, Douglas RJ, Martin KAC: A quantitative map of the circuit of the cat primary visual cortex. J Neurosci 2004, 24:8441-8453. 5. NEST http://www.nest-initiative.org. PyNN http://neuralensemble.org/PyNN. Gazeres N, Borg-Graham LJ, Fr\'egnac Y: A phenomenological model of visually evoked spike trains in cat geniculate nonlagged X-cells.Vis Neurosci 1998, 15:1157-1174.

• Mina Aliakbari Khoei, Laurent Perrinet, Guillaume S. Masson. Dynamical emergence of a neural solution for motion integration, URL . In Proceedings of Tauc, 2010

• Sylvain Fischer, Filip Sroubek, Laurent U. Perrinet, Rafael Redondo, Gabriel Crist\'obal. Self-invertible 2D log-Gabor wavelets, URL . Int. Journal of Computional Vision, 2007

• Sylvain Fischer, Rafael Redondo, Laurent Perrinet, Gabriel Crist\'obal. Sparse approximation of images inspired from the functional architecture of the primary visual areas, URL URL2 . EURASIP Journal on Advances in Signal Processing, special issue on Image Perception, :Article ID 90727, 16 pages, 2007

• Andrew P Davison, Daniel Bruderle, Jochen Eppler, Jens Kremkow, Eilif Muller, Dejan Pecevski, Laurent Perrinet, Pierre Yger. PyNN: A Common Interface for Neuronal Network Simulators., URL . Frontiers in Neuroinformatics, 2:11, 2008

  Computational neuroscience has produced a diversity of software for simulations of networks of spiking neurons, with both negative and positive consequences. On the one hand, each simulator uses its own programming or configuration language, leading to considerable difficulty in porting models from one simulator to another. This impedes communication between investigators and makes it harder to reproduce and build on the work of others. On the other hand, simulation results can be cross-checked between different simulators, giving greater confidence in their correctness, and each simulator has different optimizations, so the most appropriate simulator can be chosen for a given modelling task. A common programming interface to multiple simulators would reduce or eliminate the problems of simulator diversity while retaining the benefits. PyNN is such an interface, making it possible to write a simulation script once, using the Python programming language, and run it without modification on any supported simulator (currently NEURON, NEST, PCSIM, Brian and the Heidelberg VLSI neuromorphic hardware). PyNN increases the productivity of neuronal network modelling by providing high-level abstraction, by promoting code sharing and reuse, and by providing a foundation for simulator-agnostic analysis, visualization and data-management tools. PyNN increases the reliability of modelling studies by making it much easier to check results on multiple simulators. PyNN is open-source software and is available from http://neuralensemble.org/PyNN.

• Andrew Davison, Pierre Yger, Jens Kremkow, Laurent Perrinet, Eilif Muller. PyNN: towards a universal neural simulator API in Python, URL URL2 . In Sixteenth Annual Computational Neuroscience Meeting: CNS*2007, Toronto, Canada. 7--12 July 2007, 2007

  Trends in programming language development and adoption point to Python as the high-level systems integration language of choice. Python leverages a vast developer-base external to the neuroscience community, and promises leaps in simulation complexity and maintainability to any neural simulator that adopts it. PyNN http://neuralensemble.org/PyNN strives to provide a uniform application programming interface (API) across neural simulators. Presently NEURON and NEST are supported, and support for other simulators and neuromorphic VLSI hardware is under development.With PyNN it is possible to write a simulation script once and run it without modification on any supported simulator. It is also possible to write a script that uses capabilities specific to a single simulator. While this sacrifices simulator-independence, it adds flexibility, and can be a useful step in porting models between simulators. The design goals of PyNN include allowing access to low-level details of a simulation where necessary, while providing the capability to model at a high level of abstraction, with concomitant gains in development speed and simulation maintainability.Another of our aims with PyNN is to increase the productivity of neuroscience modeling, by making it faster to develop models de novo, by promoting code sharing and reuse across simulator communities, and by making it much easier to debug, test and validate simulations by running them on more than one simulator. Modelers would then become free to devote more software development effort to innovation, building on the simulator core with new tools such as network topology databases, stimulus programming, analysis and visualization tools, and simulation accounting. The resulting, community-developed 'meta-simulator' system would then represent a powerful tool for overcoming the so-called complexity bottleneck that is presently a major roadblock for neural modeling.

• Emmanuel Daucé, Laurent Perrinet. Computational Neuroscience, from Multiple Levels to Multi-level, URL . Journal of Physiology (Paris), 104(1--2):1--4, 2010

  Despite the long and fruitful history of neuroscience, a global, multi-level description of cardinal brain functions is still far from reach. Using analytical or numerical approaches, \emphComputational Neuroscience aims at the emergence of such common principles by using concepts from Dynamical Systems and Information Theory. The aim of this Special Issue of the Journal of Physiology (Paris) is to reflect the latest advances in this field which has been presented during the NeuroComp08 conference that took place in October 2008 in Marseille (France). By highlighting a selection of works presented at the conference, we wish to illustrate the intrinsic diversity of this field of research but also the need of an unification effort that is becoming more and more necessary to understand the brain in its full complexity, from multiple levels of description to a multi-level understanding.

• B. Cessac, E. Daucé, Laurent U. Perrinet, M. Samuelides. Topics in Dynamical Neural Networks: From Large Scale Neural Networks to Motor Control and Vision, `URL <https://laurentperrinet.github.io/publication/cessac-07>`__ . . Springer Berlin / Heidelberg, 2007.

• Amarender Bogadhi, Anna Montagnini, Pascal Mamassian, Laurent U. Perrinet, Guillaume S. Masson. A recurrent Bayesian model of dynamic motion integration for smooth pursuit. In Vision Science Society, 2010

• Amarender Bogadhi, Anna Montagnini, Pascal Mamassian, Laurent U. Perrinet, Guillaume S. Masson. Pursuing motion illusions: a realistic oculomotor framework for Bayesian inference, URL . Vision Research, 51(8):867--80, 2011

  Accuracy in estimating an object's global motion over time is not only affected by the noise in visual motion information but also by the spatial limitation of the local motion analyzers (aperture problem). Perceptual and oculomotor data demonstrate that during the initial stages of the motion information processing, 1D motion cues related to the object's edges have a dominating influence over the estimate of the object's global motion. However, during the later stages, 2D motion cues related to terminators (edge-endings) progressively take over, leading to a final correct estimate of the object's global motion. Here, we propose a recursive extension to the Bayesian framework for motion processing (Weiss, Simoncelli, & Adelson, 2002) cascaded with a model oculomotor plant to describe the dynamic integration of 1D and 2D motion information in the context of smooth pursuit eye movements. In the recurrent Bayesian framework, the prior defined in the velocity space is combined with the two independent measurement likelihood functions, representing edge-related and terminator-related information, respectively to obtain the posterior. The prior is updated with the posterior at the end of each iteration step. The maximum-a posteriori (MAP) of the posterior distribution at every time step is fed into the oculomotor plant to produce eye velocity responses that are compared to the human smooth pursuit data. The recurrent model was tuned with the variance of pursuit responses to either "pure" 1D or "pure" 2D motion. The oculomotor plant was tuned with an independent set of oculomotor data, including the effects of line length (i.e. stimulus energy) and directional anisotropies in the smooth pursuit responses. The model not only provides an accurate qualitative account of dynamic motion integration but also a quantitative account that is close to the smooth pursuit response across several conditions (three contrasts and three speeds) for two human subjects.

• Frédéric Barthélemy, Laurent Perrinet, Éric Castet, Guillaume S. Masson. Dynamics of distributed 1D and 2D motion representations for short-latency ocular following, URL . Vision Research, 48(4):501--22, 2008

on MacOsX: using MacPorts

• A basic installation procedure is to use the enthought distribution,

• Another route is to use MacPorts. It is a generic package manager inspired by what you get using Debian's apt scheme.

• Once installed, do on the command-line

  • on Leopard:

    sudo port install py25-pil py25-numpy py25-scipy py25-ipython py25-matplotlib +cairo+latex+tkinter
    sudo python_select python25

    (Note: you may also use python26 on Leopard).

  • on Snow Leopard:

    sudo port install py26-numpy py26-scipy py26-ipython py26-matplotlib
    sudo port install py26-pyobjc2-cocoa py26-pil py26-distribute py26-pip py26-py2app python_select
    sudo port install vtk5 +carbon +qt4_mac +python26 py26-mayavi
    
    sudo python_select python26

    to install a bunch of useful python packages.

  • to get a package that is not available through macports, do:

    sudo easy_install progressbar

• for visionEgg :

  sudo port install py26-opengl py26-game
  sudo easy_install visionegg

• http://ipython.scipy.org/moin/Py4Science/InstallationOSX

• on Snow Leopard, you'll have to follow these instructions.

Windows

• http://www.pythonxy.com

Debian / Ubuntu

• see http://neuro.debian.net/

  sudo aptitude install ipython python-numpy python-scipy python-matplotlib

DistUtils, PIP & Easy Install

• most of the time, there's a setup.py file:

  python setup.py install --prefix=~

• See http://peak.telecommunity.com/DevCenter/EasyInstall

• to install numpy (same for pylab, scipy, or visionegg), simply do

  easy_install numpy

• most of the cases, on a test server or a single-user machine, you may find more useful to install in your home dirtectory, for instance:

  easy_install -d ~/lib/python2.5/site-packages/ numpy

• to upgrade, use

  easy_install -U numpy

• you can browse the list of available packages.

• for pip: http://pip.openplans.org/

• you may create a script tu update all packages:

  for i in `python -c "for dist in __import__('pkg_resources').working_set: print dist.project_name"`:
  do
  echo "`easy_install -U $i`"
  echo "++++++++++++++++++++++++++++++++++++++++++++++++++"
  done

• to install PIL, use

  easy_install -d lib/python2.6/site-packages/ --find-links http://www.pythonware.com/products/pil/ Imaging

SVNs: bleeding edge versions

• numpy

  svn co http://svn.scipy.org/svn/numpy/trunk numpy
  cd numpy
  python setup.py build
  sudo python setup.py install
  rm -rf build
  cd ..

• SciPy

  svn co http://svn.scipy.org/svn/scipy/trunk scipy
  cd scipy
  python setup.py build
  sudo python setup.py install
  rm -rf build
  cd ..

  • see scipy http://www.scipy.org/Installing_SciPy/Mac_OS_X

• pylab

  svn co https://svn.sourceforge.net/svnroot/matplotlib/trunk/matplotlib matplotlib
  cd matplotlib
  python setup.py build
  sudo python setup.py install
  sudo rm -rf build
  cd ..

• SPE

  svn checkout svn://svn.berlios.de/python/spe/trunk/_spe

• PIL

  wget http://effbot.org/downloads/Imaging-1.1.6.tar.gz
  tar zxvf  Imaging-1.1.6.tar.gz
  cd Imaging-1.1.6
  python setup.py build_ext -i
  python selftest.py
  python setup.py install

• gsl

  cvs -d :pserver:anoncvs@sources.redhat.com:/cvs/gsl login
  cvs -d :pserver:anoncvs@sources.redhat.com:/cvs/gsl checkout gsl
  cd gsl/
  ./autogen.sh
  ./configure --enable-maintainer-mode
  make

• pytables

  • dependency on HDF

    wget ftp://ftp.hdfgroup.org/HDF5/current/src/hdf5-1.6.5.tar.gz
    tar zxvf hdf5-1.6.5.tar.gz
    cd hdf5-1.6.5
    ./configure --enable-cxx
    make
    make install
    h5ls -r  Documents/Sci/projets/virtualV1/experiments/benchmark_one/results/benchmark_retina_high.h5

    • and Py Rex

    wget http://www.cosc.canterbury.ac.nz/greg.ewing/python/Pyrex/Pyrex-0.9.5.1a.tar.gz
    tar zxvf Pyrex-0.9.5.1a.tar.gz
    cd Pyrex-0.9.5.1a
    python setup.py build
    sudo python setup.py install
    rm -rf build

  • install

    wget http://puzzle.dl.sourceforge.net/sourceforge/pytables/pytables-1.4.tar.gz
    #svn co http://pytables.org/svn/pytables/trunk/ pytables
    tar zxvf pytables-1.4.tar.gz
    cd pytables-1.4
    export DYLD_LIBRARY_PATH=/sw/lib # or in .bashrc
    python setup.py install --hdf5=/sw
    cd ..

