Complex Systems Seminar Masatoshi FUNABASHI : Synthetic Modeling of Intrinsic Learning with Chaotic Neural Network(CNN)、Akira MASUMI : TBA

2007-11-28 13:00 - 2007-11-28 16:00
N508, RIES, Hokkaido University
Masatoshi FUNABASHI(Centre de Recherche en Episte'mologie Applique'e Ecole Polytechnique)、Akira MASUMI(JAIST)
Masatoshi FUNABASHI: Synthetic Modeling of Intrinsic Learning with Chaotic Neural Network(CNN) As a synthetic model of brain dynamics, considering the dynamics ofneural plasticity, it is important to analyze the situation where thedynamical system's parameters are dependent on its intrinsic dynamics.For this purpose, I proposed a synthetic model conbining thebiological-based CNN and Hebbian/STDP rules. As a result, we could find certain parameter spaces that recombinedifferent memory attractors to produce the novel one. The chaoticwandering orbit during the intrinsic learning therefore has apossibility to have higher order corelation than in attractor state.This can be considered as a constructive function of chaos in brain'slearning dynamics. Applying informtaion geometry, I have developped acomputer program to detect these higher order corelations. The emerged attractor can be characterised with its invariantsubspace, and its dimension was the union space of memorizedattractors.This phenomenon is also interesting as an analogy of dialectic, thatis impossible to realize in logical computer.*****Akira MASUMI : TBA