Grants & Funding | Laurent Perrinet

MesoCentre (2018/2026)

Fri, 25 Oct 2024 00:00:00 +0000

In the field of neuroscience, modeling has long made it possible to validate and predict theories of information processing in the neural networks that make up the brain. The recent emergence of solutions inherited from machine learning, in particular deep learning, has changed the field since 2012. One reason for the effectiveness of these methods is the amount of data analyzed but above all the ability to teach these algorithms on dedicated architectures, including graphics cards (GPUs). Indeed, this architecture allows an algorithm to be parallelized into a multitude of simple and independent subprograms that allow speed gains of around 6x to 10x to be achieved over traditional visual information processing architectures. We are now using these architectures excessively to test new image processing models - targeting applications in both neuroscience and machine learning.

🇫🇷 Dans le domaine des neurosciences, la modélisation a longtemps permis d’effectuer des validations et prédictions sur les théories du traitement de l’information dans les réseaux de neurones qui constitue le cerveau. L’émergence récente de solutions héritées de l’apprentissage machine, en particulier l’apprentissage profond (Deep Learning) est venu bouleverser le champ depuis 2012. Une raison de l’efficacité de ces méthodes est la quantité de données analysées mais surtout la capacité de faire apprendre ces algorithmes sur des architectures dédiées, notamment les cartes graphiques (GPU). En effet cette architecture permet de paralléliser un algorithme en une multitude de sous programmes simples et indépendants qui permettent d’atteindre des gains de vitesse de l’ordre de 6x à 10x sur des architectures classiques de traitement de l’information visuelle. Nous utilisons maintenant excessivement ces architectures pour tester de nouveaux modèles de traitement des images - en visant des applications aussi bien aux neurosciences qu’en apprentissage machine.

Acknowledgement: This work was granted access to the HPC resources of Aix-Marseille Université financed by the project Equip@Meso (ANR-10-EQPX-29-01) of the program « Investissements d’Avenir » supervised by the Agence Nationale de la Recherche.»

ANR MarmoCatch (2023-10/2029-04)

Fri, 27 Oct 2023 00:00:00 +0000

Under natural conditions, many animals perform interceptive movements to catch small prey (Shaw22 for a marmoset study). Such movements require the processing of several features of the prey, such as its size, orientation, and position, which are ultimately expressed into the coordinated control of the arm and hand during movement execution. Furthermore, in the case of a moving prey, catching movements must also take into account the highly dynamic and intricate nature of these visual features in order to precisely control the movement to catch at the appropriate location, orientation, and timing. This behaviour relies on the capacity of our brain to overcome the intrinsic neuronal delays in visual and motor systems to anticipate as accurately as possible the prey’s trajectory in the multiple features of interest. At the visual level, moving stimuli are known to induce predictions along their trajectory (Krekelberg01, Nijhawan08), yet with fewer neurophysiological (Jancke04, Guo07, Subramaniyan2018, Benvenuti21) or theoretical evidence (Grzywacz95, Perrinet12). At the motor level, neurons in motor and parietal cortex have been reported to be involved in the predictive control of interception movements that require a precise estimation of the movement time (Port01, Merchant04, Li22). However, how visual and motor predictive responses relate to one-another remains an open issue, even more so under naturalistic conditions in which several target features may (co-)vary in parallel. Our objective is to shed light on the mechanisms underlying the coordination between visual and motor cortices in the marmoset while it prepares and executes the grasp of a real physical target in complex motion.

Acknowledgement

This work was supported by ANR project “MarmoCatch” N° ANR-XXX-YYY.

Emergences (2023 / 2027)

Thu, 05 Oct 2023 14:00:00 +0000

TL;DR: Conventional deep learning models consume too much energy. Inspired by biology, we will explore new models that are more energy efficient.

The Emergences project aims at advancing the state-of-the art on near-physics emerging models by collaboratively exploring various computation models leveraging physical devices properties. This project will focus on Event-based models, Physics-inspired models and innovative near-physics Machine Learning solutions. Emergences further intends to extend the collaborative research activities beyond the fence of the consortium by means of connecting with other projects of the PEPR IA and other research institutes.

Pilote: Marina Reyboz, CEA, Research Director
Co-Pilote: Gilles Sassatelli, CNRS, Research Director

Latest news

2026-01-29 : talk at the PEPR AI meeting
Laurent U Perrinet (2026). Neurosciences and sparsity. Séminaire au colloque du PEPR AI ``Emergences’’ 2026.
Cite Slides URL
2025-03-18: PEPR IA Days du 18 au 20 mars à CentraleSupélec.
2024-09-26: 2nd workshop meeting in Paris.
2024-05-03 : we are hiring !
Laurent U Perrinet (2024). PhD thesis 'Focus of attention: a sensory-motor task for energy reduction in spiking neural networks'.
2024-03-27 : talk at the PEPR AI meeting
Laurent U Perrinet (2024). Analyser de larges volumes de données neurobiologiques, vers une approche biomimétique.
Cite Slides URL
2023-10-05: Kick-off meeting!

Description of the “Emergences” project

Contemporary machine learning (ML) has incurred profound changes in the scientific, societal and economic landscapes alike. After a decade of sustained progress AI as a discipline is still making regular breakthroughs on many fronts, at the expense of an ever-increasing amount of consumption of compute resources. Modern language models feature hundreds of billion parameters and training energy consumption alone likely falls in the GWh range, with a logical forecast worsening the already prohibitive carbon footprint of AI.

Besides the flourishing initiatives aimed at defining AI-friendly digital compute stack, the next logical breakthrough on the horizon is undoubtedly the emergence of disruptive AI compute technologies having improved energy efficiency. This development will likely involve the utilization of models that differ from those traditionally used in ML and exhibit properties that resemble the behavior of physical components, thereby facilitating implementation.

The Emergences project aims at advancing the state-of-the art on near-physics emerging models by collaboratively exploring various computation models leveraging physical devices properties. Efforts will be put on 3 distinct fronts: Event-based models, Physics-inspired models (from physical systems dynamics) and innovative near-physics ML solutions (exploiting device properties). The investigations will be focused on embedded systems for Edge AI that call for increased energy efficiency for inference and learning, which could be incremental. They will apply to several application domains ranging for instance from the monitoring of the environment to health. Other important tasks such as common tools, performance metrics definition and model scalability analysis and will be conducted through as a collaborative transverse initiative.

Emergences further intends to extend the collaborative research activities beyond the fence of the consortium by means of connecting with other projects of the PEPR IA and other research institutes, some of which are listed in this proposal. Finally, because of the unavoidable societal and philosophical implications of AI as a whole, Emergences will concurrently to the research activities run a track aimed at analyzing and anticipating the impact of its upcoming contributions.

Description of the PhD project (WP1): Focus of attention: a sensory-motor task for energy reduction in unsupervised spiking neural networks

attention mechanisms based on our cognitive architecture using a dual pathway:
Emmanuel Daucé, Pierre Albigès, Laurent U Perrinet (2020). A dual foveal-peripheral visual processing model implements efficient saccade selection. Journal of Vision.
Cite DOI Pdf Code URL bioRxiv
implementation in a spiking neural network based:
Antoine Grimaldi, Laurent U Perrinet (2023). Learning heterogeneous delays in a layer of spiking neurons for fast motion detection. Biological Cybernetics.
PDF Cite DOI URL Code Supplementary Material

Carte des partenaires du projet Emergences.

