Skip to Main content Skip to Navigation

Réseaux causaux probabilistes à grande échelle : un nouveau formalisme pour la modélisation du traitement de l'information cérébrale

Abstract : The understanding and the prediction of the clinical outcomes of focal or degenerative cerebral lesions, as well as the assessment of rehabilitation procedures, necessitate knowing the cerebral substratum of cognitive or sensorimotor functions. This is achieved by activation studies, where subjects are asked to perform a specific task while data of their brain functioning are obtained through functional neuroimaging techniques. Such studies, as well as animal experiments, have shown that sensorimotor or cognitive functions are the offspring of the activity of large-scale networks of anatomically connected cerebral regions. However, no one-to-one correspondence between activated networks and functions can be found. Furthermore, apparently conflicting activation data can only be explained by understanding how the activation of large-scale networks derives from cerebral information processing mechanisms. At this level, cerebral mechanisms are the synthesis of more basic neurobiological, neurophysiological or neuropsychological processes. They can only be approached with the help of explicit computational models, based on the knowledge of more basic processes. In this work, we aim at defining a new and flexible formalism allowing building such models, for a better interpretation of cerebral functional images. It is based on dynamic Bayesian networks, a causal, probabilistic and graphical modelling approach. The brain is viewed as a network of anatomically connected functional areas. Each area is a processing unit which is represented by a set of nodes in the dynamic Bayesian network. In the formalism, the information processed by one node is the abstraction of the cerebral activity in the corresponding neuronal population. We use a two-sided information representation dedicated to the representation of integrated cerebral information. It is made of a numerical value corresponding to the activation level, and a symbolic value representing the pattern of activated neurons. This work is a part of a project which is dedicated to the definition of an original causal approach for the modelling of cerebral information processing in large-scale networks and the interpretation of neuroimaging data. Causality allows us to express explicitly our hypotheses on cerebral mechanisms. Our contribution can be described in two points. In an artificial intelligence perspective, the use of labelled random variables in dynamic Bayesian networks allows us to define original non-supervised learning mechanisms. From the computational neuroscience viewpoint, we propose a new causal formalism, which is more adapted than formal neural networks to model cerebral processing at the global level of cerebral areas networks. Furthermore, our formalism is more biologically plausible than other causal approach like causal qualitative networks.
Complete list of metadatas
Contributor : Vincent Labatut <>
Submitted on : Tuesday, March 2, 2004 - 5:08:13 PM
Last modification on : Saturday, November 28, 2020 - 3:05:30 AM
Long-term archiving on: : Friday, April 2, 2010 - 8:59:49 PM


  • HAL Id : tel-00005190, version 1



Vincent Labatut. Réseaux causaux probabilistes à grande échelle : un nouveau formalisme pour la modélisation du traitement de l'information cérébrale. Intelligence artificielle [cs.AI]. Université Paul Sabatier - Toulouse III, 2003. Français. ⟨tel-00005190⟩



Record views


Files downloads