• pygtk

  wget http://ftp.gnome.org/pub/GNOME/sources/pygtk/2.8/pygtk-2.8.6.tar.bz2
  tar xvfj pygtk-2.8.6.tar.bz2
  cd pygtk-2.8.6
  .configure
  make
  sudo make install    # or without sudo as root
  cd ..

Neurosciences Computationelles : émergence dans des réseaux d'information

using macports

• install py2app :

  sudo port install -u  py26-py2app

• there's sometimes a problem in py2app to check the right architecture to build on:

  find /opt/local -name apptemplate/setup.py
  sudo vim /opt/local/Library/Frameworks/Python.framework/Versions/2.6/lib/python2.6/site-packages/py2app/apptemplate/setup.py

• in this case, this can be done by adding the following lines to py2app/apptemplate/setup.py:

  gPreBuildVariants = [
      ...
      {
          'name': 'main-x86_64',
          'target': '10.6',
          'cflags': '-isysroot /Developer/SDKs/MacOSX10.6.sdk -arch x86_64',
          'cc': 'gcc-4.2',
      },
      {
          'name': 'main-i386',
          'target': '10.6',
          'cflags': '-isysroot / -arch i386',
          'cc': 'gcc-4.2',
      },
      ...
  ]

  . So, change to

  gPreBuildVariants = [
      {
          'name': 'main-x86_64',
          'target': '10.5',
          'cflags': '-isysroot /Developer/SDKs/MacOSX10.5.sdk -arch x86_64',
          'cc': 'gcc-4.2',
       },
  #     {
  #         'name': 'main-universal',
  #         'target': '10.5',
  #         'cflags': '-isysroot /Developer/SDKs/MacOSX10.5.sdk -arch i386 -arch ppc -arch ppc64 -arch x86_64',
  #         'cc': 'gcc-4.2',
  #     },
  #     {
  #         'name': 'main-fat3',
  #         'target': '10.5',
  #         'cflags': '-isysroot / -arch i386 -arch ppc -arch x86_64',
  #         'cc': 'gcc-4.2',
  #     },
  #     {
  #         'name': 'main-intel',
  #         'target': '10.5',
  #         'cflags': '-isysroot / -arch i386 -arch x86_64',
  #         'cc': 'gcc-4.2',
  #     },
  #     {
  #         'name': 'main-fat',
  #         'target': '10.3',
  #         'cflags': '-isysroot /Developer/SDKs/MacOSX10.4u.sdk -arch i386 -arch ppc',
  #         'cc': 'gcc-4.0',
  #     },
  ]

using homebrew

• another route is homebrew: http://gist.github.com/519418 /

reference

• Christopher D. Fiorillo. A neurocentric approach to Bayesian inference, URL . Nature Reviews Neuroscience, 11(8):605, 2010

  A primary function of the brain is to infer the state of the world in order to determine which motor behaviours will best promote adaptive fitness. Bayesian probability theory formally describes how rational inferences ought to be made, and it has been used with great success in recent years to explain a range of perceptual and sensorimotor phenomena1, 2, 3, 4, 5. .

• Karl Friston. Is the free-energy principle neurocentric?, URL . Nature Reviews Neuroscience, 11(8):605, 2010

  Recently, a free-energy formulation of brain function was reviewed in relation to several other neurobiological theories (The free-energy principle: a unified brain theory? Nature Rev. Neurosci. 11, 127–138 (2010) .

a bunch of existing tools

• http://sourceforge.net/projects/mussh/

• http://code.google.com/p/csshx/

• http://www.occam.com/sa/rshall.pdf

• http://web.taranis.org/shmux/#related

• http://tentakel.biskalar.de/similar/

• http://guichaz.free.fr/gsh/ (sse btw http://guichaz.free.fr/pysize/ )

what we can do

using macports

• it works now with macports:

  sudo port install -u opencv +python26 +tbb

latest SVN

• compiling here along with MacTex...

• from http://opencv.willowgarage.com/wiki/Mac_OS_X_OpenCV_Port

  svn co https://code.ros.org/svn/opencv/trunk/opencv
  cd opencv # the directory containing INSTALL, CMakeLists.txt etc.
  mkdir build
  cd build
  cmake -D CMAKE_OSX_ARCHITECTURES=x86_64 -D WITH_FFMPEG=ON -D BUILD_EXAMPLES=ON -D BUILD_LATEX_DOCS=ON -D PDFLATEX_COMPILER=/usr/texbin/pdflatex -D BUILD_NEW_PYTHON_SUPPORT=ON  -D PYTHON_LIBRARY=/opt/local/lib/libpython2.6.dylib -D PYTHON_INCLUDE_DIR=/opt/local/Library/Frameworks/Python.framework/Headers ..
  make -j4
  sudo make install

• I had to rebuild some ports

  sudo port install ilmbase
  port provides /opt/local/lib/libIlmImf.dylib
  sudo port install openexr
  sudo port install libdc1394

  and recompile

• then could run

  cd ../samples/python/
  python camera.py

using homebrew

• another route is homebrew: http://gist.github.com/519418 / :

  $ brew info opencv
  opencv 2.1.1-pre
  http://opencv.willowgarage.com/wiki/
  Depends on: cmake, pkg-config, libtiff, jasper, tbb
  /usr/local/Cellar/opencv/2.1.1-pre (96 files, 37M)
  
  The OpenCV Python module will not work until you edit your PYTHONPATH like so:
    export PYTHONPATH="/usr/local/lib/python2.6/site-packages/:$PYTHONPATH"
  
  To make this permanent, put it in your shell's profile (e.g. ~/.profile).
  
  http://github.com/mxcl/homebrew/commits/master/Library/Formula/opencv.rb

examples

This is a red square:

\usepackage{graphics,color}

%%end-prologue%%
\newsavebox{\mysquare}
\savebox{\mysquare}{\textcolor{red}{\rule{1in}{1in} } }
\usebox{\mysquare}

symboles

% Math-mode symbol & verbatim
\def\W#1#2{$#1{#2}$ &\tt\string#1\string{#2\string}}
\def\X#1{$#1$ &\tt\string#1}
\def\Y#1{$\big#1$ &\tt\string#1}
\def\Z#1{\tt\string#1}

% A non-floating table environment.
\makeatletter
\renewenvironment{table}%
   {\vskip\intextsep\parskip\z@
    \vbox\bgroup\centering\def\@captype{table}}%
   {\egroup\vskip\intextsep}
\makeatother

% All the tables are \label'ed in case this document ever gets some
% explanatory text written, however there are no \refs as yet. To save
% LaTeX-ing the file twice we go:
\renewcommand{\label}[1]{}

%%end-prologue%%
\begin{table}
\begin{tabular}{*8l}
\X\alpha        &\X\theta       &\X o           &\X\tau         \\
\X\beta         &\X\vartheta    &\X\pi          &\X\upsilon     \\
\X\gamma        &\X\gamma       &\X\varpi       &\X\phi         \\
\X\delta        &\X\kappa       &\X\rho         &\X\varphi      \\
\X\epsilon      &\X\lambda      &\X\varrho      &\X\chi         \\
\X\varepsilon   &\X\mu          &\X\sigma       &\X\psi         \\
\X\zeta         &\X\nu          &\X\varsigma    &\X\omega       \\
\X\eta          &\X\xi                                          \\
                                                                \\
\X\Gamma        &\X\Lambda      &\X\Sigma       &\X\Psi         \\
\X\Delta        &\X\Xi          &\X\Upsilon     &\X\Omega       \\
\X\Theta        &\X\Pi          &\X\Phi
\end{tabular}
\caption{Greek Letters}\label{greek}
\end{table}

ou

\begin{equation}
x^3 =\int_{0}^{\infty} f(x,y) dy
\end{equation}

• et encore

  $$x^3 =\int_{0}^{\infty} f(x,y) dy + c$$

inline

Because people requested an easier way to enter latex, I've added the possibility to write $ ... $ to obtain inline formulas. This is equivalent to writing \$ ...\$ and has the same single-line limitation (but everything else isn't really useful in formulas anyway). In order to do this, install the inline\_latex.py parser add #format inline\_latex to your page (alternatively, configure the default parser to be ``inline\_latex). This parser accepts all regular wiki syntax, but additionally the $ ... $' syntax. Additionally, the ``inline_latex` formatter supports $$....$$ style formulas (still limited to a single line though!) which puts the formula into a paragraph on its own.

Note: in the nikola blog, this is directly accomplished by using ReST : \$\\lambda\$ = $lambda$

installing SUMATRA

• notes from the CodeJamNr4 from FACETS

• this was using a fresh install of ETS 6.2

525  svn export ../sci/dyva/Motion/particles hg_particles
526  cd hg_particles/
527  hg init
528  hg add MotionParticles.py experiment_all.py
529  hg commit
530  hg commit -m 'test'
531  echo $USER
532  vim .hgrc
533  vim ~/.hgrc
534  hg commit -m 'my first HG commit'
535  vim ~/.hgrc
536  ipython
537  ls
538  smt init sumatraTest_hg
539  smt info

 501  cd sci/dyva/Motion/particles/
 502  smt init -h
 503  smt init sumatraTest
 504  smt info
511  smt configure --simulator=python --main=experiment_all.py
 512  smt info
 513  smtweb &
 514  ls -a
 515  rm -fr .smt
 516  smt init sumatraTest
 517  smtweb &
 518  open experiment_all.py
 519  touch fake.param
 520  smt run
 521  smt run -s python -m experiment_dot.py fake.param
 522  smt info
 523  smt configure -h
 524  smt configure -c diff
 525  smt info
 526  smt run -s python -m experiment_dot.py fake.param
 529  smt run -s python -m experiment_dot.py fake.param
 534  rm mat/dot.npy
 535  python experiment_dot.py fake.param
 536  ls
 537  smt help configure
 538  smt configure -d ./figures/
 539  smt info
 540  smt configure -s python -m experiment_dot.py
 541  smt run fake.param
 542  rm mat/dot.npy
 543  smt run fake.param
 544  ls figures/
 545  rm figures/dot_*
 546  smt run fake.param
 547  smt info
 548  smt configure -d ./figures
 549  smt info
 550  rm figures/dot_*png
 551  smt configure -d ./figures
 552  smt run fake.param
 553  smt comment "apparently, it is worth NaN shekels."
 554  smt tag codejam
 558  rm figures/dot_*png
 559  rm mat/dot.npy
 560  smt run --reason="test effect of a bigger dot" fake.param dot_size=0.1
 561  ls
 562  ls -al .smt/
 563  less .smt/simulation_records
 564  sqlite3 .smt/simulation_records

The reStructuredText Cheat Sheet: Syntax Reminders

Info

See <http://docutils.sf.net/rst.html> for introductory docs.