L’intelligence artificielle induit des changements profonds dans les paysages scientifiques, économiques, politiques et sociétaux contemporains. Une décennie après sa « renaissance », l’apprentissage automatique continue à réaliser des avancées sur de nombreux fronts, au prix cependant d’une boulimie de ressources informatiques induisant une consommation électrique préoccupante. Les modèles de langage actuels comportent quelques centaines de milliards de paramètres et consomment pour leur entraînement seul plusieurs GWh, ce qui aujourd’hui motive la recherche d’approches (de rupture) plus sobres. En plus des diverses initiatives visant à développer des composants et systèmes numériques pensés pour l’IA et dotés d’une meilleure efficacité énergétique, des approches disruptives en IA doivent être développées pour viser des gains énergétiques encore plus importants. Cette évolution passera par l’utilisation de modèles différents de ceux utilisés traditionnellement en apprentissage et présentant des propriétés proches des comportements de composants physiques, en facilitant par là-même l’implantation. Le projet Emergences fait avancer l’état de l’art sur les modèles émergents proches de la physique en explorant de manière collaborative divers modèles de calcul en utilisant les propriétés de différents dispositifs physiques. Les efforts concentrés sur trois fronts distincts : i) les modèles événementiels bio-inspirés pour lesquels des avancées sont réalisées sur la compréhension du fonctionnement de ces modèles et leur optimisation notamment dans le cas d’implémentation sur semiconducteurs (FPGA et ASIC) mais aussi de lois d’apprentissage émergentes (Sparse Forward forward) ou encore sur la parcimonie des données ii) les modèles inspirés de la physique pour lesquelles des premières propositions concrètes de solutions permettant de réaliser des circuits émergent (création d’un testchip, modèles stochastiques, apprentissage continu bayésien, accélération matériel de couches MHA, entre autres) et enfin iii) les solutions d’apprentissage automatique innovantes proches de la physique, avec des propositions d’implémentation d’apprentissage on-chip pour des composants émergents (apprentissage forward-only sur réseaux memristifs, apprentissage multimodal et incrémental). Toutes ces investigations sont menées dans un cadre d’expérimentations basés sur des jeu de données et indicateurs de performances décidés communément, et un effort très significatif à l’endroit de la soutenabilité est réalisé avec une méthodologie d’évaluation de l’empreinte environnementale en cours de mise en place. Ainsi Emergences vise à proposer un cadre de travail pour qualifier ses propositions, face à l’état de l’art académique et industriel dans le domaine de l’IA à la périphérie (Edge AI) qui nous sert de référentiel d’analyse.

Key figures

starting date: September 1, 2023
Duration: 48 months (until August 31, 2027)
14 partners
Nb of PhD: 19
Nb of Post doc: 13
TRL: basic research
Total grant requested: 6.8 M€

This work is supported by a public grant overseen by the French National Research Agency (ANR) under the grant number ANR-23-PEIA-0002 EMERGENCES.

Polychronies (2022 / 2025)

Mon, 18 Jul 2022 14:00:00 +0000

THE POSITION HAS BEEN FILLED.

Description

Why do neurons communicate through action potentials, or spikes? An action potential is a binary event —it can occur or not, without further details— and asynchronous, i.e. it can occur at any time. In the living world, neurons almost systematically use this so-called event-based representation, though we do not yet have a clear idea why. A better understanding of this phenomenon remains a fundamental challenge in neurobiology in order to better interpret the masses of recorded data. It is also an emerging challenge in computer science to allow the efficient exploitation of a new class of sensors and impulse computers, called neuromorphic, which could allow significant gains in computing time and energy consumption —a major societal challenge in the age of the digital economy and of global warming.

The goal of this project is to bring an interdisciplinary perspective on the computational advantage of time series representations for the brain and for information processing machines. In particular, we will formalize mathematically a representation in an assembly of neurons based on a set of patterns of different relative spike times called polychronous groups. This hypothesis is directly inspired by neurobiological observations in the hippocampus, and the innovative aspect is to expand the capabilities of analog representations based on the firing rate by considering a representation based on repetitions of these polychronous groups at precise times of occurrence. This formalization is particularly well suited to neuromorphic computing, and allows for supervised or self-supervised learning of polychronous groups in any event-driven data. By extending this paradigm to a hierarchy, we envision practical applications of this approach in audio, video or neurobiological signal processing. The cross-fertilization of neuroscience and neuromimetic approaches will be instrumental in understanding the typical or pathological development of such spiking neural networks.

grant number AMX-21-RID-025:

" Ce travail a bénéficié d’une aide du gouvernement français au titre de France 2030, dans le cadre de l’Initiative d’Excellence d’Aix-Marseille Université – A*MIDEX, projet numero AMX-21-RID-025 "
" This work received support from the french government under the France 2030 investment plan, as part of the Initiative d’Excellence d’Aix-Marseille Université – A*MIDEX, under grant number AMX-21-RID-025 ”

Latest news

2023-09-11: Start of post-doc position
2023-05-01: Opening of post-doc position (THE POSITION HAS BEEN FILLED!)
2022-12-29: check out our review paper:
Antoine Grimaldi, Amélie Gruel, Camille Besnainou, Jean-Nicolas Jérémie, Jean Martinet, Laurent U Perrinet (2022). Precise spiking motifs in neurobiological and neuromorphic data.
Cite DOI Hal Code URL arXiv
2022-11-28: Pilot project at the INT brainhack
2022-07-18: Le projet Polychronies est lauréat de l’appel à projets « Interdisciplinarité » !
2022-02-27: read our complete proposal.

ANR ACES (2022/2026)

Tue, 13 Jul 2021 00:00:00 +0000

Contextual motor adaptation is the ability to produce different motor responses depending on different contingencies signaled by specific cues or contexts. This requires to learn the relation between antecedent stimuli, that signal the future state of the environment, motor responses, and outcomes. A wealth of research have demonstrated that motor systems such as the saccadic or the pursuit eye movement system may simultaneously adapt in two opposite directions (for instance increasing and decreasing the saccade amplitudes) when a context, such as the orbital position of the eye before the movement, signals different contingencies for each response.

However, it has also been repeatedly reported that some cues, such as the target color or its shape, do not come to control the adaptation of the motor response. These observations remain unexplained and we lack adequate theoretical concepts to account for them: any stimulus, or context, that is perfectly correlated with the experimental manipulation should, in theory, induce contextual adaptation as it is conventionally thought that outcome predictability is the main factor controlling contextual learning. This has been a particularly vexing problem for the past 25 years as motor adaptation has become one of the main experimental model to study learning in humans.

To solve this problem, the ACEs project relies on a general conceptual framework that elaborates on the active-inference view as well as recent proposals regarding the relation between value-based decision making and attention. Our conceptual model is grounded on the notion that, at each moment, several hypotheses regarding credit assignment (what causes what?) are competing to produce a behavioral policy. The inputs are categorized, somehow arbitrarily, as internal status, prior knowledge and sensory inputs. Sensory inputs might be viewed as affecting the hypothesis space while prior knowledge and internal status would provide bias in favor of various credit assignment hypothesis. Competition in the hypothesis space, relying on Bayesian inference, determines a unique motor response. Because out of all the different credit assignment hypotheses only one will prevail and determine the actual behavioral policy, the influence of the inputs on behavior are limited by their specific contribution to the dominating hypothesis, i.e. their weight.

Fiche d’identité

Acronyme : ACES (ANR-21-CE28-0013)
Title : Assignment of credit and constraints on eye movement learning
Coordinateur Scientifique : Laurent Madelain (ScaLab)
Responsable Scientifique local : Anna Montagnini (UMR7289)
Durée: 4 ans, à partir du 1er mars 2022 - 1er mars 2026
Budget total: 435 k€
https://anr.fr/Projet-ANR-21-CE28-0013

Acknowledgement

This work was supported by ANR project ANR-21-CE28-0013 “ANR ACES”.

