Émergence et apprentissage d'information dans un modèle multimodal de cartes CNFT impulsionnelles

Selma Belgacem (1) , Yann Boniface (2) , Alain Dutech (3)

1 DocApp - LITIS - Equipe Apprentissage
2 CORTEX - Neuromimetic intelligence
3 MAIA - Autonomous intelligent machine

Given the diversity of data, it is essential to provide artificial systems with a capacity of association of information. The aim of the work we present in this paper is to provide a new neuromimetic system with a capacity of association of correlated information coming from different external inputs. This system has a multimodal architecture, consisting of CNTF cards constituted of spiking neurons with inter-cards and intra-cards connections. We present a study of the functional properties of this system and the conditions of convergence to a stable state preparing the system to emerge this association and to learn it. Then, we present a model of the STDP learning rule that we adapted to the dynamics of our system. To evaluate the learning ability of our system, we developed a scenario of memorization and reminder that we explain in this article.

Mots clés: système neuromimétique; multimodalité; cartes CNFT; neurone impulsionnel; association; apprentissage; loi STDP; rappel

Domaines: Neurosciences

Selma Belgacem, Yann Boniface, Alain Dutech. Émergence et apprentissage d'information dans un modèle multimodal de cartes CNFT impulsionnelles. 2009. ⟨hal-00802908⟩