Author

David Goodger <goodger@python.org>

Date

$Date: 2006-01-23 02:13:55 +0100 (Mon, 23 Jän 2006) $

Revision

$Revision: 4321 $

Description

This is a "docinfo block", or bibliographic field list

Body Elements

Grid table:

Paragraphs are flush-left, separated by blank lines. Block quotes are indented.	Literal block, preceded by "::": Indented or: > Quoted
>>> print 'Doctest block' Doctest block
Line blocks preserve line breaks & indents. [new in 0.3.6] Useful for addresses, verse, and adornment-free lists; long lines can be wrapped with continuation lines.

Simple tables:

List Type	Examples
Bullet list	items begin with "-", "+", or "*"
Enumerated list	items use any variation of "1.", "A)", and "(i)" also auto-enumerated
Definition list	Term is flush-leftoptional classifier Definition is indented, no blank line between
Field list	field name field body
Option list	-o at least 2 spaces between option & description

Explicit Markup	Examples (visible in the text source)
Footnote	1 Manually numbered or [#] auto-numbered (even [#labelled]) or [*] auto-symbol
Citation	CIT2002 A citation.
Hyperlink Target
Anonymous Target
Directive ("::")
Substitution Def
Comment
Empty Comment	(".." on a line by itself, with blank lines before & after, used to separate indentation contexts)

using sed

• The UNIX command sed is useful to find and replace text in single or multiple files. This page lists some common commands in using sed to improve editing code.

• To replace foo with foo_bar in a single file:

  sed -i 's/foo/foo_bar/g' my_script.py

  • -i = edit the file "in-place": sed will directly modify the file if it finds anything to replace

  • s = substitute the following text

  • foo = the text string to be substituted

  • foo_bar = the replacement string

  • g = global, match all occurrences in the line

• To replace foo with foo_bar in multiple files:

  sed -i 's/foo/foo_bar/g'  *.py

• Consult the manual pages of the operating system that you use: man sed

• in the particular case of changing a scaling parameter in a set of experiment files:

  sed -i 's/size = 6/size = 7/g'  experiment*.py
  sed -i 's/size = 7/size = 6/g'  experiment*.py

using vim

• on the current buffer, with confirmation

  :%s/old_text/new_text/cg

• on the current buffer

  :%s/old_text/new_text/g

• to get help

  :help substitute

• one could pass the required files to 'args' and apply whatever command to all these files using the command 'argdo'. First I will apply the substitute 's' command and then 'update' which will only save the modified files.

  :args *.py
  :argdo :%s/old_text/new_text/g | update

using python

• http://muharem.wordpress.com/2007/05/20/python-find-files-using-unix-shell-style-wildcards/

La création de tableaux de données Numpy

Un petit exemple pour commencer:

>>> import numpy as np
>>> a = np.array([0, 1, 2])
>>> a
array([0, 1, 2])
>>> print a
[0 1 2]
>>> b = np.array([[0., 1.], [2., 3.]])
>>> b
array([[ 0.,  1.],
       [ 2.,  3.]])

Dans la pratique, on rentre rarement les éléments un par un...

• Valeurs espacées régulièrement:

  >>> import numpy as np
  >>> a = np.arange(10) # de 0 a n-1
  >>> a
  array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])
  >>> b = np.arange(1., 9., 2) # syntaxe : debut, fin, saut
  >>> b
  array([ 1.,  3.,  5.,  7.])

  ou encore, en spécifiant le nombre de points:

  >>> c = np.linspace(0, 1, 6)
  >>> c
  array([ 0. ,  0.2,  0.4,  0.6,  0.8,  1. ])
  >>> d = np.linspace(0, 1, 5, endpoint=False)
  >>> d
  array([ 0. ,  0.2,  0.4,  0.6,  0.8])

• Constructeurs pour des tableaux classiques:

  >>> a = np.ones((3,3))
  >>> a
  array([[ 1.,  1.,  1.],
         [ 1.,  1.,  1.],
         [ 1.,  1.,  1.]])
  >>> a.dtype
  dtype('float64')
  >>> b = np.ones(5, dtype=np.int)
  >>> b
  array([1, 1, 1, 1, 1])
  >>> c = np.zeros((2,2))
  >>> c
  array([[ 0.,  0.],
         [ 0.,  0.]])
  >>> d = np.eye(3)
  >>> d
  array([[ 1.,  0.,  0.],
         [ 0.,  1.,  0.],
         [ 0.,  0.,  1.]])

La représentation graphique des données : matplotlib et mayavi

Maintenant que nous avons nos premiers tableaux de données, nous allons les visualiser. Matplotlib est un package de plot 2-D, on importe ces fonctions de la manière suivante

>>> import pylab
>>> # ou
>>> from pylab import * # pour tout importer dans le namespace

Si vous avec lancé Ipython avec python(x,y), ou avec l'option ipython -pylab (sous linux), toutes les fonctions/objets de pylab ont déjà été importées, comme si on avait fait from pylab import *. Dans la suite on suppose qu'on a fait from pylab import * ou lancé ipython -pylab: on n'écrira donc pas pylab.fonction() mais directement fonction.

Tracé de courbes 1-D

In [6]: a = np.arange(20)
In [7]: plot(a, a**2) # line plot
Out[7]: [<matplotlib.lines.Line2D object at 0x95abd0c>]
In [8]: plot(a, a**2, 'o') # symboles ronds
Out[8]: [<matplotlib.lines.Line2D object at 0x95b1c8c>]
In [9]: clf() # clear figure
In [10]: loglog(a, a**2)
Out[10]: [<matplotlib.lines.Line2D object at 0x95abf6c>]
In [11]: xlabel('x') # un peu petit
Out[11]: <matplotlib.text.Text object at 0x98923ec>
In [12]: xlabel('x', fontsize=26) # plus gros
Out[12]: <matplotlib.text.Text object at 0x98923ec>
In [13]: ylabel('y')
Out[13]: <matplotlib.text.Text object at 0x9892b8c>
In [14]: grid()
In [15]: axvline(2)
Out[15]: <matplotlib.lines.Line2D object at 0x9b633cc>

Tableaux 2-D (images par exemple)

In [48]: # Tableaux 30x30 de nombres aleatoires entre 0 et 1
In [49]: image = np.random.rand(30,30)
In [50]: imshow(image)
Out[50]: <matplotlib.image.AxesImage object at 0x9e954ac>
In [51]: gray()
In [52]: hot()
In [53]: imshow(image, cmap=cm.gray)
Out[53]: <matplotlib.image.AxesImage object at 0xa23972c>
In [54]: axis('off') # on enleve les ticks et les labels

Il y a bien d'autres fonctionnalités dans matplotlib : choix de couleurs ou des tailles de marqueurs, fontes latex, inserts à l'intérieur d'une figure, histogrammes, etc.

Pour aller plus loin :

• la documentation de matplotlib http://matplotlib.sourceforge.net/contents.html

• une gallerie d'exemples accompagnés du code source http://matplotlib.sourceforge.net/gallery.html

Représentation en 3-D

Pour la visualisation 3-D, on utilise un autre package : Mayavi. Un exemple rapide : commencez par relancer ipython avec les options ipython -pylab -wthread

In [59]: from enthought.mayavi import mlab
In [60]: mlab.figure()
get fences failed: -1
param: 6, val: 0
Out[60]: <enthought.mayavi.core.scene.Scene object at 0xcb2677c>
In [61]: mlab.surf(image)
Out[61]: <enthought.mayavi.modules.surface.Surface object at 0xd0862fc>
In [62]: mlab.axes()
Out[62]: <enthought.mayavi.modules.axes.Axes object at 0xd07892c>

La fenêtre mayavi/mlab qui s'ouvre est interactive : en cliquant sur le bouton gauche de la souris vous pouvez faire tourner l'image, on peut zoomer avec la molette, etc.

Pour plus d'informations sur Mayavi : http://code.enthought.com/projects/mayavi/docs/development/html/mayavi/index.html

Indexage

On peut accéder aux éléments des tableaux Numpy (indexer) d'une manière similaire que pour les autres séquences Python (list, tuple)

>>> a = np.arange(10)
>>> a
array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])
>>> a[0], a[2], a[-1]
(0, 2, 9)

Attention ! L'indexage commence à partir de 0, comme pour les autres séquences Python (et comme en C/C++). En Fortran ou Matlab, l'indexage commence à 1.

Pour les tableaux multidimensionnels, l'indice d'un élément est donné par un n-uplet d'entiers

>>> a = np.diag(np.arange(5))
>>> a
array([[0, 0, 0, 0, 0],
       [0, 1, 0, 0, 0],
       [0, 0, 2, 0, 0],
       [0, 0, 0, 3, 0],
       [0, 0, 0, 0, 4]])
>>> a[1,1]
1
>>> a[2,1] = 10 # deuxième ligne, première colonne
>>> a
array([[ 0,  0,  0,  0,  0],
       [ 0,  1,  0,  0,  0],
       [ 0, 10,  2,  0,  0],
       [ 0,  0,  0,  3,  0],
       [ 0,  0,  0,  0,  4]])
>>> a[1]
array([0, 1, 0, 0, 0])

A retenir :

• En 2-D, la première dimension correspond aux lignes, la seconde aux colonnes.

• Pour un tableau a à plus qu'une dimension,`a[0]` est interprété en prenant tous les éléments dans les dimensions non-spécifiés.

Slicing (parcours régulier des éléments)

Comme l'indexage, similaire au slicing des autres séquences Python:

>>> a = np.arange(10)
>>> a
array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])
>>> a[2:9:3] # [début:fin:pas]
array([2, 5, 8])

Attention, le dernier indice n'est pas inclus

>>> a[:4]
array([0, 1, 2, 3])

début:fin:pas est un objet slice, qui représente l'ensemble d'indices range(début, fin, pas). On peut créer explicitement un slice

>>> sl = slice(1, 9, 2)
>>> a = np.arange(10)
>>> b = 2*a + 1
>>> a, b
(array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9]), array([ 1,  3,  5,  7,  9, 11, 13, 15, 17, 19]))
>>> a[sl], b[sl]
(array([1, 3, 5, 7]), array([ 3,  7, 11, 15]))

On n'est pas obligé de spécifier à la fois le début (indice 0 par défaut), la fin (dernier indice par défaut) et le pas (1 par défaut):

>>> a[1:3]
array([1, 2])
>>> a[::2]
array([0, 2, 4, 6, 8])
>>> a[3:]
array([3, 4, 5, 6, 7, 8, 9])

Et bien sûr, ça marche pour les tableaux à plusieurs dimensions:

>>> a = np.eye(5)
>>> a
array([[ 1.,  0.,  0.,  0.,  0.],
       [ 0.,  1.,  0.,  0.,  0.],
       [ 0.,  0.,  1.,  0.,  0.],
       [ 0.,  0.,  0.,  1.,  0.],
       [ 0.,  0.,  0.,  0.,  1.]])
>>> a[2:4,:3] #2è et 3è lignes, trois premières colonnes
array([[ 0.,  0.,  1.],
       [ 0.,  0.,  0.]])

On peut changer la valeur de tous les éléments indexés par une slice de façon très simple

>>> a[:3,:3] = 4
>>> a
array([[ 4.,  4.,  4.,  0.,  0.],
       [ 4.,  4.,  4.,  0.,  0.],
       [ 4.,  4.,  4.,  0.,  0.],
       [ 0.,  0.,  0.,  1.,  0.],
       [ 0.,  0.,  0.,  0.,  1.]])

Une opération de slicing crée une vue (view) du tableau d'origine, c'est-à-dire une manière d'aller lire dans la mémoire. Le tableau d'origine n'est donc pas copié. Quand on modifie la vue, on modife aussi le tableau d'origine.:

>>> a = np.arange(10)
>>> a
array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])
>>> b = a[::2]; b
array([0, 2, 4, 6, 8])
>>> b[0] = 12
>>> b
array([12,  2,  4,  6,  8])
>>> a # a a été modifié aussi !
array([12,  1,  2,  3,  4,  5,  6,  7,  8,  9])

Ce comportement peut surprendre au début... mais est bien pratique pour gérer la mémoire de façon économe.