ANR RubinVase (2021/2024)

Thu, 01 Apr 2021 00:00:00 +0000

Le but de RUBIN-VASE est de concevoir et valider de nouveaux modèles variationnels pour l’évolution des activations neuronales dans les systèmes visuel et auditif, codant naturellement le principe neurobiologique de représentation efficace. En nous concentrant sur des modifications des équations de Wilson-Cowan pour la dynamique neuronale, nous visons à i) valider cette approche pour le cortex visuel primaire, à travers l’étude des patterns hallucinatoires ; ii) développer un cadre neuro-inspiré pour le traitement sonore et la reconstruction vocale, à partir des mêmes principes ; iii) comparer les modèles variationnels proposés à des modèles data-driven. Pour atteindre nos objectifs nous couplerons le développement de théories mathématiques rigoureuses avec leur validation numérique et expérimentale. Cela se fera à travers une interaction originale entre des techniques variationnelles ou issues de la théorie du contrôle et des expériences psycho-physiques.

carte d’identité du projet

Durée: 4 ans, à partir du 1er avril 2021
Budget total (partenaire français): 665 k€
Coordinateur Scientifique : Dario PRANDI (Laboratoire des Signaux et Systèmes)
Partenaire(s) : AGENCE NATIONALE DE LA RECHERCHE, CE48 - Fondements du numérique: informatique, automatique, traitement du signal
Responsable Scientifique INT : Laurent PERRINET (UMR7289)
https://anr.fr/Projet-ANR-20-CE48-0003

Acknowledgement

This work was supported by ANR project “RubinVase” N° ANR-20-CE48-0003.

ANR AgileNeuRobot (2021/2025)

Mon, 07 Dec 2020 00:00:00 +0000

An Unmanned aerial vehicle (UAV) flying autonomously in a cluttered environment would require the agility to navigate rapidly by detecting as fast as possible potential obstacles, as represented here by the collision zone, given a cruising speed, associated to slow or fast latencies (respectively red and blue shaded areas). This project will provide with a novel neuromorphic architecture designed to meet these requirements thanks to an event-based, two-way processing.

Fiche d’identité

Acronyme : AgileNeuRobot (ANR-20-CE23-0021)
Titre : Robots aériens agiles bio-mimetiques pour le vol en conditions réelles
Title : Bio-mimetic agile aerial robots flying in real-life conditions
CES : CE23 - Intelligence Artificielle / Instrument de financement : Projet de recherche collaborative (PRC) / Catégorie R&D : Recherche fondamentale
Coordinateur Scientifique : PERRINET Laurent (UMR7289)
Durée: 4 ans, à partir du 1er mars 2021 - 1er décembre 2025
Budget total: 435 k€
Responsables Scientifiques : Stéphane Viollet (BioRobotique, Inst Sciences Mouvement), Ryad Benosman (Inst de la Vision ) | Laurent Perrinet (NeOpTo, Inst Neurosciences de la Timone, coordinateur)

A miniature, event-based ATIS sensor. Contrary to a classical frame-based camera for which a full dense image representation is given at discrete, regularly spaced timings, the event-based camera provides with events at the micro-second resolution. These are sparse as they represent luminance increments or decrements (ON and OFF events, respectively).

Résumé

Des robots aériens autonomes seraient des outils essentiels dans les opérations de recherche et de sauvetage. Toutefois, voler dans des environnements complexes exige un haut niveau d’agilité, ce qui implique par exemple la capacité de déclencher des manœuvres agressives pour esquiver les obstacles: Les caméras et algorithmes d’intelligence artificielle conventionnels n’ont pas ces capacités. Dans ce projet, nous proposerons une solution associant de manière bio-inspirée une dynamique rapide de détection visuelle et de stabilisation. Nous intégrerons ces différents aspects dans un système neuromorphique événementiel de bout en bout. La clé de cette approche est l’optimisation des délais du système par traitement prédictif. Ceci permettra de voler indépendamment, sans aucune intervention de l’utilisateur. Notre objectif à plus long terme est de satisfaire ces besoins avec un minimum d’énergie et de fournir des solutions novatrices aux défis des algorithmes traditionnels d’IA.

Our system is divided into 3 units to process visual inputs (ATIS) until the rotors: the Camera, Processor and Motor units. Each represents respectively multi-channel feature maps ($C_i$), an estimate of the depth-of-field ($P$) and a navigation map, for instance time-of-contacts on a polar map ($M$). Compared to a discrete-time pipeline, we will design an integrated, back-to-back event-driven system based on a fast, two-way processing between the C, P and M units. Event-driven, feed-forward and feed-back communications are denoted respectively in yellow, black and red. Notice the attention module $A$ from $P$ to $C$ and the feed-back of navigation information from $M$ and the IMU to $P$.

Abstract

Autonomous aerial robots would be essential tools in search and rescue operations. But flying in complex environments requires a high level of agility, which implies the ability to initiate aggressive maneuvers to avoid obstacles: Conventional AI cameras and algorithms do not have these capabilities. In this project, we propose a solution that will integrate bio-inspired rapid visual detection and stabilization dynamics into an end-to-end event based neuromorphic system. The key to this approach will be the optimization of delays through predictive processing. This will allow these robots to fly independently, without any user intervention. Our longer-term goal is to meet the requirements with very little power and provide innovative solutions to the challenges of traditional AI algorithms.

Acknowledgement

This work was supported by ANR project “AgileNeuRobot” N° ANR-20-CE23-0021.

ANR PRIOSENS (2021/2025)

Mon, 27 Apr 2020 00:00:00 +0000

A fundamental goal of systems neuroscience is to describe how sensory inputs are integrated and guide an animal’s behavior. To be able to integrate these inputs, early sensory systems have developed selectivities for specific stimulus features that allow them to analyze the inputs using these features as basis. We aim to uncover how disparate motion signals are integrated to produce a global percept of motion, and to understand the conditions in which such integration fails. Our proposal reflects the fact that adaptive behaviors in complex environments face numerous challenges, from processing noisy and uncertain visual motion information to predict future events on target trajectory contingencies and its interactions with a dynamic, cluttered environment. We propose to use dynamic inference as an efficient theoretical framework to understand how the brain integrates Prior knowledges elaborated from statistical regularities of natural environments with different sources of information across different time scales in order to extract relevant motion information from the sensory flow and predict future events or actions. The smooth pursuit system is an excellent probe of such hierarchical dynamical inferences from target motion computation to target trajectory prediction. In marmosets, we have access to populations of neurons in pivotal cortical areas along the occipito-parieto- frontal network that have been identified in non-human and human primates. We seek to uncover a unifying empirical and theoretical framework to capture inference across different time scales.

With Guilhem Ibos, Guillaume Masson & Nicholas Priebe.

Aim 3, modelling behavioural and neuronal data within the active inference framework

Type de contrat : CRCNS US-French Research Proposal - ANR-CRCNS-2020
Durée: 4 ans, à partir du 1er novembre 2020 - prolongatio au 10/2025
Budget total (partenaire français): 341 k€
to be recruited: Post-doctoral fellow: A post-post-doctoral fellow in computational neuroscience will be recruited. With a 2-5 years experience, salary cost is of 52K€/year, for 2 years (total: 104K€).
Coordinateur Scientifique : MONTAGNINI, Anna & PERRINET Laurent (UMR7289)
Partenaire(s) : AGENCE NATIONALE DE LA RECHERCHE
Responsable Scientifique INT : MASSON Guillaume (UMR7289)

Acknowledgement

This work was supported by ANR project “PRIOSENS” N° ANR-20-NEUC-0002.

ANR ShootingStar (2021/2024)

Mon, 27 Apr 2020 00:00:00 +0000

The natural visual environments in which we have evolved have shaped and constrained the neural mechanisms of vision. Rapid progress has been made in recent years in understanding how the retina and visual cortex are specifically adapted to processing natural scenes.1–3 However, studies in this research tradition have mainly addressed the processing of natural images in the spatial domain. Although the processing of temporal properties of visual stimuli is just as important as spatial properties, stimuli with naturalistically valid temporal dynamics have not been sufficiently investigated. Although objects and creatures we view undergo a variety of intrinsic movements, probably the most common motions on the retina are image shifts due to our own eye movements: in free viewing in humans, ocular saccades occur about three times every second, shifting the retinal image at speeds of 100-500 degrees of visual angle per second.4 How these very fast shifts are suppressed, leading to clear, accurate and stable representations of the visual scene is an fundamental unsolved problem in visual neuroscience known as saccadic suppression. One reason why this problem is difficult is technological: to make progress we need to visually simulate these fast retinal shifts, but computer displays have been too slow to produce adequate simulations.

