Thierry Viéville | Institut National de Recherche en Informatique et Automatique (INRIA) (original) (raw)

Papers by Thierry Viéville

We share a new exploratory action known as Artificial Intelligence Devoted to Education (AIDE) la... more We share a new exploratory action known as Artificial Intelligence Devoted to Education (AIDE) launched with the support of Inria (Mnemosyne Team) and Nice INSPE from Cote d´Azur University (LINE laboratory) in connection with the Bordeaux NeuroCampus. It positions artificial intelligence in a somewhat original way ... not [only] as a disruptive tool, but as a formalism allowing to model learning human in problem-solving activities.

The aim of this document is to present the design of an ontology allowing to carry out a modeling... more The aim of this document is to present the design of an ontology allowing to carry out a modeling of the learner, the task and the observables during a learning activity, in order to develop a model applicable to the observed learning analytics which can be exploited to analyze them with computational approaches. The challenge here is to work from a relatively small batch of data (a few dozen to compare with the thousands of data used with classic statistical methods), highly structured, therefore to introduce a maximum of a priori information upstream to the analysis in order the results to be meaningful. The learner is modeled on the basis of knowledge from the educational science and cognitive neurosciences, including machine learning formalisms, in the very precise framework of a task, named CreaCube, related to initiation to computational thinking presented as an open-ended problem, which involves solving a problem and appealing to creativity. This document presents these elements and discusses the exploration and exploitation issues, the different goals (for example of performance, speed or mastery of the task), before relating this to the different types of memory and discussing the basics of problem solving, including engaging in a learning activity. It then describes the very precise construction of an ontology which formalizes this process of task resolution and knowledge construction, taking into account the stimuli received, the discovery of affordances, the setting of hypotheses, clearly distinguished from the notion of belief, without forgetting contextual knowledge. The production is shared as a free and open resource, and both the implications and the perspectives of this pioneering work of formalizing such a human learning task are discussed in conclusion.

Noname manuscript No. (will be inserted by the editor)

Ce texte s'interesse au changement profond qu'a connu la publication scientifique sur une... more Ce texte s'interesse au changement profond qu'a connu la publication scientifique sur une generation de chercheurs. Cette apparition du document numerique s'est accompagnee d'une mise en ligne progressive mais surtout le document a evolue : il est devenu hypertexte, multimedia (son, video) ou logiciel executable sans oublier les donnees d'experimentations et les cours en ligne qui deviennent a leur tour publications. C'est en fait la nature meme de la publication scientifique qui est bouleversee par l'apparition de ces documents 3.0.

L’introduction de la programmation a l’ecole peut etre un levier pour developper la pensee inform... more L’introduction de la programmation a l’ecole peut etre un levier pour developper la pensee informatique en lien avec une demarche de resolution de problemes. Dans ce contexte, nous nous interessons aux differents types d’activites d’apprentissage de la programmation dans le but d’etablir un protocole pour comparer les activites branchees et debranchees a l’ecole, et plus particulierement de voir dans quelle mesure une activite debranchee permet un transfert de competences vers l’apprentissage de la programmation. Nous discutons la methodologie et les resultats en lien aux observations realisees au cours des activites de formation Class’Code.

Our whole society is and will be deeply impacted by digital science and this takes a new qualitat... more Our whole society is and will be deeply impacted by digital science and this takes a new qualitative and quantitative turn with what is named artificial intelligence (AI). We must allow everyone to master, thus understand how all this works. This means computational thinking discovery and machine learning initiation. Directly inspired by the Finnish initiative to train 1% of its population on these subjects and after our success in providing a hybrid formation on computational thinking for teachers and citizens not familiar with computer science where more 45000 persons have been reached, yielding a satisfaction level higher than 90%, we have built and now operate a citizen training in AI in the broad sense, intended to a large public beyond the school domain, with the goal of giving rise to an ubiquitary citizen university in digital science and culture

Oui binaire s’adresse aussi aux jeunes de tous âges que le numerique laisse parfois perplexes. Av... more Oui binaire s’adresse aussi aux jeunes de tous âges que le numerique laisse parfois perplexes. Avec « Petit binaire », osons ici expliquer de maniere simple et accessible un des mecanismes qui nous donne l’impression d’une intelligence artificielle : les reseaux antagonistes generatifs (Generative Adversial Network ou GAN en anglais)

We propose a method that allows the interpretation of the data representation obtained by CNN, th... more We propose a method that allows the interpretation of the data representation obtained by CNN, through introducing prototypes in the feature space, that are later classified into a certain category. This way we can see how the feature space is structured in link with the categories and the related task.