Si on veut faire une copie différente du tableau d'origine

>>> a = np.arange(10)
>>> a
array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])
>>> b = np.copy(a[::2]); b
array([0, 2, 4, 6, 8])
>>> b[0] = 12
>>> b
array([12,  2,  4,  6,  8])
>>> a # a n'a pas été modifié
array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])

Manipuler la forme des tableaux

On obtient la forme d'un tableau grâce à la méthode ndarray.shape qui retourne un tuple des dimensions du tableau

>>> a = np.arange(10)
>>> a.shape
(10,)
>>> b = np.ones((3,4))
>>> b.shape
(3, 4)
>>> b.shape[0] # on peut accéder aux élements du tuple b.shape
3
>>> # et on peut aussi faire
>>> np.shape(b)
(3, 4)

Par ailleurs on obtient la longueur de la première dimension avec np.alen (par analogie avec len pour une liste) et le nombre total d'éléments avec ndarray.size:

>>> np.alen(b)
3
>>> b.size
12

Il existe plusieurs fonctions Numpy qui permettent de créer un tableau de taille différente à partir d'un tableau de départ.:

>>> a = np.arange(36)
>>> b = a.reshape((6, 6))
>>> b
array([[ 0,  1,  2,  3,  4,  5],
       [ 6,  7,  8,  9, 10, 11],
       [12, 13, 14, 15, 16, 17],
       [18, 19, 20, 21, 22, 23],
       [24, 25, 26, 27, 28, 29],
       [30, 31, 32, 33, 34, 35]])

ndarray.reshape renvoie une vue, et pas une copie

>>> b[0,0] = 10
>>> a
array([10,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15, 16,
       17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33,
       34, 35])

On peut aussi créer un tableau avec un nombre d'éléments différents avec ndarray.resize:

>>> a = np.arange(36)
>>> a.resize((4,2))
>>> a
array([[0, 1],
       [2, 3],
       [4, 5],
       [6, 7]])
>>> b = np.arange(4)
>>> b.resize(3, 2)
>>> b
array([[0, 1],
       [2, 3],
       [0, 0]])

Ou paver un grand tableau à partir d'un tableau plus petit

>>> a = np.arange(4).reshape((2,2))
>>> a
array([[0, 1],
       [2, 3]])
>>> np.tile(a, (2,3))
array([[0, 1, 0, 1, 0, 1],
       [2, 3, 2, 3, 2, 3],
       [0, 1, 0, 1, 0, 1],
       [2, 3, 2, 3, 2, 3]])

Exercices : quelques gammes avec les tableaux numpy

Grâce aux divers constructeurs, à l'indexage et au slicing, et aux opérations simples sur les tableaux (+/-/x/:), on peut facilement créer des tableaux de grande taille correspondant à des motifs variés.

Exemple : comment créer le tableau:

[[ 0  1  2  3  4]
 [ 5  6  7  8  9]
 [10 11 12 13  0]
 [15 16 17 18 19]
 [20 21 22 23 24]]

Réponse

>>> a = np.arange(25).reshape((5,5))
>>> a[2, 4] = 0

Exercices : créer les tableaux suivants de la manière la plus simple possible (pas élément par élement)

[[ 1.  1.  1.  1.]
 [ 1.  1.  1.  1.]
 [ 1.  1.  1.  2.]
 [ 1.  6.  1.  1.]]

[[0 0 0 0 0]
 [2 0 0 0 0]
 [0 3 0 0 0]
 [0 0 4 0 0]
 [0 0 0 5 0]
 [0 0 0 0 6]]

De "vraies données" : lire et écrire des tableaux dans des fichiers

Bien souvent, nos expériences ou nos simulations écrivent leurs résultats dans des fichiers. Il faut ensuite les charger dans Python sous la forme de tableaux Numpy pour les manipuler. De même, on peut vouloir sauver les tableaux qu'on a obtenus dans des fichiers.

Aller dans le bon répertoire

Pour se déplacer dans une arborescence de fichiers :

• utiliser les fonctionnalités d'Ipython : cd, pwd, tab-completion.

• modules os (routines système) et os.path (gestion des chemins)

  >>> import os, os.path
  >>> current_dir = os.getcwd()
  >>> current_dir
  '/home/gouillar/sandbox'
  >>> data_dir = os.path.join(current_dir, 'data')
  >>> data_dir
  '/home/gouillar/sandbox/data'
  >>> if not(os.path.exists(data_dir)):
  ...     os.mkdir('data')
  ...     print "creation du repertoire 'data'"
  ...
  >>> os.chdir(data_dir) # ou dans Ipython : cd data

On peut en fait se servir de Ipython comme d'un véritable shell grâce aux fonctionnalités d'Ipython et au module os.

Ecrire un tableau de données dans un fichier

>>> a = np.arange(100)
>>> a = a.reshape((10, 10))

• Ecriture dans un fichier texte (en ascii)

  >>> np.savetxt('data_a.txt', a)

• Ecriture dans un fichier en binaire (extension .npy)

  >>> np.save('data_a.npy', a)

Charger un tableau de données à partir d'un fichier

• Lecture dans un fichier texte

  >>> b = np.loadtxt('data_a.txt')

• Lecture dans un fichier binaire

  >>> c = np.load('data_a.npy')

Pour sélectionner un fichier dans une liste

On va sauver chaque ligne de a dans un fichier différent

>>> for i, l in enumerate(a):
...     print i, l
...     np.savetxt('ligne_'+str(i), l)
...
0 [0 1 2 3 4 5 6 7 8 9]
1 [10 11 12 13 14 15 16 17 18 19]
2 [20 21 22 23 24 25 26 27 28 29]
3 [30 31 32 33 34 35 36 37 38 39]
4 [40 41 42 43 44 45 46 47 48 49]
5 [50 51 52 53 54 55 56 57 58 59]
6 [60 61 62 63 64 65 66 67 68 69]
7 [70 71 72 73 74 75 76 77 78 79]
8 [80 81 82 83 84 85 86 87 88 89]
9 [90 91 92 93 94 95 96 97 98 99]

Pour obtenir une liste de tous les fichiers commençant par ligne, on fait appel au module glob qui "gobe" tous les chemins correspondant à un motif. Exemple

>>> import glob
>>> filelist = glob.glob('ligne*')
>>> filelist
['ligne_0', 'ligne_1', 'ligne_2', 'ligne_3', 'ligne_4', 'ligne_5', 'ligne_6', 'ligne_7', 'ligne_8', 'ligne_9']
>>> # attention la liste n'est pas toujours ordonnee
>>> filelist.sort()
>>> l2 = np.loadtxt(filelist[2])

Remarque : il est aussi possible de créer des tableaux à partir de fichiers Excel/Calc, de fichiers hdf5, etc. (mais à l'aide de modules supplémentaires non décrits ici : xlrd, pytables, etc.).

Opérations mathématiques et statistiques simples sur les tableaux

Un certain nombre d'opérations sur les tableaux sont codées directement dans numpy (et sont donc en général très efficaces):

>>> a = np.arange(10)
>>> a.min() # ou np.min(a)
0
>>> a.max() # ou np.max(a)
9
>>> a.sum() # ou np.sum(a)
45

Il est possible de réaliser l'opération le long d'un axe uniquement, plutôt que sur tous les éléments

>>> a = np.array([[1, 3], [9, 6]])
>>> a
array([[1, 3],
       [9, 6]])
>>> a.mean(axis=0) # tableau contenant la moyenne de chaque colonne
array([ 5. ,  4.5])
>>> a.mean(axis=1) # tableau contenant la moyenne de chaque ligne
array([ 2. ,  7.5])

Il y en a encore bien d'autres opérations possibles : on en découvrira quelques unes au fil de ce cours.

>>> a = np.ones((2,2))
>>> a*a
array([[ 1.,  1.],
       [ 1.,  1.]])
>>> np.dot(a,a)
array([[ 2.,  2.],
       [ 2.,  2.]])

Exemple : simulation de diffusion avec un marcheur aléatoire

Quelle est la distance typique d'un marcheur aléatoire à l'origine, après t sauts à droite ou à gauche ?

>>> nreal = 1000 # nombre de réalisations de la marche
>>> tmax = 200 # temps sur lequel on suit le marcheur
>>> # On tire au hasard tous les pas 1 ou -1 de la marche
>>> walk = 2 * ( np.random.random_integers(0, 1, (nreal,tmax)) - 0.5 )
>>> np.unique(walk) # Vérification : tous les pas font bien 1 ou -1
array([-1.,  1.])
>>> # On construit les marches en sommant ces pas au cours du temps
>>> cumwalk = np.cumsum(walk, axis=1) # axis = 1 : dimension du temps
>>> sq_distance = cumwalk**2
>>> # On moyenne dans le sens des réalisations
>>> mean_sq_distance = np.mean(sq_distance, axis=0)

In [39]: figure()
In [40]: plot(mean_sq_distance)
In [41]: figure()
In [42]: plot(np.sqrt(mean_sq_distance))

On retrouve bien que la distance grandit comme la racine carrée du temps !

Exercice : statistiques des femmes dans la recherche (données INSEE)

1. Récupérer les fichiers organismes.txt et pourcentage_femmes.txt (clé USB du cours ou http://www.dakarlug.org/pat/scientifique/data/).

2. Créer un tableau data en ouvrant le fichier pourcentage_femmes.txt avec np.loadtxt. Quelle est la taille de ce tableau ?

3. Les colonnes correspondent aux années 2006 à 2001. Créer un tableau annees (sans accent !) contenant les entiers correspondant à ces années.

4. Les différentes lignes correspondent à différents organismes de recherche dont les noms sont stockés dans le fichier organismes.txt. Créer un tableau organisms en ouvrant ce fichier. Attention : comme np.loadtxt crée par défaut des tableaux de flottant, il faut lui préciser qu'on veut créer un tableau de strings : organisms = np.loadtxt('organismes.txt', dtype=str)

5. Vérifier que le nombre de lignes de data est égal au nombre d'organismes.

6. Quel est le pourcentage maximal de femmes dans tous les organismes, toutes années confondues ?

7. Créer un tableau contenant la moyenne temporelle du pourcentage de femmes pour chaque organisme (i.e., faire la moyenne de data suivant l'axe No 1).

8. Quel organisme avait le pourcentage de femmes le plus élevé en 2004 ? (Indice np.argmax).

9. Représenter un histogramme du pourcentage de femmes dans les

   différents organismes en 2006 (indice : np.histogram, puis bar ou plot de matplotlib pour la visualisation).

L'indexage avancé (fancy indexing)

On peut indexer des tableaux numpy avec des slices, mais aussi par des tableaux de booléens (les masques) ou d'entiers : on appelle ces opérations plus évoluées du fancy indexing.