In this project we propose a unique convergence between neurophysiology, modeling and psychophysics, aided by recent technological developments. Some of the partners have been at the forefront of recent developments that have led to a realization that moving stimuli lead to traveling waves of activity in primary visual cortex, propagating at speeds similar to those produced by saccades. Other partners have developed detailed models of the retina and primary visual cortex based on multielectrode recordings from the retina and optical imaging of the cortex that have been able to account for these wave phenomena. Finally, another partner recently made psychophysical observations—aided by new, ultrafast computer displays that allow us to realistically simulate saccadic dynamics on a static retina—that show how image dynamics alone can account for saccadic suppression phenomena.

We expect that the convergence of these three research currents and methodologies will lead to rapid progress in understanding how the visual system is adapted to naturalistic dynamics. The psychophysical observations will provide new leads and targets for the neurophysiology and modeling, which in turn may provide detailed neural explanations for the psychophysics. Our main hypothesis is that the neural architectures that have been uncovered in the retina and the primary visual cortex will be revealed as most effective when processing naturalistic, fast stimuli that arise as the consequence of eye movements.

carte d’identité du projet

Durée: 4 ans, à partir du 1er avril 2021
Budget total (partenaire français): 665 k€
Coordinateur Scientifique : Mark WEXLER (CNRS‐INCC)
Partenaire(s) : AGENCE NATIONALE DE LA RECHERCHE
Responsable Scientifique INT : Frédéric Chavane (UMR7289)

Acknowledgement

This work was supported by ANR project “ShootingStar” N° ANR-XX-XXX-XXXX.

APROVIS3D (2019/2023)

Tue, 10 Sep 2019 10:00:00 +0000

Le projet APROVIS3D est lauréat de l’appel à projets 2018 CHIST-ERA :

The APROVIS3D project targetted analog computing for artificial intelligence in the form of Spiking Neural Networks (SNNs) on a mixed analog and digital architecture. The project included field programmable analog array (FPAA) and SpiNNaker applied to a stereopsis system dedicated to coastal surveillance using an aerial robot. Computer vision systems widely rely on artificial intelligence and especially neural network based machine learning, which recently gained huge visibility. The training stage for deep convolutional neural networks is both time and energy consuming. In contrast, the human brain has the ability to perform visual tasks with unrivalled computational and energy efficiency. It is believed that one major factor of this efficiency is the fact that information is vastly represented by short pulses (spikes) at analog – not discrete – times. However, computer vision algorithms using such representation still lack in practice, and its high potential is largely underexploited. Inspired from biology, the project addresses the scientific question of developing a low-power, end-to-end analog sensing and processing architecture of 3D visual scenes, running on analog devices, without a central clock and aims to validate them in real-life situations. More specifically, the project will develop new paradigms for biologically inspired vision, from sensing to processing, in order to help machines such as Unmanned Autonomous Vehicles (UAV), autonomous vehicles, or robots gain high-level understanding from visual scenes. The ambitious long-term vision of the project is to develop the next generation AI paradigm that will eventually compete with deep learning. We believe that neuromorphic computing, mainly studied in EU countries, will be a key technology in the next decade. It is therefore both a scientific and strategic challenge for the EU to foster this technological breakthrough. The consortium from four EU countries offers a unique combination of expertise that the project requires. SNNs specialists from various fields, such as visual sensors (IMSE, Spain), neural network architecture and computer vision (Uni. of Lille, France) and computational neuroscience (INT, France) will team up with robotics and automatic control specialists (NTUA, Greece), and low power integrated systems designers (ETHZ, Switzerland) to help geoinformatics researchers (UNIWA, Greece) build a demonstrator UAV for coastal surveillance (TRL5). Adding up to the shared interest regarding analog based computing and computer vision, all team members have a lot to offer given their different and complementary points of view and expertise. Key challenges of this project will be end-to-end analog system design (from sensing to AI-based control of the UAV and 3D coastal volumetric reconstruction), energy efficiency, and practical usability in real conditions. We aim to show that such a bioinspired analog design will bring large benefits in terms of power efficiency, adaptability and efficiency needed to make coastal surveillance with UAVs practical and more efficient than digital approaches.

Type de contrat : Subvention / Aide
Durée: 3 ans, à partir du 1er avril 2020 (prolongation demandée)
Budget total: 867 k€ , bugdget INT: 150 k€
Partenaire(s) : AGENCE NATIONALE DE LA RECHERCHE
Objet : AAP 2019 - CHIST-ERA “Analog PROcessing of bioinspired VIsion Sensors for 3D reconstruction” ANR-19-CHR3-0008-03
Responsable Scientifique INT : PERRINET Laurent (UMR7289)
“This project has received funding from the European Union’s ERA-NET CHIST-ERA 2018 research and innovation programme under grant agreement No ANR-19-CHR3-0008-03”
Find more on the official website

SpikeAI: laureat du Défi Biomimétisme (2019)

Mon, 15 Apr 2019 10:00:00 +0000

Description

Le projet SpikeAI est lauréat de l’appel à projets 2019 Biomimétisme :

The SpikeAI project targets analog computing for artificial intelligence in the form of Spiking Neural Networks (SNNs). Computer vision systems widely rely on artificial intelligence and especially neural network based machine learning, which recently gained huge visibility. The training stage for deep convolutional neural networks is time-consuming and yields enormous energy consumption. In contrast, the human brain has the ability to perform visual tasks with unrivaled computational and energy efficiency. It is believed that one major factor of this efficiency is the fact that information is vastly represented by short pulses (spikes) at analog –not discrete– times. However, computer vision algorithms using such representation still lack in practice, and its high potential is largely underexploited. Inspired from biology, the project addresses the scientific question of developing a low-power, end-to-end analog sensing and processing architecture. This will be applied on the particular context of a field programmable analog array (FPAA) applied to a stereovision system dedicated to coastal surveillance using an aerial robot of 3D visual scenes, running on analog devices, without a central clock and to validate them in real-life situations. The ambitious long-term vision of the project is to develop the next generation AI paradigm that will at term compete with deep learning. We believe that neuromorphic computing, mainly studied in EU countries, will be a key technology in the next decade. It is therefore both a scientific and strategic challenge for France and EU to foster this technological breakthrough. This call will help kickstart collaboration within this European consortium to help leverage the chance to successfully apply to future large-scale grant proposals (e.g. ANR, CHIST-ERA, ERC).

Get more information.

outcomes

The main goal was mainly to build a network of actors and to answer to relevant calls in the field of biomimetic research. We have communicated through the diffusion of computational frameworks and actions which are gathered online @ https://github.com/SpikeAI.

Summary of the actions taken:

We had a APROVIS3D FPP meeting April 23-24, 2019 in Lille with all partners. The call could support the travel of the 5 participants from outside Lille. During these two days, we had a first day to know each other better and a second day devoted to writing the grant proposal.
With the help of this call, we could kickstart a collaboration within this European consortium which helped successfully achieve a large-scale grant proposal (CHIST-ERA : https://laurentperrinet.github.io/grant/aprovis-3-d/ ).
We organized a tutorial on deep learning during the GDR Vision, see https://github.com/SpikeAI/2019-10-10_ML-tutorial and https://laurentperrinet.github.io/post/2019-10-10_gdrvision-atelier/
We could invite a major actor of the modeling of biomimetic computations, Ryad Benosman, to the GDR vision meeting. The call could support his travel and accommodation and was acknowledged in all communications (see https://gdrvision2019.sciencesconf.org/resource/page/id/6).