Lecture Notes in Computer Science, 2021

Complex problem solving involves representing structured knowledge, reasoning and learning, all a... more Complex problem solving involves representing structured knowledge, reasoning and learning, all at once. In this prospective study, we make explicit how a reinforcement learning paradigm can be applied to a symbolic representation of a concrete problem-solving task, modeled here by an ontology. This preliminary paper is only a set of ideas while feasibility verification is still a perspective of this work.

Frontiers in Neuroinformatics, 2017

The retina encodes visual scenes by trains of action potentials that are sent to the brain via th... more The retina encodes visual scenes by trains of action potentials that are sent to the brain via the optic nerve. In this paper, we describe a new free access user-end software allowing to better understand this coding. It is called PRANAS (https://pranas.inria.fr), standing for Platform for Retinal ANalysis And Simulation. PRANAS targets neuroscientists and modelers by providing a unique set of retina-related tools. PRANAS integrates a retina simulator allowing large scale simulations while keeping a strong biological plausibility and a toolbox for the analysis of spike train population statistics. The statistical method (entropy maximization under constraints) takes into account both spatial and temporal correlations as constraints, allowing to analyze the effects of memory on statistics. PRANAS also integrates a tool computing and representing in 3D (time-space) receptive fields. All these tools are accessible through a friendly graphical user interface. The most CPU-costly of them have been implemented to run in parallel.

Biosystems & Biorobotics, 2015

Studying and modeling the brain as a whole is a real challenge. For such systemic models (in cont... more Studying and modeling the brain as a whole is a real challenge. For such systemic models (in contrast to models of one brain area or aspect), there is a real need for new tools designed to perform complex numerical experiments, beyond usual tools distributed in the computer science and neuroscience communities. Here, we describe an effective solution, freely available on line and already in use, to validate such models of the brain functions. We explain why this is the best choice, as a complement to robotic setup, and what are the general requirements for such a benchmarking platform. In this experimental setup, the brainy-bot implementing the model to study is embedded in a simplified but realistic controlled environment. From visual, tactile and olfactory input, to body, arm and eye motor command, in addition to vital interoceptive cues, complex survival behaviors can be experimented. We also discuss here algorithmic high-level cognitive modules, making the job of building biologically plausible bots easier. The key point is to possibly alternate the use of symbolic representation and of complementary and usual neural coding. As a consequence, algorithmic principles have to be considered at higher abstract level, beyond a given data representation, which is an interesting challenge.

ABSTRACT In the early visual system, regarding the detection of a visual event, local motion info... more ABSTRACT In the early visual system, regarding the detection of a visual event, local motion information is pre-processed in the Magnocellular pathway, while it has been shown that the Koniocellular pathway also plays an important role, providing a global a-priori estimation about such a kind of information processing. However, the functional interplay between these two parallel pathways remains partially understood. To fill this gap, we investigate, through simulations, the impact of the multi-scale characteristics of the Konio pathway in the Magno-driven thalamocortical system of the mammal. We propose a systemic retino-thalamo-cortico-collicular model, including feedforward, feedback and inhibitory connections. In order to provide a discriminative framework, the model implements minimal algorithms, such as spatio-temporal filters and dynamic neural fields, and is fed with a sequence of images. Our results show that the multi-scale interactions between both pathways, integrating local and larger image cues, account for target discrimination, selection and tracking in the presence of threats or targets. This approach proposes an innovative answer to the interplay issues between these pathways, and it is generalizable to other visuo-motor functions.