Les masques

>>> np.random.seed(3)
>>> a = np.random.random_integers(0, 20, 15)
>>> a
array([10,  3,  8,  0, 19, 10, 11,  9, 10,  6,  0, 20, 12,  7, 14])
>>> (a%3 == 0)
array([False,  True, False,  True, False, False, False,  True, False,
        True,  True, False,  True, False, False], dtype=bool)
>>> mask = (a%3 == 0)
>>> extract_from_a = a[mask] #on pourrait écrire directement a[a%3==0]
>>> extract_from_a #on extrait un sous-tableau grâce au masque
array([ 3,  0,  9,  6,  0, 12])

Extraire un sous-tableau avec un masque produit une copie de ce sous-tableau, et non une vue

>>> extract_from_a = -1
>>> a
array([10,  3,  8,  0, 19, 10, 11,  9, 10,  6,  0, 20, 12,  7, 14])

L'indexation grâce masques peut être très utile pour l'assignation d'une nouvelle valeur à un sous-tableau

>>> a[mask] = 0
>>> a
array([10,  0,  8,  0, 19, 10, 11,  0, 10,  0,  0, 20,  0,  7, 14])

Indexer avec un tableau d'entiers

>>> a = np.arange(10)
>>> a[::2] +=3 #pour ne pas avoir toujours le même np.arange(10)...
>>> a
array([ 3,  1,  5,  3,  7,  5,  9,  7, 11,  9])
>>> a[[2, 5, 1, 8]] # ou a[np.array([2, 5, 1, 8])]
array([ 5,  5,  1, 11])

On peut indexer avec des tableaux d'entiers où le même indice est répété plusieurs fois

>>> a[[2, 3, 2, 4, 2]]
array([5, 3, 5, 7, 5])

On peut assigner de nouvelles valeurs avec ce type d'indexation

>>> a[[9, 7]] = -10
>>> a
array([  3,   1,   5,   3,   7,   5,   9, -10,  11, -10])
>>> a[[2, 3, 2, 4, 2]] +=1
>>> a
array([  3,   1,   6,   4,   8,   5,   9, -10,  11, -10])

Quand on crée un tableau en indexant avec un tableau d'entiers, le nouveau tableau a la même forme que le tableau d'entiers

>>> a = np.arange(10)
>>> idx = np.array([[3, 4], [9, 7]])
>>> a[idx]
array([[3, 4],
       [9, 7]])
>>> b = np.arange(10)

>>> a = np.arange(12).reshape(3,4)
>>> a
array([[ 0,  1,  2,  3],
       [ 4,  5,  6,  7],
       [ 8,  9, 10, 11]])
>>> i = np.array( [ [0,1],
...              [1,2] ] )
>>> j = np.array( [ [2,1],
...              [3,3] ] )
>>> a[i,j]
array([[ 2,  5],
       [ 7, 11]])

Exercice

Reprenons nos données de statistiques du pourcentage de femmes dans la recherche (tableaux data et organisms)

1. Créer un tableau sup30 de même taille que data valant 1 si la valeur de data est supérieure à 30%, et 0 sinon.

2. Créez un tableau contenant l'organisme avec le pourcentage de femmes le plus élévé de chaque année.

Le broadcasting

Les opérations élémentaires sur les tableaux numpy (addition, etc.) sont faites élément par élément et opèrent donc des tableaux de même taille. Il est néanmoins possible de faire des opérations sur des tableaux de taille différente si numpy` arrive à transformer les tableaux pour qu'ils aient tous la même taille : on appelle cette transformation le broadcasting (jeu de mots intraduisible en français).

L'image ci-dessous donne un exemple de broadcasting

ce qui donne dans Ipython:

>>> a = np.arange(0, 40, 10)
>>> b = np.arange(0, 3)
>>> a = a.reshape((4,1)) #il faut transformer a en tableau "vertical"
>>> a + b
array([[ 0,  1,  2],
       [10, 11, 12],
       [20, 21, 22],
       [30, 31, 32]])

On a déjà utilisé le broadcasting sans le savoir

>>> a = np.arange(20).reshape((4,5))
>>> a
array([[ 0,  1,  2,  3,  4],
       [ 5,  6,  7,  8,  9],
       [10, 11, 12, 13, 14],
       [15, 16, 17, 18, 19]])
>>> a[0] = 1 # on égale deux tableaux de dimension 1 et 0
>>> a[:3] = np.arange(1,6)
>>> a
array([[ 1,  2,  3,  4,  5],
       [ 1,  2,  3,  4,  5],
       [ 1,  2,  3,  4,  5],
       [15, 16, 17, 18, 19]])

On peut même utiliser en même temps le fancy indexing et le broadcasting : reprenons un exemple déjà utilisé plus haut

>>> a = np.arange(12).reshape(3,4)
>>> a
array([[ 0,  1,  2,  3],
       [ 4,  5,  6,  7],
       [ 8,  9, 10, 11]])
>>> i = np.array( [ [0,1],
...              [1,2] ] )
>>> a[i, 2] # même chose que a[i, 2*np.ones((2,2), dtype=int)]
array([[ 2,  6],
       [ 6, 10]])

Le broadcasting peut sembler un peu magique, mais il est en fait assez naturel de l'utiliser dès qu'on veut veut résoudre un problème où on obtient en sortie un tableau avec plus de dimensions que les données en entrée.

Exemple : construisons un tableau de distances (en miles) entre les villes de la route 66 : Chicago, Springfield, Saint-Louis, Tulsa, Oklahoma City, Amarillo, Santa Fe, Albucquerque, Flagstaff et Los Angeles.

>>> mileposts = np.array([0, 198, 303, 736, 871, 1175, 1475, 1544,
...                         1913, 2448])
>>> tableau_de_distances = np.abs(mileposts - mileposts[:,np.newaxis])
>>> tableau_de_distances
array([[   0,  198,  303,  736,  871, 1175, 1475, 1544, 1913, 2448],
       [ 198,    0,  105,  538,  673,  977, 1277, 1346, 1715, 2250],
       [ 303,  105,    0,  433,  568,  872, 1172, 1241, 1610, 2145],
       [ 736,  538,  433,    0,  135,  439,  739,  808, 1177, 1712],
       [ 871,  673,  568,  135,    0,  304,  604,  673, 1042, 1577],
       [1175,  977,  872,  439,  304,    0,  300,  369,  738, 1273],
       [1475, 1277, 1172,  739,  604,  300,    0,   69,  438,  973],
       [1544, 1346, 1241,  808,  673,  369,   69,    0,  369,  904],
       [1913, 1715, 1610, 1177, 1042,  738,  438,  369,    0,  535],
       [2448, 2250, 2145, 1712, 1577, 1273,  973,  904,  535,    0]])

Beaucoup de problèmes sur grille ou réseau peuvent aussi utiliser du broadcasting. Par exemple, si on veut calculer la distance à l'origine des points sur une grille 10x10, on peut faire

>>> x, y = np.arange(5), np.arange(5)
>>> distance = np.sqrt(x**2 + y[:, np.newaxis]**2)
>>> distance
array([[ 0.        ,  1.        ,  2.        ,  3.        ,  4.        ],
       [ 1.        ,  1.41421356,  2.23606798,  3.16227766,  4.12310563],
       [ 2.        ,  2.23606798,  2.82842712,  3.60555128,  4.47213595],
       [ 3.        ,  3.16227766,  3.60555128,  4.24264069,  5.        ],
       [ 4.        ,  4.12310563,  4.47213595,  5.        ,  5.65685425]])

On peut représenter les valeurs du tableau distance en niveau de couleurs grâce à la fonction pylab.imshow (syntaxe : pylab.imshow(distance). voir l'aide pour plus d'options).

Remarque : la fonction numpy.ogrid permet de créer directement les vecteurs x et y de l'exemple précédent avec deux "dimensions significatives" différentes

>>> x, y = np.ogrid[0:5, 0:5]
>>> x, y
(array([[0],
       [1],
       [2],
       [3],
       [4]]), array([[0, 1, 2, 3, 4]]))
>>> x.shape, y.shape
((5, 1), (1, 5))
>>> distance = np.sqrt(x**2 + y**2)

np.ogrid est donc très utile dès qu'on a des calculs à faire sur un réseau. np.mgrid fournit par contre directement des matrices pleines d'indices pour les cas où on ne peut/veut pas profiter du broadcasting

>>> x, y = np.mgrid[0:4, 0:4]
>>> x
array([[0, 0, 0, 0],
       [1, 1, 1, 1],
       [2, 2, 2, 2],
       [3, 3, 3, 3]])
>>> y
array([[0, 1, 2, 3],
       [0, 1, 2, 3],
       [0, 1, 2, 3],
       [0, 1, 2, 3]])

Exercice de synthèse : un médaillon pour Lena

Nous allons faire quelques manipulations sur les tableaux numpy en partant de la célébre image de Lena (http://www.cs.cmu.edu/~chuck/lennapg/). scipy fournit un tableau 2D de l'image de Lena avec la fonction scipy.lena

>>> import scipy
>>> lena = scipy.lena()

Voici quelques images que nous allons obtenir grâce à nos manipulations : utiliser différentes colormaps, recadrer l'image, modifier certaines parties de l'image.

• Utilisons la fonction imshow de pylab pour afficher l'image de Lena.

In [3]: import pylab
In [4]: lena = scipy.lena()
In [5]: pylab.imshow(lena)

• Lena s'affiche alors en fausses couleurs, il faut spécifier une colormap pour qu'elle s'affiche en niveaux de gris.

In [6]: pylab.imshow(lena, pl.cm.gray)
In [7]: # ou
In [8]: gray()

• Créez un tableau où le cadrage de Lena est plus serré : enlevez par exemple 30 pixels de tous les côtés de l'image. Affichez ce nouveau tableau avec imshow pour vérifier.

In [9]: crop_lena = lena[30:-30,30:-30]

• On veut maintenant entourer le visage de Lena d'un médaillon noir. Pour cela, il faut

  • créer un masque correspondant aux pixels qu'on veut mettre en noir. Le masque est défini par la condition (y-256)**2 + (x-256)**2

In [15]: y, x = np.ogrid[0:512,0:512] # les indices x et y des pixels
In [16]: y.shape, x.shape
Out[16]: ((512, 1), (1, 512))
In [17]: centerx, centery = (256, 256) # centre de l'image
In [18]: mask = ((y - centery)**2 + (x - centerx)**2)> 230**2

puis

• affecter la valeur 0 aux pixels de l'image correspondant au masque. La syntaxe pour cela est extrêmement simple et intuitive :

In [19]: lena[mask]=0
In [20]: imshow(lena)
Out[20]: <matplotlib.image.AxesImage object at 0xa36534c>

• Question subsidiaire : recopier toutes les instructions de cet exercice dans un script medaillon_lena.py puis exécuter ce script dans Ipython avec %run medaillon_lena.py.

Setting up SSH: spreading the good keys

1. There are many ways to authenticate your session, but mainly password or keys. Keys are to be preferred to avoid typing your password 10 times a day. It is also most secure (you type your key's password locally and not remotely).

2. Generate a private/public key pair. Simple command to do this:

   ssh-keygen -t rsa

3. Copy the key to the

   ssh-copy-id -i ~/.ssh/id_rsa.pub username@host

   . this can be also be done using

   scp ~/.ssh/id_rsa.pub username@host:~/mykey.pub
   ssh username@host
   cat mykey.pub >> .ssh/authorized_keys

4. Now try logging into the remote machine again from local

   ssh REMOTE_USERNAME@remote_host

5. Check that your public key is in the list of authorized keys: .ssh/authorized_keys.

6. Change password regularly:

   ssh-keygen -p

   It is not advised to put an empty pass-phrase, rather use key agent (see below).

key agent

• An agent loads your keys on the local machines:

  • it's more secure, since all passwords are typed locally, you only send encrypted authentifications

  • it's more practical, since you type your password once per session

• http://www.sshkeychain.org/mirrors/SSH-with-Keys-HOWTO/

• GUI interface on MacOsX : http://www.sshkeychain.org/

  • install with macports using sudo port install SSHKeychain, you'll find it in /Applications/MacPorts

tunnels

• http://souptonuts.sourceforge.net/sshtips.htm

• http://projects.tynsoe.org/en/stm/doc.php

securing the server

• Robots usually try common name / password combinations on your SSH server. If you're the only user admin_name of your server you may use in the SSH server configuration file (usually /etc/ssh/sshd_config) the option AllowUsers admin_name to restrict access to user admin_name and avoid brute force attacks. Since robots are most of the time dumb, they'll get an immediate acces denied response to any connection request.

