Skip to Main content Skip to Navigation
Theses

Une modélisation évolutionniste du liage temporel

David Meunier 1
1 DM2L - Data Mining and Machine Learning
LIRIS - Laboratoire d'InfoRmatique en Image et Systèmes d'information
Abstract : The hypothesis of temporal binding by synchrony raises a major challenge in neurobiology, because it explains how different cerebral areas can establish a functionnal relation between each others, linked to a cognitive process. However, there is no model allowing different groups of neurons to communicate by the mean of their emissions.

We have developped a model of spiking neuron network, where the topology is modified by an evolutionary algorithm. The fitness criterion used for the evolutionary algorithm is evaluated by mean of the behaviour of an individual, controlled by a spiking neuron network, and placed in a virtual environment. The use of a spiking neuron model, with the property of synchrony detection, obliges evolution to build a system using this property at a global level, and results in the emergence of large-scale neural synchronisation. Topological and dynamical properties are not taken into account in the fitness computation, but are studied a posteriori by comparing individuals before and after evolution.

On the one hand, with the help of complex network tools, we show the emergence of several topological properties, including ``small-world" network properties. These topological properties are similar to those obtained from the anatomy of biological nervous systems. On the other hand, at the dynamical level, we show that the property of large-scale neural synchronisation, induced by the presentation of a stimulus, is found for evolved individuals. To show this result, we use some tools classically used in electrophysiology, and extend these tools in order to analyze the big quantity of data obtained from the model.

The model shows that it is possible to build artificial neural networks based on the hypothesis of temporal binding, by using an ecological fitness criterion (the behaviour of an individual in a virtual environment) for artificial evolution. The tools extended for analyzing the model properties can be used in other domains, including electrophysiology. The quantity of data in this domain and those obtained with the model are indeed similar, because of recent technical innovations in electrophysiological recordings.
Document type :
Theses
Complete list of metadatas

https://tel.archives-ouvertes.fr/tel-00198797
Contributor : David Meunier <>
Submitted on : Monday, December 17, 2007 - 9:45:33 PM
Last modification on : Wednesday, July 8, 2020 - 12:43:49 PM
Long-term archiving on: : Monday, April 12, 2010 - 8:15:13 AM

Identifiers

  • HAL Id : tel-00198797, version 1

Citation

David Meunier. Une modélisation évolutionniste du liage temporel. Informatique [cs]. Université Lumière - Lyon II, 2007. Français. ⟨tel-00198797⟩

Share

Metrics

Record views

315

Files downloads

2079