We propose a computational approach using a variational specication of the visual front-end, wher... more We propose a computational approach using a variational specication of the visual front-end, where ganglion cells with properties of retinal Konio cells (K-cells), are considered as a network, yielding a mesoscopic view of the retinal process. The variational framework is implemented as a simple mechanism of diusion in a two-layered non-linear ltering mechanism with feedback, as observed in synaptic layers of the retina, while its biological plausibility, and capture functionalities as (i) stimulus adapted response; (ii) non-local noise reduction (i.e. segmentation); (iii) visual event detection, taking several visual cues into account: contrast and local texture, color or edge channels, and motion base in natural images. Those functionalities could be implemented in the biological tissues We use computer vision methods to propose an eective link between the observed functions and their possible implementation in the retinal network base on a two-layers network with nonseparable local spatio-temporal convolution as input, and recurrent connections performing nonlinear diusion before prototype based visual event detection. The numerical robustness of the proposed model has been experimentally checked on real natural images. Finally, we discuss in base of experimental biological and computational results the generality of our description.

Journal of Physiology-Paris, 2011

This article introduces general concepts and definitions related to the notion of asynchronous co... more This article introduces general concepts and definitions related to the notion of asynchronous computation in the framework of artificial neural networks. Using the dynamic field theory as an illustrative example, we explain why one may want to perform such asynchronous computation and how one can implement it since this computational scheme draws several consequences on both the trajectories and the stability of the whole system. After giving an unequivocal definition of asynchronous computation, we present a few practically usable methods and quantitative bounds that can guarantee most of the mesoscopic properties of the system.

Research Report RR- …, 2011

The problem of adjusting the parameters of an event-based network model is addressed here at the ... more The problem of adjusting the parameters of an event-based network model is addressed here at the programmatic level. Considering temporal processing, the goal is to adjust the network units weights so that the outcoming events correspond to what is desired. The present work proposes, in the deterministic and discrete case, a way to adapt usual alignment metrics in order to derive suitable adjustment rules. At the numerical level, the stability and unbiasness of the method is verified.

We share a new exploratory action known as Artificial Intelligence Devoted to Education (AIDE) la... more We share a new exploratory action known as Artificial Intelligence Devoted to Education (AIDE) launched with the support of Inria (Mnemosyne Team) and Nice INSPE from Cote d´Azur University (LINE laboratory) in connection with the Bordeaux NeuroCampus. It positions artificial intelligence in a somewhat original way ... not [only] as a disruptive tool, but as a formalism allowing to model learning human in problem-solving activities.

The aim of this document is to present the design of an ontology allowing to carry out a modeling... more The aim of this document is to present the design of an ontology allowing to carry out a modeling of the learner, the task and the observables during a learning activity, in order to develop a model applicable to the observed learning analytics which can be exploited to analyze them with computational approaches. The challenge here is to work from a relatively small batch of data (a few dozen to compare with the thousands of data used with classic statistical methods), highly structured, therefore to introduce a maximum of a priori information upstream to the analysis in order the results to be meaningful. The learner is modeled on the basis of knowledge from the educational science and cognitive neurosciences, including machine learning formalisms, in the very precise framework of a task, named CreaCube, related to initiation to computational thinking presented as an open-ended problem, which involves solving a problem and appealing to creativity. This document presents these elements and discusses the exploration and exploitation issues, the different goals (for example of performance, speed or mastery of the task), before relating this to the different types of memory and discussing the basics of problem solving, including engaging in a learning activity. It then describes the very precise construction of an ontology which formalizes this process of task resolution and knowledge construction, taking into account the stimuli received, the discovery of affordances, the setting of hypotheses, clearly distinguished from the notion of belief, without forgetting contextual knowledge. The production is shared as a free and open resource, and both the implications and the perspectives of this pioneering work of formalizing such a human learning task are discussed in conclusion.