ANR BalaV1 (2013/2016)

Wed, 27 Apr 2016 00:00:00 +0000

ANR BalaV1: Balanced states in area V1 (2013/2016)

Official website

In carnivores and primates the orientation selectivity (OS) of the cells in the primary visual cortex (V1) is organized in maps in which preferred orientations (POs) of the cells change gradually except near “pin- wheels”, around which all orientations are present. Over the last half-century the mechanism for OS has been hotly debated. However the theories that purport to explain OS have almost all considered cortical networks in which the neurons receive input preferentially from cells with similar PO. Such theories certainly capture the connectivity for neurons in orientation domains where neurons are surrounded by other cells with similar PO. However this does not necessarily hold near pinwheels: because of the discontinuous change in orientation preference at the pinwheel, neurons in this area are surrounded by cells of all preferred orientations. Thus if the probability of connection is solely dependent on anatomical distance, the inputs that these neurons receive should represent all orientations by roughly the same amount. Thus one may expect that the response of the cells near pinwheels should hardly vary with orientation, in contrast to experimental data. As a result, the common belief is that, at least near pinwheels, the connectivity depends also on the differences between preferred orientation. The situation near pinwheels in V1 of carnivores and primates is similar to that in the whole of V1 of rodents. In these species, neurons in V1 are OS but the network does not exhibit an orientation map and the surround of the cells represents all orientations roughly equally. In a recent theoretical paper (Hansel and van Vreeswijk 2012) we have demonstrated that in this situation, the response of the cells can still be orientation selective provided that the network operates in the balanced regime. Here we hypothesize that V1 with an orientation map operates in the balanced regime and therefore neurons can exhibit OS near pinwheels even in the absence of functional specific connectivity. The goal of this interdisciplinary project is to investigate whether the “balance hypothesis” holds for layer 2/3 in V1 of primate and carnivore and whether the functional organization observed in that layer can be accounted for without feature specific connectivity. We will combine modeling and experiments to investigate how the response of the neurons – the mean firing, the mean voltage, the inhibitory and excitatory conductances and importantly, the power spectrum of their fluctuations – vary with the location in the map, and also how a population of neurons – LFP, voltage-sensitive dye imaging or 2 photons – is affected by the various para- meters used to test the system. Whether V1 indeed operates in the balanced regime in more realistic conditions will be further investigated by determining how the local network responds to visual stimuli beyond the classical receptive field. We will investigate this issue in models of layer 2/3 representing multiple hyper- columns to characterize center-surround interactions and their dependence on the long-range connectivity. This will provide us with predictions for center-surround interactions for cells near pinwheels and in orientation domains. These predictions will be tested experimentally.

The proposed project is new and ambitious. It aims at building a comprehensive and coherent understand- ing of the physiology of V1 layer 2/3 on several spatial scales from single cells to several hypercolumns and to account for this in mechanistic models. To accomplish these ambitious aims, we propose a combination of experimental and computational studies that take advantage of the unique strengths and the complementarity of expertise of 3 research teams. The Paris team has extensive experience in large-scale modeling of V1. The Toulouse and Marseille teams master both intra- and extracellular electrophysiology. In addition, the Marseille team is expert in microscopic and mesoscopic imaging techniques in V1.

Acknowledgement

This work was supported by ANR project "BalaV1" N° ANR-13-BSV4-0014-02.

ANR CausaL (2018/2020)

Wed, 27 Apr 2016 00:00:00 +0000

With Andrea Brovelli (INT), Mateus Joffily (GATE)…

See https://anr.fr/Project-ANR-18-CE28-0016

Humans have an extraordinary capacity to infer cause-effect relations. In particular, we excel in forming beliefs about the causal effect of actions. Causal learning provides the basis for rational decision-making and allows people to engage in meaningful life and social interactions. Causal learning is a form of goal-directed learning, defined as the capacity to rapidly learn the consequence of actions and to select behaviours according to goals and motivational state. This ability is based on internal models of the consequence of our behaviors and relies on learning rules determined by the contingency between actions and outcomes. At a first approximation, contingency ΔP is operationalized as the difference between two conditional probabilities: i) P(O|A), the probability of outcome O given action A; ii) P(O|¬A), the probability of the outcome when the action is withheld. In everyday life, people perceive their actions as causing a given outcome if the contingency is positive, whereas they perceive them as preventing it if negative; when P(O|A) and P(O|¬A) are equal, people report no causal effect . Despite the centrality of causal learning, a clear understanding of both the internal computations and neural substrates (the so-called cognitive architectures) is currently missing. Our project will therefore address two key questions:

What are the key internal representations of causal beliefs and what are the computational processes that enable their formation during learning?
How are internal representations and computational processes implemented in the brain? CausaL will address these two objectives through two dedicated research work packages (WPs).

Acknowledgement : This work was supported by ANR project ANR-18-AAPG–“CAUSAL, Cognitive Architectures of Causal Learning”.

ANR Horizontal-V1 (2017/2021)

Wed, 27 Apr 2016 00:00:00 +0000

Description on the official website of the ANR

The Horizontal-V1 project aimed at understanding the emergence of sensory predictions linking local shape attributes (orientation, contour) to global indices of movement (direction, speed, trajectory) at the earliest stage of cortical processing (primary visual cortex, i.e. V1). We studied how the long-distance “horizontal” connectivity, intrinsic to V1 and the feedback from higher cortical areas contribute to a dynamic processing of local-to-global features as a function of the context (eg displacement along a trajectory; during reafference change induced by eye-movements…). We characterized the dynamic processes based on lateral propagation intra-V1, through which spatio-temporal inferences (continuous movement or apparent motion sequences) facilitating spatial (“filling-in”) or positional (“flash-lag”) future expected responses may be generated.

Our main contributions to the project:

Victor Boutin, Angelo Franciosini, Frédéric Chavane, Franck Ruffier, Laurent U Perrinet (2021). Sparse Deep Predictive Coding captures contour integration capabilities of the early visual system. PLoS Computational Biology.
PDF Cite DOI Code URL arXiv
Victor Boutin, Angelo Franciosini, Franck Ruffier, Laurent U Perrinet (2020). Effect of top-down connections in Hierarchical Sparse Coding. Neural Computation.
PDF Cite DOI URL arXiv
Victor Boutin, Angelo Franciosini, Frédéric Chavane, Laurent U Perrinet (2022). Pooling in a predictive model of V1 explains functional and structural diversity across species. PLoS Computational Biology.
Cite DOI URL bioRxiv
Alberto Arturo Vergani, Laurent U Perrinet (2021). Simulating anticipatory activity in a 1D Spiking Neural Network Model. Bernstein Conference 2021.
Cite DOI
Alberto Arturo Vergani, Laurent U Perrinet (2021). Neural Turing Patterns.
Project

WP3 - Design of novel visual paradigms, probabilistic model of V1 and data-driven simulations - co lead UNIC-INT.

Objectives : This WP will have two primary goals. The first one is theoretically driven, and for sake of simplicity will ignore the dynamic features of neural integration (as expected from a statistical model of image analysis). Binding the different features of visual objects at the local scale (contours) as well as a more global level involves understanding the statistical regularities of the sensory inflow. In particular, titrating the predictions that can be done at the statistical level can be seen as a first pass to better search for critical parameters constraining the network behaviour. From these, we will build probabilistic predictive models optimized for edge co-occurrence classification and generate novel visual statistics 1) which obey rules imposed by the functional horizontal connectivity anisotropies, such as co- circularity, and 2) which facilitate binding in the orientation domain, such as log-polar planforms. These statistics generated in the first half of the grant will be implemented and tested experimentally in the second half of the grant. The second one is more data-driven (as well as phenomenological for feedback from higher cortical areas, since it will not be explored in the grant). Since model fitting will depend on close interactions with WP1 and WP2 measurements, it will be done in the second half of the grant.