• Robots usually sniff port 22. To change the port which is listened by the SSH server, either modify the default port in the SSH server configuration file (usually /etc/ssh/sshd_config). Another way is to use your router to redirect the outside port (for instance 2221) to the default port of your server.

notre proposition

• APPEL A PROPOSITIONS

  PROGRAMME INTERDISCIPLINAIRE CNRS

  • Neurosciences et neuroinformatiquecomputationnelle Neuro-IC

  • Déclaration de candidature

    • Titre bref du Projet:(maximum 20 caractères) Émergencein computo

Titre long du Projet:(maximum 3 lignes) Collaborations multi-disciplinaires entre des équipes de neurosciences marseillaises pour une approche intégrative de l'étude de stratégies computationnelles.

Coordinateur du projet : M. Prénom : Laurent Nom :PERRINET Fonction :CR CNRS Laboratoire (nom complet et sigle, le cas échéant) : Institut de Neurosciences Cognitives de la Méditerranée (INCM – UMR 6193) Adresse complète du laboratoire : 31, chemin Joseph Aiguier 13402 Marseille cedex Courriel : Laurent.Perrinet@incm.cnrs-mrs.fr Tél :(0-33) 4 91 16 43 08 Fax :(0-33) 4 91 22 08 75

court CV du porteur du projet (1 page)

• Chercheur (CR1) à l**'INCM,CNRS, depuis 10/2004 · Sous la conduite de Guillaume Masson à l'Institut de Neurosciences Cognitives de la Méditerranée (INCM-UMR 6193, CNRS) à Marseille, j'étudie des modèles d'inférence spatio-temporels dans des flux video. Cette étude a pour but de définir un algorithme neuro-mimétique adaptatif de représentation sur-complète d'un flux video que nous désirons appliquer à l'extraction du champ de vitesse.

• Doctorat de Sciences cognitivesONERA/DTIM, Toulouse (France) 10/1999-2/2003 - Titre : Comment déchiffrer le code impulsionnel de la Vision? Étude du flux parallèle, asynchrone et épars dans le traitement visuel ultra-rapide. · Allocataire d'une bourse MENRT, accueil à l'ONERA/DTIM. · Cette thèse a été initiée par les résultats de la collaboration pendant le stage de DEA. Elle a été dirigée par Manuel Samuelides (professeur à Supaéro et chargé de recherche à l'ONERA/DTIM) et co-dirigée par Simon Thorpe (directeur de recherche au CerCo)

• DEA de Sciences cognitives (Univ. Paris VII, P. Sabatier, EHESS, Polytechnique), Paris (France), mention TB. Allocataire d'une bourse de DEA. 9/1998-9/1999 · 3/1999-7/1999 Assistant de recherche, ONERA/DTIM (Département de Traitement de l'Image et de Modélisation), Toulouse (stage de DEA). · 7/1999-8/1999 Assistant de recherche, USAFB (Rome, NY) / University of San Diego in California (États-Unis).

• Diplôme d'ingénieurSupaéro, Toulouse, France. 1993 - 1998 Spécialisation dans le traitement du signal et de l'image et plus particulièrement dans les techniques des réseaux de neurones artificiels. · 4/1998-9/1998 Assistant de recherche, CerCo (CNRS, UMR5549), Toulouse (stage de fin d'études d'ingénieur). Développement d'un réseau de neurones asynchrone appliqué à la reconnaissance de caractère. · 4/1997-9/1997 Assistant de recherche, Jet Propulsion Laboratory (Nasa), Pasadena, Californie. Département des Sciences de la Terre, Laboratoire d'imagerie radar, Interférométrie SAR · 9/1995-6/1996 Ingénieur Alcatel, Vienne (Autriche). Département du Voice Processing Systems. Budget demandé (sommairement présenté) et durée du projet: 30 k€ /2 ans Le budget sera notifié en une fois. Le programme n’attribuera aucun moyen humain. - 20 k€ : matériel computationnel (voir devis inclus), - 7k€: organisation d'un colloque et d'un atelier, - 3k€: missions, communication. Ce projet a-t-il été déjà évalué ou fait-il l’objet d’une demande en cours ? Dans quel cadre ? A quelle date ? Ce projet n'a pas été évalué précédemment. Plus précisément, ce projet fait-il partie d’une demande présentée à l’ANR dans les deux dernières années ou sera –t-il présenté dans l’année en cours ? Non. A-t-il été accepté ? Si en cours, à quelle date prévoyez-vous une réponse ? Demandez-vous un autre financement ? Précisez dans quel cadre, la somme demandée et la durée du contrat. Non.

Équipes participant au projet:

Neuroinformatique et neurosciences computationnelles

• Contexte

• Objectifs et plus-value attendue

• Descriptif du programme

• Enjeu scientifique interdisciplinaire

finder nags when changing a file's extension

• read the default value (should be 0)

  defaults read com.apple.finder FXEnableExtensionChangeWarning

• change it:

  defaults write com.apple.finder FXEnableExtensionChangeWarning False

Make spell check show only desired languages

• http://www.macosxhints.com/article.php?story=20090512175846504

HFS+

• Mac OS X Filesystems

• One feature of HFS volumes is that fle are referred to as links: for instance you can read a PDF file while changing the name at the same time. No problem!

• Safely remove '._' files created by HFS(+)

  find . -name '._*' -print0 | xargs -0 rm

Spotlight

• to remove the remaining index files on the volume using the command: `` sudo mdutil -E /Volumes/volume_name `` utilitaires CLI :

  sudo mdutil --help
  mdutil: unrecognized option `--help'
  usage: mdutil -pE volume ...
          mdutil can be used to manage the metadata stores used by Spotlight.
          -p              publish metadata for the provided volumes.
          -i (on|off)     set indexing status for the provided volumes.
          -E              erase the master copy of the metadata stores for the provided volumes.
          -s              print indexing status for the provided volumes.

System & Security

• http://www.macgeekery.com/tips/security/basic_mac_os_x_security

• Synthesis of unauthorized accesses to your machine:

  sudo grep "failed to auth" /var/log/secure.log | sed 's/^.*user \(.*\) for.*$/\1/' | sort | uniq -c

nmap

• to scan for open ports on a remote machine :

  sudo nmap -T3 -vv -sS -p 1-65535 -P0 google.com
  Starting nmap 3.81 ( http://www.insecure.org/nmap/ ) at 2006-04-05 16:32 CEST
  Initiating SYN Stealth Scan against google.com [65535 ports] at 16:32
  Discovered open port 80/tcp on google.com
  SYN Stealth Scan Timing: About 2.18% done; ETC: 16:55 (0:22:26 remaining)

• to scan for open ports on your local network (see https://stackoverflow.com/questions/13669585/how-to-get-a-list-of-all-valid-ip-addresses-in-a-local-network#15351073 ):

  nmap -sP 192.168.1.*

Dash board

• Don't use Dashboard? No particular reason to leave it running, consuming memory. Following http://www.macosxhints.com/article.php?story=20050723123302403, you can turn Dashboard off by doing:

  defaults write com.apple.dashboard mcx-disabled -boolean YES
  killall Dock

• Unsurprisingly, you change YES to NO to re-enable Dashboard:

  defaults write com.apple.dashboard mcx-disabled -boolean NO
  killall Dock

Unix - X11

• http://xquartz.macosforge.org/

• install quickly http://trac.macosforge.org/projects/xquartz :

  wget http://xquartz.macosforge.org/downloads/X11-2.4.0.dmg
  sudo installer -pkg X11-2.4.0.pkg  -target /
  rm  X11-2.4.0.pkg

Change login window on (Snow) Leopard

• disrupted by the look of the plasma flames? think it looks like a cheap star trek sundae? check http://paulstamatiou.com/2007/10/31/how-to-change-leopards-login-wallpaper :

  cd /System/Library/CoreServices
  sudo rm DefaultDesktop.jpg
  #sudo mv DefaultDesktop.jpg DefaultDesktop_old.jpg # if you want to keep the Aurora stuff (it stills around, do a 'locate Aurora'
  sudo ln -s /Library/Desktop\ Pictures/Nature/Stones.jpg DefaultDesktop.jpg

symbols

• Use Detexify2 - LaTeX symbol classifier to find out a particular symbol.

• list of symbols: http://web.ift.uib.no/Fysisk/Teori/KURS/WRK/TeX/symALL.html

% Math-mode symbol & verbatim
\def\W#1#2{$#1{#2}$ &\tt\string#1\string{#2\string}}
\def\X#1{$#1$ &\tt\string#1}
\def\Y#1{$\big#1$ &\tt\string#1}
\def\Z#1{\tt\string#1}

% A non-floating table environment.
\makeatletter
\renewenvironment{table}%
   {\vskip\intextsep\parskip\z@
    \vbox\bgroup\centering\def\@captype{table}}%
   {\egroup\vskip\intextsep}
\makeatother

% All the tables are \label'ed in case this document ever gets some
% explanatory text written, however there are no \refs as yet. To save
% LaTeX-ing the file twice we go:
\renewcommand{\label}[1]{}

%%end-prologue%%
\begin{table}
\begin{tabular}{*8l}
\X\alpha        &\X\theta       &\X o           &\X\tau         \\
\X\beta         &\X\vartheta    &\X\pi          &\X\upsilon     \\
\X\gamma        &\X\gamma       &\X\varpi       &\X\phi         \\
\X\delta        &\X\kappa       &\X\rho         &\X\varphi      \\
\X\epsilon      &\X\lambda      &\X\varrho      &\X\chi         \\
\X\varepsilon   &\X\mu          &\X\sigma       &\X\psi         \\
\X\zeta         &\X\nu          &\X\varsigma    &\X\omega       \\
\X\eta          &\X\xi                                          \\
                                                                \\
\X\Gamma        &\X\Lambda      &\X\Sigma       &\X\Psi         \\
\X\Delta        &\X\Xi          &\X\Upsilon     &\X\Omega       \\
\X\Theta        &\X\Pi          &\X\Phi

\end{tabular}
\caption{Greek Letters}\label{greek}
\end{table}



\begin{table}
\begin{tabular}{*8l}
\X\pm           &\X\cap         &\X\diamond             &\X\oplus     \\
\X\mp           &\X\cup         &\X\bigtriangleup       &\X\ominus    \\
\X\times        &\X\uplus       &\X\bigtriangledown     &\X\otimes    \\
\X\div          &\X\sqcap       &\X\triangleleft        &\X\oslash    \\
\X\ast          &\X\sqcup       &\X\triangleright       &\X\odot      \\
\X\star         &\X\vee         &             &\X\bigcirc   \\
\X\circ         &\X\wedge       &              &\X\dagger    \\
\X\bullet       &\X\setminus    &            &\X\ddagger   \\
\X\cdot         &\X\wr          &          &\X\amalg     \\
\X+             &\X-
\end{tabular}

\caption{Binary Operation Symbols}\label{bin}
\end{table}



\begin{table}
\begin{tabular}{*8l}
\X\leq          &\X\geq         &\X\equiv       &\X\models      \\
\X\prec         &\X\succ        &\X\sim         &\X\perp        \\
\X\preceq       &\X\succeq      &\X\simeq       &\X\mid         \\
\X\ll           &\X\gg          &\X\asymp       &\X\parallel    \\
\X\subset       &\X\supset      &\X\approx      &\X\bowtie      \\
\X\subseteq     &\X\supseteq    &\X\cong        &    \\
  & &\X\neq         &\X\smile       \\
\X\sqsubseteq   &\X\sqsupseteq  &\X\doteq       &\X\frown       \\
\X\in           &\X\ni          &\X\propto      &\X=            \\
\X\vdash        &\X\dashv       &\X<            &\X>            \\
\X:
\end{tabular}

\caption{Relation Symbols}\label{rel}
\end{table}

checking typographic style

• Lint for LaTeX : http://baruch.ev-en.org/proj/chktex/

make links

• hyperref quick reference list :

  \href{URL}{text }
  \url{URL}
  \nolinkurl{URL}

managing margins

• to adjust margins, use

  \usepackage[margin=2.5cm]{geometry}

  then play around with the 2.5cm value until it fits.