Noname manuscript No. (will be inserted by the editor)

Ce texte s'interesse au changement profond qu'a connu la publication scientifique sur une... more Ce texte s'interesse au changement profond qu'a connu la publication scientifique sur une generation de chercheurs. Cette apparition du document numerique s'est accompagnee d'une mise en ligne progressive mais surtout le document a evolue : il est devenu hypertexte, multimedia (son, video) ou logiciel executable sans oublier les donnees d'experimentations et les cours en ligne qui deviennent a leur tour publications. C'est en fait la nature meme de la publication scientifique qui est bouleversee par l'apparition de ces documents 3.0.

L’introduction de la programmation a l’ecole peut etre un levier pour developper la pensee inform... more L’introduction de la programmation a l’ecole peut etre un levier pour developper la pensee informatique en lien avec une demarche de resolution de problemes. Dans ce contexte, nous nous interessons aux differents types d’activites d’apprentissage de la programmation dans le but d’etablir un protocole pour comparer les activites branchees et debranchees a l’ecole, et plus particulierement de voir dans quelle mesure une activite debranchee permet un transfert de competences vers l’apprentissage de la programmation. Nous discutons la methodologie et les resultats en lien aux observations realisees au cours des activites de formation Class’Code.

Our whole society is and will be deeply impacted by digital science and this takes a new qualitat... more Our whole society is and will be deeply impacted by digital science and this takes a new qualitative and quantitative turn with what is named artificial intelligence (AI). We must allow everyone to master, thus understand how all this works. This means computational thinking discovery and machine learning initiation. Directly inspired by the Finnish initiative to train 1% of its population on these subjects and after our success in providing a hybrid formation on computational thinking for teachers and citizens not familiar with computer science where more 45000 persons have been reached, yielding a satisfaction level higher than 90%, we have built and now operate a citizen training in AI in the broad sense, intended to a large public beyond the school domain, with the goal of giving rise to an ubiquitary citizen university in digital science and culture

Oui binaire s’adresse aussi aux jeunes de tous âges que le numerique laisse parfois perplexes. Av... more Oui binaire s’adresse aussi aux jeunes de tous âges que le numerique laisse parfois perplexes. Avec « Petit binaire », osons ici expliquer de maniere simple et accessible un des mecanismes qui nous donne l’impression d’une intelligence artificielle : les reseaux antagonistes generatifs (Generative Adversial Network ou GAN en anglais)

We propose a method that allows the interpretation of the data representation obtained by CNN, th... more We propose a method that allows the interpretation of the data representation obtained by CNN, through introducing prototypes in the feature space, that are later classified into a certain category. This way we can see how the feature space is structured in link with the categories and the related task.

Lecture Notes in Computer Science, 2021

Complex problem solving involves representing structured knowledge, reasoning and learning, all a... more Complex problem solving involves representing structured knowledge, reasoning and learning, all at once. In this prospective study, we make explicit how a reinforcement learning paradigm can be applied to a symbolic representation of a concrete problem-solving task, modeled here by an ontology. This preliminary paper is only a set of ideas while feasibility verification is still a perspective of this work.

Frontiers in Neuroinformatics, 2017

The retina encodes visual scenes by trains of action potentials that are sent to the brain via th... more The retina encodes visual scenes by trains of action potentials that are sent to the brain via the optic nerve. In this paper, we describe a new free access user-end software allowing to better understand this coding. It is called PRANAS (https://pranas.inria.fr), standing for Platform for Retinal ANalysis And Simulation. PRANAS targets neuroscientists and modelers by providing a unique set of retina-related tools. PRANAS integrates a retina simulator allowing large scale simulations while keeping a strong biological plausibility and a toolbox for the analysis of spike train population statistics. The statistical method (entropy maximization under constraints) takes into account both spatial and temporal correlations as constraints, allowing to analyze the effects of memory on statistics. PRANAS also integrates a tool computing and representing in 3D (time-space) receptive fields. All these tools are accessible through a friendly graphical user interface. The most CPU-costly of them have been implemented to run in parallel.