WP3-Task 1: Theoretically oriented workplan – Lead INT (Laurent Perrinet)

WP3-Task 1.1 - theory : we will exploit our current expertise in integrating these statistics in the form of probabilistic models to make predictions both at the physiological and modelling levels. First, we will take advantage of our previous work on the quantification of the association field in different classes of natural images (Perrinet & Bednar, 2015). Using an existing library (https://github.com/bicv/SparseEdges), we will use the sparse representation of static natural images to compute histograms of edge co-occurrences. Using an existing algorithm for unsupervised learning (https://github.com/bicv/SparseHebbianLearning), we will learn the different independent components of edge co-occurrences. Such an algorithm fits well a traditional deep-learning convolutional neural network, but, in addition, will include constraints imposed by intra-layer horizontal connectivity. We expect that relevant features will be co-linear or co-circular pairs of edges, but also T-junctions or end-stopping features.
WP3-Task 1.2 - image/film synthesis : We have previously found that random synthetic textures, coined “Motion Clouds”, can be used to quantify V1 implication in visual motion perception (Leon et al, 2012; Simoncini et al, 2012). Recently, the INT and UNIC, partners proved mathematically that these stimuli were optimal with respect to some common geometrical transformations, such as translation, zoom or rotations (Vacher et al, 2015). A main characteristic of these textures is to be generated with a maximally entropic arrangement of elementary textures (so-called textons).

** Informed by the generative model of edge co-occurrences studied in subtask 1, we will be able to extend the family of motion cloud stimuli (Leon et al, 2012; Simoncini et al, 2012) to include joint dependencies between different elements in position or orientation. An exact solution to this problem is hard to achieve as it involves a combinatorial search of all possible combinations of pairs of edges. However, numerous variational approaches are possible and fit well with our probabilistic framework. We will use the convolutional neural network described above but using a back-propagating stream to generate different images. Such a representation will then be optimized using an unsupervised learning method. This is similar to the process used in Generative Adversarial Networks in deep-learning architectures (Radford et al, Archives). ** Finally, the regularities observed in static images will be extended to dynamical scenes by observing that a co-occurrence can be implemented by simple geometrical operations as they are operated in time. For instance a co-circularity is easily described as the set of smooth roto-translational transformations of an edge in time using the group of Galilean transformations (Sarti and Citti, 2006). This theory calls for a first prediction to understand the set of whole possible spatio-temporal co-occurrences of edges as geodesics in the lifted space of all possible trajectories. We predict that such decomposition should allow us to better understand the different classes of features that emerged in the first task.

WP3-Task 1.3 - Feedback of theory on experimentation : An essential aspect of this work would be to apply these stimuli in neurophysiological experiments and in the modelling. In particular, the ability to select different types of dependencies from the different classes learned above (for instance, co-circularities of a certain curvature range) will make it possible to evaluate the relative contribution of different components of the contextual information. This justifies the fact that the WP3 post-doctoral fellow should have the mobility (between INT and UNIC) and multi-disciplinar profile (theoretical and experimental) to perform this task.
WP3-Task 1.4 - Generic modelling : These various subtasks will allow us to determine the hierarchy of critical features relevant to describe the full statistics of the space of spatio-temporal edge co-occurrences. Indeed, in static images, we will be able to find independent component in the histograms of edge co-occurrences between metric aspect (distance or scale between edge) from configurational aspects (difference of angle or co-circularity angle).

Similarly, we expect to see that the different independent features should decompose at various scales both in space and in time. For instance, we expect configurational aspects to be more local while aspects related to a motion (Perrinet and Masson, 2012; Khoei et al, 2016) or global shape (form) should be more global. This translates into a probabilistic hierarchical model that would combine dependencies from different cues. In particular, through the emergence of differential pathways for form and motion. These quantitative predictions should finally be confronted at the modelling and neurophysiological levels.

WP3-Task 2 : Data-driven comprehensive model of V1 – Co-lead UNIC and INT

The second task is more data-driven (as well as phenomenological for the feedback circuit part, since largely unknown). Since simulations will depend on close interactions with WP1 and WP2 measurements, it will be developed by the WP3-Post-Doc in the second half of the grant. It will benefit from existing structuro-functional models addressing separately two distinct levels of neural integration, microscopic (conductance-based in Kremkow et al, 2016; Antolik et al, submitted, Chariker et al, 2016) and mesoscopic (VSD-like mean field in Rankin and Chavane, 2017). Efforts will be made to merge these models to fit - in a unified multiscale biologically realistic model - the cellular and VSD data targeting critically horizontal propagation. The parameterization should be flexible enough to produce a generic cortical architecture accounting possibly for species-specificity (Antolik for cat; Chaliker for monkey)

Acknowledgement

This work was supported by ANR project “Horizontal-V1” N° ANR-17-CE37-0006.

ANR PredictEye (2018/2020)

Wed, 27 Apr 2016 00:00:00 +0000

The objectives of PREDICTEYE is to rigorously test and define the functional and neurophysiological grounds of probabilistic oculomotor internal models by investigating the multiple timescales at which the trajectory of a moving target is learned and represented in a probabilistic framework (Aim #1). Second, we will investigate the role of (pre)frontal oculomotor networks in building such probabilistic representations and their impact upon two of their downstream neural targets of the brainstem premotor centers (superior colliculus for saccades; NRTP for pursuit) (Aim #2). Our third objective is to model and simulate the dynamics of target motion prediction and eye movement performance. A key question is to unveil how probabilistic information about target timing and motion (i.e. direction and speed) is sampled over trial history by neuronal populations and integrated with Prior knowledge (i.e. sequence properties and rules of conditional probabilities) in order to coordinate saccades and pursuit and optimize their precisions (Aim #3).

ANR-2018 Project PREDICTEYE - Agence Nationale de la Recherche (2018-2022). This project starts november 2018, for 4 years. It will investigate the neural networks in human volunteers supporting anticipatory pursuit eye movements using magnetic transcranial stimulation (TMS) to perturb frontal networks during ocular tracking of predictable targets. In complementary studies conducted in macaque monkeys, perturbations will be applied pharmacologically to subcortical targets of this frontal network, namely superior colliculus and NTRP, a pontine nucleus relaying information to the pursuit networks of the cerebellum. The project involves 4 CNRS permanent researchers from the INVIBE team headed by G Masson. The funding is 507K€ for 4 years. PI: G Masson, co-PI: A Montagnini, L Perrinet, L Goffart
related grant by the Fondation pour le Recherche Médicale, under the program Équipe FRM (DEQ20180339203/PredictEye/PI: G Masson/ A. Montagnini and L. Perrinet as participants).

Acknowledgement

This work was supported by ANR project "PredictEye" ANR-XXXX.

ANR REM (2013/2016)

Wed, 27 Apr 2016 00:00:00 +0000

Reinforcement learning theory provides a general conceptual framework to account for behavioral changes. Recently the idea that reinforcement may be used to explain learning in motor responses has emerged. In particular, there is a growing interest in studying the effects of reinforcement learning in arm movements trajectories (Dam, Kording, & Wei, 2013), pointing movements (Trommershauser, Landy, & Maloney, 2006), or eye movements (Madelain, Champrenaut, & Chauvin, 2007; Madelain & Krauzlis, 2003b; Madelain, Paeye, & Wallman, 2011; Sugrue, Corrado, & Newsome, 2004; Takikawa, Kawagoe, Itoh, Nakahara, & Hikosaka, 2002; Xu-Wilson, Zee, & Shadmehr, 2009). However, and despite these few seminal studies, much is still unknown about both the details of the effects of reward on motor control and the underlying mechanisms. This proposal aims at a better understanding of how skilled motor responses are learned focusing on voluntary eye movements.