• tips for fitting your text in the required size : LaTeX Tips n Tricks for Conference Paper

citations

• If you give LaTeX \cite{fred,joe,harry,min}, its default commands could give something like "[2,6,4,3]"; this looks awful. One can of course get the things in order by rearranging the keys in the \cite command, but who wants to do that sort of thing for no more improvement than "[2,3,4,6]"

  • The cite package sorts the numbers and detects consecutive sequences, so creating "[2-4,6]". The natbib package, with the numbers and sort&compress options, will do the same when working with its own numeric bibliography styles (plainnat.bst and unsrtnat.bst).

  • If you might need to make hyperreferences to your citations, cite isn't adequate. If you add the hypernat package:

    \usepackage[...]{hyperref}
    \usepackage[numbers,sort&compress]{natbib}
    \usepackage{hypernat}
    ...
    \bibliographystyle{plainnat}

    See for example http://www.tex.ac.uk/cgi-bin/texfaq2html?label=citesort

Useful draft tips

• “LaTeX and Subversion”

  • set a keyword with ``svn propset svn:keywords "Id" index.tex `` so that every occurrence of

    System Message: WARNING/2 (<string>, line 70); backlink

    Inline literal start-string without end-string.

  • use latex-svninfo http://www.ctan.org/tex-archive/macros/latex/contrib/svninfo/ for instance with

    \usepackage[fancyhdr,today,draft]{svninfo}%
    %\usepackage[fancyhdr]{svninfo}%
    \pagestyle{fancyplain}
    \fancyhead{}

  • now at every commit $Id$ will be replaced by useful data that will show up in the foot of the page

  • for a reference on using keywords see http://wiki.loria.fr/wiki/Variables_automatiques

using pdfLaTeX

• but... sites like arXiV use only plain LaTeX so that you should keep the 2 versions of your directives for better portability (see \ifpdf ...)

  • in particular arXiV rejects the microtype package

• le package hyperref permet même de faire des références vers les différents chapitres.

• PDfLaTeX ne permet pas d'inclure des eps pour cela il faut les convertir en pdf avec epstopdf ou le script suivant qui permet de convertir tous les .eps d'un dossier (à sauver et rendre executable):

  for f in $* ;do
      if echo "$f" | grep -i eps*   ; then
           epstopdf --nocompress $f
           echo "converting  $f to pdf ..."
      else
      echo "$f is not a eps file, ignored"
      fi
  done

  il suffit alors d'executer en console `` ./mon_script la-ou-ya-tout-mes-eps/*.eps ``

  System Message: WARNING/2 (<string>, line 116); backlink

  Inline emphasis start-string without end-string.

count number of words / compter le nombre de mots

• Pour compter le nombre de mots et de caractères d'un document latex, il suffit d'installer deTeX et de lancer la simple ligne de commande

  detex MonFichier.tex | wc -w

• alternatively, you may use

  pdftotext MonFichier.pdf - | wc -w

• in TexShop there's a "Statistics..." interface to the same technique.

• on MacOsX, to install appropriate tools, use MacPorts and

  sudo port install detex
  sudo port install xpdf +a4 +with_poppler

including source code in a document with pretty printing

• use

  \usepackage{attachfile}

  or

  \usepackage{filecontents}

• see documentation:

  texdoc attachfile

referring to table or image

• referring to table or image (and not to the bottom of it)

  \usepackage{hypcap}

framed box

• make a framed box around text (and configure space) :

  \setlength\fboxsep{1pt}
  \setlength\fboxrule{0.5pt}
  \fbox{text}

convert a collection of JPGs to a pdf

\listfiles
\documentclass{minimal}
\usepackage{graphicx}
\usepackage[active,graphics,tightpage]{preview}
\begin{document}
\includegraphics{pic1}
\includegraphics{pic2}
\includegraphics{pic3}
\end{document}

• or

for f in *.jpg ; do convert $f `basename $f .jpg`.pdf ; done

• or `` slideshow.tex`` TeX file

___________________________________________________________
\pdfcatalog{/PageMode/FullScreen}\pdfcompresslevel=0
\pdfhorigin0pt\pdfvorigin0pt
\def\process#1 {\setbox0\hbox{\pdfximage width 20cm {#1}%
  \pdfrefximage\pdflastximage}%
  \pdfpagewidth=\wd0 \pdfpageheight=\ht0 \shipout\box0\par}
\everypar{\setbox0\lastbox\process} \input dir \end
___________________________________________________________
Usage:
ls *.jpg > dir
pdftex slideshow

Tex on MacOsX

• !TexLive is the most recent /easy distribution. You may add new packages easilly in $HOME/Library/texmf (see a reference) or using the TexLive tool: tlmgr

• to install :

  wget http://ftp.klid.dk/ftp/texlive/tlnet/mactex-2009-sept-20.mpkg.zip
  unzip mactex-2009-sept-20.mpkg.zip
  sudo installer -pkg MacTeX-2009.mpkg -target /

  (check before on http://ftp.klid.dk/ftp/texlive/tlnet/ the correct name)

• I had to set up a new source repository :

  sudo tlmgr option location http://ftp.klid.dk/ftp/texlive/tlnet

• to upgrade

  sudo tlmgr update --self
  sudo tlmgr update --all

preferred: Hardlink

From http://code.google.com/p/hardlinkpy/ , "hardlink.py is a tool to hardlink together identical files in order to save space.". Thus the filesystem is the same, but duplicate files are checked so they are actually written once on the hard drive.

Dupinator

Dupinator, tries to find duplicates and to report them in order to clean-up the organization of your files.

Touches spéciales (unix) sur clavier Mac francais

Merci a http://busy.lab.free.fr/mac/ !

Apple also got the habit to give it's own special names to some keys:

• ctrl is called, well, control. So far, so good.

• the alt key (with a funy sign under it) is called option.

• the key with a funny sharp and the Apple logo ⌘ is called command (for we are supposed to be so creative Apple users I think), or sometimes more logically apple.

keyboard shortcuts

• in Finder (and most other applications)

  • ⌃⌘Z to zoom the window

  • ⌘ + ` to switch between your windows

  • ⇧⌘[ and ⇧⌘] to switch tabs

  • ⌃⌘D to show the dictionary

• in Firefox:

  • Fn + Alt + Left arrow gets you to the home page.

  • Command + L : link bar

  • Command + K: search box

  • Ctrl + Tab: Next Tab

• see Mac OS X keyboard shortcuts

Program

Afternoon of Monday, 22 October 07: Framework

12:30-14:00

Registration and Lunch

14:00-18:00 (First session)

Introduction: coordination of modeling efforts for WP9T2

14:00-14:10

Guillaume Masson (INCM): Welcome

14:10-14:30

• Laurent presented some examples of using the benchmark structure proposed in NeuroTools and the new SpikeList object. The benchmark will be the privileged way of spreading the different benchmarks to all partners and is a practical way of describing a benchmark as a list of experiments, storing the result of the experiments, distribute and run the experiments on a cluster and finally for plotting figures. In a second part, Laurent presented the SpikeList object as an answer to the difficulty of finding a common format for describing lists of spikes. Some experimental example and an interface with PyNN was shown as a proof of concept. Jan suggested that the internal representation could be sparse. This should be easily done by using the format methods (See SpikeList).

14:40-15:10

• Andrew and Jens presented a system's approach to modeling the visual system. Partners in FACETS need to exchange knowledge on their modeling progress and this approach allowed to construct a model from composite parts (retina, LGN, V1) from the “LEGO bricks” of the different groups. A practical example with INRIA/INCM retina and a INCM V1-layer 4 was shown. An LGN brick is to be developed. Brick for higher areas (MT) can also be developed.

15:20-16:00

Coffee

16:00-16:20

• Andrew proposed a format specification for benchmarks : 1) the input is provided as zipped PNG files, storing is done in the FACETS knowledge base (every stimulus has a unique URL), 2) A specification scheme for writing benchmarks and a whole experiment was presented with a proposed format using a XML interface.

16:30-16:50

• Adrien presented his physiologically realistic retina, emphasizing on the contrast gain control. His model uses the INRIA simulator but with an open architecture and specification compatible with the FACETS specification. The retina uses a feed-back on the bipolar cells to achieve realistic contrast gain control which was compared with classical experiments (Shapley and Victor, 79, Enroth and Cugell XX). A possible extent is to study the importance of the spike profile at image onset which reveals first the luminance image and then the edges (Enroth Cugell 83 Bernadete-Kaplan 99). He also approached a mean field approach valid for gratings.

17:00-17:20

• Klaus presented the retina used at the TUG. It is based on a spatio-temporal decorrelation model from Dong and Attick (1995) for the linear part and on a gamma renewal process (Gazères, 1998) for the spiking part. It is available to FACETS partners in the SVN1 and will converted to the format specified in the VisualSytem class (See https://www.kip.uni-heidelberg.de/repos/FACETSCOMMON/facetsmodel/LGN/trunk/graz ).

17:30-18:00

General Discussion

• conclusions on benchmarking progress and decisions about their future development and use.

• We concluded on benchmarking progress and decided about their future development and use. In particular, the unification proposed by the Benchmark class and the SpikeList object from Laurent and the specification proposed by Andrew seemed to fit to the needs of the partners and we agreed to use the proposed schemes.

• We agreed on further definition of the visual benchmarks WP9T2-VisionBenchmarks (while respecting the standards from FACETS_Benchmarks). This will be implemented in a coming deliverable.

• The question if we can reach the goal a single, common framework for multiple V1 models in FACETS was left open due to the wide variety of approaches in the consortium. However, the discussion suggested some collaborations of groups on some specific scientific questions.

Morning of Tuesday, 23rd of october 07

09:00-13:00 (Second session)

Progress in Modeling V1

• 09:00-09:20*

• Mike described current progress towards a neural model of motion perception in V1/MT, a collaborative project with CNRS/INCM. The model is based on the recurrent neural circuitry between a V1 hypercolumn and area MT, and uses a Kalman-Bucy approach to estimate the velocity of the moving object by integration of local motion information. The model will be suitable for testing using the FACETS visual motion benchmark stimuli, in particular the CNRS/INCM data on motion integration for smooth eye pursuit.*

09:30-09:50

• The KTH model was presented by Jan, with the latest changes and additions and some preliminary results. HH neuronal models, hierarchical columnar structure. The horizontal connectivity is specified by the LISSOM model while the vertical connectivity is inspired by biology.*

10:00-10:20

• Klaus presented the TUG model with the latest changes and some preliminary results. He illustrated the use of Izhikievitch neurons, the data from Alex Thomson (and not from Binzegger) by showing some results of the columnar model in particular by studying the influence of NMDA.*

10:30-10:50

• Jens presented progress of the INCM/ALUF cooperation emphasizing on progress in the specification of the thalamo-cortical projections. He presented results of applying the different benchmarks proposed to reveal the functional properties of the model. It emerged that the inhibition scheme used allowed a normalization of the input at the network level. This was put into evidence using the spike-triggered conductance profile which were consistent with some recent data from Rudolph et al. (2007).*

11:00-11:30

Coffee

11:30-13:00

• Round-table discussion: questions, problems, and issues of modeling V1. Moderator: G. Masson

• Retina model: shouldn't we use a more standardized input? How should we set up background noise? Biologists need to specify what they intend by background (on-going) activity to be simulated in large scale neural networks.