Biosystems & Biorobotics, 2015

Studying and modeling the brain as a whole is a real challenge. For such systemic models (in cont... more Studying and modeling the brain as a whole is a real challenge. For such systemic models (in contrast to models of one brain area or aspect), there is a real need for new tools designed to perform complex numerical experiments, beyond usual tools distributed in the computer science and neuroscience communities. Here, we describe an effective solution, freely available on line and already in use, to validate such models of the brain functions. We explain why this is the best choice, as a complement to robotic setup, and what are the general requirements for such a benchmarking platform. In this experimental setup, the brainy-bot implementing the model to study is embedded in a simplified but realistic controlled environment. From visual, tactile and olfactory input, to body, arm and eye motor command, in addition to vital interoceptive cues, complex survival behaviors can be experimented. We also discuss here algorithmic high-level cognitive modules, making the job of building biologically plausible bots easier. The key point is to possibly alternate the use of symbolic representation and of complementary and usual neural coding. As a consequence, algorithmic principles have to be considered at higher abstract level, beyond a given data representation, which is an interesting challenge.

ABSTRACT In the early visual system, regarding the detection of a visual event, local motion info... more ABSTRACT In the early visual system, regarding the detection of a visual event, local motion information is pre-processed in the Magnocellular pathway, while it has been shown that the Koniocellular pathway also plays an important role, providing a global a-priori estimation about such a kind of information processing. However, the functional interplay between these two parallel pathways remains partially understood. To fill this gap, we investigate, through simulations, the impact of the multi-scale characteristics of the Konio pathway in the Magno-driven thalamocortical system of the mammal. We propose a systemic retino-thalamo-cortico-collicular model, including feedforward, feedback and inhibitory connections. In order to provide a discriminative framework, the model implements minimal algorithms, such as spatio-temporal filters and dynamic neural fields, and is fed with a sequence of images. Our results show that the multi-scale interactions between both pathways, integrating local and larger image cues, account for target discrimination, selection and tracking in the presence of threats or targets. This approach proposes an innovative answer to the interplay issues between these pathways, and it is generalizable to other visuo-motor functions.

We propose a computational approach using a variational specication of the visual front-end, wher... more We propose a computational approach using a variational specication of the visual front-end, where ganglion cells with properties of retinal Konio cells (K-cells), are considered as a network, yielding a mesoscopic view of the retinal process. The variational framework is implemented as a simple mechanism of diusion in a two-layered non-linear ltering mechanism with feedback, as observed in synaptic layers of the retina, while its biological plausibility, and capture functionalities as (i) stimulus adapted response; (ii) non-local noise reduction (i.e. segmentation); (iii) visual event detection, taking several visual cues into account: contrast and local texture, color or edge channels, and motion base in natural images. Those functionalities could be implemented in the biological tissues We use computer vision methods to propose an eective link between the observed functions and their possible implementation in the retinal network base on a two-layers network with nonseparable local spatio-temporal convolution as input, and recurrent connections performing nonlinear diusion before prototype based visual event detection. The numerical robustness of the proposed model has been experimentally checked on real natural images. Finally, we discuss in base of experimental biological and computational results the generality of our description.

Journal of Physiology-Paris, 2011

This article introduces general concepts and definitions related to the notion of asynchronous co... more This article introduces general concepts and definitions related to the notion of asynchronous computation in the framework of artificial neural networks. Using the dynamic field theory as an illustrative example, we explain why one may want to perform such asynchronous computation and how one can implement it since this computational scheme draws several consequences on both the trajectories and the stability of the whole system. After giving an unequivocal definition of asynchronous computation, we present a few practically usable methods and quantitative bounds that can guarantee most of the mesoscopic properties of the system.

Research Report RR- …, 2011

The problem of adjusting the parameters of an event-based network model is addressed here at the ... more The problem of adjusting the parameters of an event-based network model is addressed here at the programmatic level. Considering temporal processing, the goal is to adjust the network units weights so that the outcoming events correspond to what is desired. The present work proposes, in the deterministic and discrete case, a way to adapt usual alignment metrics in order to derive suitable adjustment rules. At the numerical level, the stability and unbiasness of the method is verified.