Although learning is often regarded as a restricted period of time during which a behavior undergo some changes we view learning as a continuously ongoing process. In the case of motor control every instance of a behavior is followed by some consequences that will affect some dimensions of the future response. These changes will in return affect the functional relations with the environment and this feedback process continues through lifetime. Therefore we do not regard motor learning as a special phase that allows the emergence of a particular motor response but as a continuous adaptation to the changes within the organism that affect the functional relations with her environment. This distinction is important because the learning situations that are experimentally tested over a short period of time may then be viewed as a condensed version of motor learning in the real life: the same adaptive processes are responsible for the changes in the response in both situations.

An important aspect of this fundamental research project is that the theoretical propositions addressed provide a new view on motor learning that departs from conventional wisdom. We expect to gain considerable knowledge on learning by constructing new experimental paradigms to collect behavioural data, implementing new learning models based on Bayesian theories and testing dynamical mathematical models of behavioural changes. Whichever way the results turn out, we anticipate that these studies will provide a better understanding of motor learning and provide a well-defined and solid framework for studying other forms of motor plasticity. If eye movement learning follows the rules of other operant responses (i.e. responses reinforced by their consequences), this will constitute a minor revolution in the study of motor control, both at the behavioral and neural levels, with important implications for the understanding of plasticity in other motor systems.

Acknowledgement

This work was supported by ANR project ANR-13-APPR-0008 "ANR R.E.M.".

ANR SPEED (2013/2016)

Wed, 27 Apr 2016 00:00:00 +0000

Measuring speed and direction of moving objects is an essential computational step in order to move our eyes, hands or other body parts with respect to the environment. Whereas encoding and decoding of direction information is now largely understood in various neuronal systems, how the human brain accurately represents speed information remains largely unknown. Speed tuned neurons have been identified in several early cortical visual areas in monkeys. However, how such speed tuning emerges is not yet understood. A working hypothesis is that speed tuned neurons nonlinearly combine motion information extracted at different spatial and temporal scales, taking advantage of the statistical spatiotemporal properties of natural scenes. However, such pooling of information must be context dependent, varying with the spatial perceptual organization of the visual scenes. Furthermore, the population code underlying perceived speed is not elucidated either and therefore we are still far from understanding how speed information is decoded to drive and control motor responses or perceptual judgments.

Recently, we have proposed that speed estimation is intrinsically a multi-scale, task-dependent problem (Simoncini et al., Nature Neuroscience 2012) and we have defined a new set of motion stimuli, constructed as random phase dynamical textures that mimic the statistics of natural scenes (Sanz-Leon et al., Journal of Neurophysiology 2012). This approach has proved to be fruitful to investigate nonlinear properties of motion integration.

The current proposal brings together psychophysicists, oculomotor scientists and modelers to investigate speed processing in human. We aim at expanding this framework in order to understand how tracking eye movements and motion perception can take advantage of multiple scale processing for estimating target speed. We will design sets of high dimensional stimuli by extending our generative model. Using these natural-statistics stimuli, we will investigate how speed information is encoded by computing motion energy across different spatial and temporal filters. By analysing both perceptual and oculomotor responses we will probe the nonlinear mechanisms underlying the integration of the outputs of multiple spatiotemporal filters and implement these processes in a refined version of our model. Furthermore, we will test our working hypothesis that in natural scenes such nonlinear integration provides precise and reliable motion estimates, which leads to efficient motion-based behaviors. By comparing tracking responses with perception, we will also test a second critical hypothesis, that nonlinear speed computations are task-dependent. In particular, we will explore the extent to which the geometrical structures of visual scenes are decisive for perception beyond the motion energy computation used for early sensorimotor transformation. Finally we will investigate the role of contextual and extra-retinal, predictive information in building an efficient dynamic estimate of objects’ speed for perception and action.

Acknowledgement

This work was supported by ANR project "ANR Speed" ANR-13-BSHS2-0006.

ANR TRAJECTORY (2016/2019)

Wed, 27 Apr 2016 00:00:00 +0000

Global motion processing is a major computational task of biological visual systems. When an object moves across the visual field, the sequence of visited positions is strongly correlated in space and time, forming a trajectory. These correlated images generate a sequence of local activation of the feed-forward stream. Local properties such as position, direction and orientation can be extracted at each time step by a feed-forward cascade of linear filters and static non-linearities. However such local, piecewise, analysis ignores the recent history of motion and faces several difficulties, such as systematic delays, ambiguous information processing (e.g., aperture and correspondence problems61) high sensitivity to noise and segmentation problems when several objects are present. Indeed, two main aspects of visual processing have been largely ignored by the dominant, classical feed-forward scheme. First, natural inputs are often ambiguous, dynamic and non-stationary as, e.g., objects moving along complex trajectories. To process them, the visual system must segment them from the scene, estimate their position and direction over time and predict their future location and velocity. Second, each of these processing steps, from the retina to the highest cortical areas, is implemented by an intricate interplay of feed-forward, feedback and horizontal interactions1. Thus, at each stage, a moving object will not only be processed locally, but also generate a lateral propagation of information. Despite decades of motion processing research, it is still unclear how the early visual system processes motion trajectories. We, among others, have proposed that anisotropic diffusion of motion information in retinotopic maps can contribute resolving many of these difficulties25 13. Under this perspective, motion integration, anticipation and prediction would be jointly achieved through the interactions between feed-forward, lateral and feedback propagations within a common spatial reference frame, the retinotopic maps.

Addressing this question is particularly challenging, as it requires to probe these sequences of events at multiple scales (from individual cells to large networks) and multiple stages (retina, primary visual cortex (V1)). “TRAJECTORY” proposes such an integrated approach. Using state-of-the-art micro- and mesoscopic recording techniques combined with modeling approaches, we aim at dissecting, for the first time, the population responses at two key stages of visual motion encoding: the retina and V1. Preliminary experiments and previous computational studies demonstrate the feasibility of our work. We plan three coordinated physiology and modeling work-packages aimed to explore two crucial early visual stages in order to answer the following questions: How is a translating bar represented and encoded within a hierarchy of visual networks and for which condition does it elicit anticipatory responses? How is visual processing shaped by the recent history of motion along a more or less predictable trajectory? How much processing happens in V1 as opposed to simply reflecting transformations occurring already in the retina?

The project is timely because partners master new tools such as multi-electrode arrays and voltage-sensitive dye imaging for investigating the dynamics of neuronal populations covering a large segment of the motion trajectory, both in retina and V1. Second, it is strategic: motion trajectories are a fundamental aspect of visual processing that is also a technological obstacle in computer vision and neuroprostheses design. Third, this project is unique by proposing to jointly investigate retinal and V1 levels within a single experimental and theoretical framework. Lastly, it is mature being grounded on (i) preliminary data paving the way of the three different aims and (ii) a history of strong interactions between the different groups that have decided to join their efforts.

The Marseille team

Frédéric Chavane (DR, CNRS, NEOPTO team) is working in the field of vision research for about 20 years with a special interest in the role of lateral interactions in the integration of sensory input in the primary visual cortex. His recent work suggest that lateral interactions mediated by horizontal intracortical connectivity participates actively in the input normalization that controls a wide range of function, from the contrast-response gain to the representation of illusory or real motion. His expertise range from microscopic (intracellular recordings) to mesoscopic (optical imaging, multi-electrode array) recording scales in the primary visual cortex of anesthetized and awake behaving animals.
Laurent Perrinet (CR, CNRS, NEOPTO team). His scientific interests focus on bridging computational understanding of neural dynamics and low-level sensory processing by focusing on motion perception. He is the author of papers in machine learning, computational neuroscience and behavioral psychology. One key concept is the use of statistical regularities from natural scenes as a main drive to integrate local neural information into a global understanding of the scene. In a recent paper that he coauthored (in Nature Neuroscience), he developed a method to use synthesized stimuli targeted to analyze physiological data in a system-identification approach.
Ivo Vanzetta (CR, CNRS, NEOPTO team). His scientific interests focus on how to optimally use photonics-based imaging methods to investigate visual information processing in low-level visual areas, in the anesthetized and awake animal (rodent & primate). As can be seen from his bibliographic record, these methods include optical imaging of intrinsic signals and voltage sensitive dyes and, recently, 2 photon microscopy. Finally I. Vanzetta has an ongoing collaboration with L. Perrinet on the utilization of well-controlled, synthesized nature-like visual stimuli to probe the response characteristics of the primate’s visual system (Sanz-Leon & al. 2012).