• Back to back cooperation between experimentalists and modelers for defining benchmarks in connections with dissemination of experimental data.

• Link with more high-level task, as the one presented by Mike and in WP9T3 or the work done in collaboration INCM-INRIA. Since they use the same benchmarks (such as motion integration), these approaches can better defined the computational rules to be implemented in large scale neural networks. One good example in the role of asymmetric diffusion of information/activity in the network, thanks to feedback from higher areas.

13:00-14:00

Lunch

Afternoon of Tuesday, 23rd of october 07

14:00-16:30 (Third session)

Scientific questions coming from the Biology of V1

14:00-14:20

• Julian presented a review of different results on the physiology and anatomy of the cat retino-thalamo-cortical projections. In fact, though the question of the architecture of these projections was questioned during the workshop, little is known with general agreement. Correlating different works from the literature, a detailed quantification was reviewed suggesting a disagreement between anatomy and physiology. Presenting the work of Ringach (2004), he concluded by showing own simulations suggesting that the properties and diversity of the receptive field of cat's area 17 simple cells may be captured by a wiring scheme based on the specific quantization of the parameters of the retino-thalamo-cortical pathway.*

14:30-14:50

• Cyril reviewed different models of the emergence of orientation and direction selectivity before emphasizing on the results of different groups on the role of conductance profile in this function. This revealed a diversity of behavior between push-pull model where inhibitory and excitatory profiles are in overlap and other configuration This was put in light with results obtained at the UNIC.*

15:00-15:20

• Alex presented preliminary results of center-surround interactions using VSD optical imaging in the primate V1 cortex. In the retinotopic position of the center the response to the center appears with decreasing latency for increasing contrast. The response of the 80% contrast surround reaches the center at a latency equivalent to approx. 15% contrast, leaving open the question of the interaction of these two information streams. Preliminary results show suppression for high contrasts but facilitation for low center contrast. Further analysis (of latency, propagation) suggests the functional role of horizontal propagation in this configuration.*

15:30-16:00

Coffee

16:00-16:30

General Discussion Moderator: Yves Fregnac (UNIC)

16:30-17:30

Outcome: Planning of Implementation plans / priorities for WP9T2.

• Several actions need to be taken. 1) We keep the idea of one annual meeting on V1 modeling. The meeting shall be held in June instead of October, to prepare for Annual reports and implementation plan. We will post a call for the organization of the 2008 meeting. We should also try to bring more biologists in these meetings. 2) We will organize a phone/video conference once every 3 months to exchange information and compare outputs for each benchmark steps. 3) We shall provide a timeline for delivering benchmark tools and objectives, as well as deadline for collecting results for testing models. Such a timeline will be added to the D9-2 in which we will describe the different benchmarks. 4) We will set a discussion list on the FACETS Wiki website to propose new questions from modelers to biologist and vice-versa. The idea is to put information for which there is a general agreement rather than having an on-going forum. Answers shall be concise, with reference to published work and or available data. 5) We shall promote active collaboration between sites with the objectives of common publication of one specific aspects of visual tasks to be develop in FACETS (on-going activity, local cortical point spread function ….)

Questions

Below are some questions modelers (please add to this list) would like the biologists at the meeting to answer during the round-table discussion (11:30-13:30 on 23 October):

• What is the purpose of the Y-type pathway input to layer 4 of cat area 17?

• Is the tuning of cortical neurons dynamic or not (e.g. for orientation)?

• Can simple or complex cortical cells be directionally selective but untuned for orientation?

• Are inhibitory neurons in cortex generally tuned for orientation or not?

• Do inhibitory fast-spiking (FS) neurons have a higher spiking threshold than excitatory regular-spiking (RS) neurons?

• Is modelling corticothalamic feedback essential for models of V1?

• What common mistakes do modellers make that annoy you and fellow biologists the most?

• How can modellers best help the biologists?

For the sake of fairness, we would also like some questions from the biologists for modellers to answer during the same discussion session.

Organization

Main points and decisions

• review / demonstration of coding strategies by different partners. This definition has impact on the "Meta Simulator",

• review of existing V1 models inside and outside FACETS (see also !Review_of_V1_models),

• definition of the common benchmark FACETS_Benchmarks to qualitatively compare different models and strategies in order to break the model complexity barrier, as we began to do for the simulator,

• definition of the architecture of the FACETS model of an "hyper-column" by defining the priorities of different partners: effect of including different neuron types, including layers, specific lateral connectivity.

• Partner 13 (Plymouth, Mike Denham) will provide an hypercolumn for D25.

• Partner 6b will provide a retina before Dec. 15 for implementing benchmark zero.

Monday, 20 November 2006

Monday, 20 November 2006

morning, 09:00 - 13:00

13:00

Lunch

afternoon

Tuesday, 21 November 2006

Tuesday, 21 November 2006

morning, 09:00 - 13:00

13:00

Lunch

afternoon

CRAAC 2005

2005-10-24

Récapitulatif de votre saisie : ce document sera soumis au visa de votre directeur

Compte rendu annuel d'activité des chercheurs du CNRS Année 2004 - 2005

Identité

Nom (nom de jeune fille) PERRINET Prénom Laurent Date de naissance 23/02/1973 Grade CR2 N° d'agent QAF195447 Téléphone 04 91 16 43 64 Télécopie 04 91 16 43 64 Adresse électronique laurent.perrinet@incm.cnrs-mrs.fr Section(s) du Comité National 7 Département scientifique Sciences de la vie Délégation régionale Provence Affectation

Intitulé de l'unité Institut de neurosciences cognitives de la méditerranée - INCM Code unité UMR6193 Directeur Driss BOUSSAOUD Adresse électronique du directeur driss.boussaoud@incm.cnrs-mrs.fr Adresse 31 Chemin Joseph Aiguier

• 13402 MARSEILLE CEDEX 20

Téléphone 04 91 16 43 18 Télécopie 04 91 77 49 69 Délégation Provence Site Web Distinction(s)

Qualification

Habilitation à diriger des recherches non Doctorat d'Etat non Doctorat oui Année d'obtention 2003 Qualification "Maître de conférences" non Qualification "Professeur" non Période d'inactivité

Mobilité(s) antérieure(s)

Activités de recherche développées

Rattachement à(aux) activité(s) de recherche de l'unité UMR6193

Intitulé d'activité Date fin du rattachement

• DYNAMIQUE DE LA PERCEPTION VISUELLE ET DE L'ACTION (DyVA)

Mots clés des sections/CID du Comité national

Section 1 : Équations aux dérivées partielles, optimisation, contrôle, théorie du signal Section 1 : Probabilités, processus et algorithmes stochastiques, statistique, analyses des données Section 7 : Intelligence artificielle : raisonnement, décision, apprentissage Section 7 : Modélisation, analyse, commande et supervision des systèmes dynamiques Section 27 : Neurosciences comportementales et cognitives normales et pathologiques, homme et modèles animaux, imagerie cérébrale fonctionnelle Section 27 : Modélisation des processus cognitifs et neurosciences computationnelles Publication(s), parue(s) ou sous presse, dans des revues à comité de lecture

Référence FISCHER, S.; REDONDO, R.; PERRINET, L.; CRISTOBAL, G. (2005), «Sparse gabor wavelets by local operations.», Proceedings SPIE, volume 5839 of Bioengineered and Bioinspired Systems, 75 86 PERRINET L. Feature detection using spikes : the greedy approach. Journal of Physiology, Paris (special issue) - Sous presse PERRINET L., SAMUELIDES S., THORPE S.,. Coding static natural images using spiking event times: do neuron cooperate?. IEEE Transactions on Neural Networks. . . . 15. . . 1164-1175. . PERRINET L.,. Finding independent components using spikes: A natural result of hebbian learning in a spike coding scheme. Natural Computing . . . . 3. . . 159-175-. . PERRINET L.;. Emergence of filters from natural scenes in a sparse spike coding scheme. Neurocomputing. . . . 58-60. . . 821-826. . PERRINET, L; (2005), «Efficient source detection using integrate-and-fire neurons.», In W. Duch et al., editor, ICANN 2005, Lecture Notes in Computer Science, volume 3696, pages 167-72 Rafael Redondo, Sylvain Fischer, Laurent Perrinet, and Gabriel Cristobal. Simple cells modeling through a sparse overcomplete gabor wavelet representation based on local inhibition and facilitation. In Gustavo Linan-Cembrano Ricardo A. Carmona, editor, Perception, volume 34 of Bioengineered and Bioinspired Systems II, page 238, August 2005. Publication(s), parue(s) ou sous presse, dans des revues sans comité de lecture

Ouvrage(s) ou chapitre(s) d'ouvrage(s), paru(s) ou sous presse

Participation à des manifestations scientifiques

Manifestation Dynn - ACI Temps et Cerveau Type de manifestation ( national ) Lieu Cannes ( FRANCE ) Durée 3 (jour(s))

Manifestation ECVP Type de manifestation ( international ) Lieu La Corogne ( ESPAGNE ) Durée 5 (jour(s))

Manifestation ICANN Type de manifestation ( international ) Lieu Varsovie ( POLOGNE ) Durée 5 (jour(s))

Manifestation Maths et Cerveau Type de manifestation ( international ) Lieu Paris ( FRANCE ) Durée 15 (jour(s))

Manifestation Pre-FACETS Workshop on Simulation and Computation Type de manifestation ( international ) Lieu Graz ( AUTRICHE ) Durée 4 (jour(s))

Activité éditoriale

Rapporteur/Relecteur dans des revues Informations complémentaires Advanced Concepts for Intelligent Vision Systems (Sept 20-23, 2005, University of Antwerp, Antwerp, Belgium, http://acivs.org/acivs2005/)

Rapporteur/Relecteur dans des revues Informations complémentaires Neural Information Processing

Rapporteur/Relecteur dans des revues Informations complémentaires Neural Processing Letters

Rapporteur/Relecteur dans des revues Informations complémentaires International Journal of Neural Systems (IJNS)

Séjour(s) dans d'autres laboratoires

Objet Visite et collaboration avec Bruno Olshausen Organisme Redwood Neuroscience Institute Pays ETATS UNIS Unité RNI Durée annuelle 30 (j)

Objet Collaboration avec Sylvain Fischer et Gabriel Cristobal - co-publications Organisme CSIC Pays ESPAGNE Unité Instituto de Optica Durée annuelle 30 (j)

Mission(s) sur le terrain

Formation personnelle

Collaborations

Organisme partenaire CNRS Pays FRANCE ( Europe ) Unité partenaire UMR Mouvement et Perception Intitulé ANR RETINAE Cadre de la coopération Nature de l'activité

Organisme partenaire INRIA Pays FRANCE ( Europe ) Unité partenaire Odyssee Intitulé FACETS Cadre de la coopération AUTRE - contrat europeen Nature de l'activité Participation à un réseau

Organisme partenaire SUPAERO - CERT Pays FRANCE ( Europe ) Unité partenaire ONERA Intitulé Reseau Dynn - ACI Temps et Cerveau Cadre de la coopération Nature de l'activité Participation à un réseau

Encadrement et animation scientifique

Animation scientifique Participation et organisation dans la formation interne organisee par l'INCM. Enseignement

Valorisation et partenariat

Vulgarisation

Type d'information Intitulé Type de participation Intervention en milieu scolaire Fete de la Science Participation ponctuelle Administration de la recherche