Progress meeting ANR TRAJECTORY

Time January 15th, 2018
Location INT
General presentation of the grant, see Anr TRAJECTORY
Overview of my current projects https://laurentperrinet.github.io/sciblog/files/2017-11-15_ColloqueMaster.html
MotionClouds with trajectories https://laurentperrinet.github.io/sciblog/posts/2018-01-16-testing-more-complex-trajectories.html or https://laurentperrinet.github.io/sciblog/posts/2018-11-13-testing-more-complex-trajectories.html

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

Acknowledgement

This work was supported by ANR project “TRAJECTORY” N° ANR-15-CE37-0011.

DOC2AMU (2016/2019)

Wed, 27 Apr 2016 00:00:00 +0000

DOC2AMU is co-funded by the prestigious Marie Skłodowska-Curie COFUND action within the H2020 Research and Innovation programme of the European Union and by the Regional Council of Provence-Alpes-Côte d’Azur, with a contribution from A*MIDEX Foundation.

Within this programme, the PhD fellows will sign a three-year work contract with one of the 12 Doctoral Schools of AMU. Numerous advantages

These PhD fellowships are remunerated above that of a standard French PhD contract with a gross monthly salary of 2600 € and a gross monthly mobility allowance of 300 €, which after standard deductions will amount to a net salary of approximately 1625€/month (net amount may vary slightly). A 500€ travel allowance per year and per fellow is also provided for the fellows to travel between Marseille and their place of origin. Tailored training and personalised mentoring: Fellows will define and follow a Personal Career Development Plan at the beginning of their Doctoral thesis and will have access to a variety of training options and workshops. Financial support for international research training and conferences participations. A contribution to the research costs will be provided for the benefit of the fellow.

“This work was supported by the Doc2Amu project which received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No 713750. Projet cofinancé par le Conseil Régional Provence-Alpes-Côte d’Azur. Projet cofinancé par le Conseil Régional Provence-Alpes-Côte d’Azur, la commission européenne et les Investissements d’Avenir.”

PhD ICN (2017 / 2021)

Wed, 27 Apr 2016 00:00:00 +0000

Description

The Ph.D. program in Integrative and Clinical Neuroscience (Aix-Marseille University) is offering in 2017 three Ph.D. scholarships to Master students graduated from highly ranked international universities (outside France). We were awarded with one PhD position for Angelo Franciosini at the “Institut de Neurosciences de la Timone” (team “Inference and Visual Behavior”), CNRS, Marseille (France) to study trajectories in natural images and the sensory processing of contours.

##Funding

This project is funded by the Aix-Marseille Université, which was awarded the prestigious status of “Excellence Initiative” (A*MIDEX) by the French Government and considering interdisciplinary studies as one of its main axes of growth. Within this program, the PhD fellow will sign a three-year work contract. They will enroll the ICN PhD program offering personalized follow-up to the students, a wide spectrum of scientific and professional training activities including specialized courses and career development activities and interactions with multi-disciplinary researchers at Aix-Marseille University and top world-wide visiting speakers, in a vibrant international community of students.

Acknowledgement

This work has received support from the French government under the Programme Investissements d’Avenir, Initiative d’Excellence d’Aix-Marseille Université via A*Midex (AMX-19-IET-004) and ANR (ANR-17-EURE-0029) funding.

BrainScaleS (2011/2014)

Mon, 27 Apr 2015 00:00:00 +0000

List of publications that were funded by European Union’s project Number FP7-269921, “BrainScales”.

CODDE (2008/2012)

Mon, 27 Apr 2015 00:00:00 +0000

The CODDE network studies the links between sensory input, brain activity and motor output. It does this by combining behavioural techniques, brain imaging, movement recording and computational modelling.

FACETS (2006/2010)

Mon, 27 Apr 2015 00:00:00 +0000

List of publications that were funded by the FACETS project (more info).

also available on the FACET’s website

FACETS-ITN (2010/2013)

Mon, 27 Apr 2015 00:00:00 +0000

FACETS ITN project (EU funding, grant number 237955) is a ‘Marie-Curie Initial Training Network’ involves 15 groups at European Research Universities, Research Centers and Industrial Partners in 6 countries. 22 Ph.D. Positions are funded in the FACETS-ITN project in the following scientific work areas: Neurobiology of Cells and Networks, Modelling of Neural Systems, Neuromorphic Hardware, Neuro-Electronic Interfaces, Computational Principles in Neural Architectures, Mechanisms of Learning and Plasticity.

PACE-ITN (2015/2019)

Mon, 27 Apr 2015 00:00:00 +0000

The PACE ITN project involved over 50 researchers spread across 10 full and 5 associated partners, from academia and the private sector, established in 7 different European and Associated countries, the PACE network gathers a broad range of expertise from experimental psychology, cognitive neurosciences, brain imaging, technology and clinical sciences.

The PACE Project has received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie grant agreement No 642961

Grants & Funding | Laurent Perrinet

MesoCentre (2018/2026)

ANR MarmoCatch (2023-10/2029-04)

Acknowledgement

Emergences (2023 / 2027)

Latest news

Description of the “Emergences” project

Description of the PhD project (WP1): Focus of attention: a sensory-motor task for energy reduction in unsupervised spiking neural networks

Key figures

Polychronies (2022 / 2025)

Description

Latest news

ANR ACES (2022/2026)

Fiche d’identité

Acknowledgement

ANR RubinVase (2021/2024)

carte d’identité du projet

Acknowledgement

ANR AgileNeuRobot (2021/2025)

Fiche d’identité

Résumé

Abstract

Acknowledgement

ANR PRIOSENS (2021/2025)

Aim 3, modelling behavioural and neuronal data within the active inference framework

Acknowledgement

ANR ShootingStar (2021/2024)

carte d’identité du projet

Acknowledgement

APROVIS3D (2019/2023)

SpikeAI: laureat du Défi Biomimétisme (2019)

Description

outcomes

ANR BalaV1 (2013/2016)

ANR BalaV1: Balanced states in area V1 (2013/2016)

ANR CausaL (2018/2020)

ANR Horizontal-V1 (2017/2021)

Our main contributions to the project:

WP3 - Design of novel visual paradigms, probabilistic model of V1 and data-driven simulations - co lead UNIC-INT.

WP3-Task 1: Theoretically oriented workplan – Lead INT (Laurent Perrinet)

WP3-Task 2 : Data-driven comprehensive model of V1 – Co-lead UNIC and INT

Acknowledgement

ANR PredictEye (2018/2020)

ANR REM (2013/2016)

ANR SPEED (2013/2016)

ANR TRAJECTORY (2016/2019)

The Marseille team

Progress meeting ANR TRAJECTORY

Acknowledgement

DOC2AMU (2016/2019)

PhD ICN (2017 / 2021)

Description

Acknowledgement

BrainScaleS (2011/2014)

CODDE (2008/2012)

FACETS (2006/2010)

FACETS-ITN (2010/2013)

FACETS-ITN: From Neuroscience to neuro-inspired computing (2010/2013)

PACE-ITN (2015/2019)