Bavafa-Toosi, Yazdan (Keio Univ.), Ohmori, Hiromitsu (Keio Univ.)

Flexible Neural Network-Based Associative Memories – Part I: Behavior

Scheduled for presentation during the Regular Session "Neural Networks I" (TuP01), Tuesday, July 25, 2006, 16:35−17:00, Room B2

17th International Symposium on Mathematical Theory of Networks and Systems, July 24-28, 2006, Kyoto, Japan

This information is tentative and subject to change. Compiled on April 22, 2024

Abstract

Abstract——The successful applications of flexible neural networks (FNNs) notwithstanding, nothing was rigorously known about them until recently that we examined them in a theoretical framework. Following those works, two models of FNN-based associative memories (AMs) are introduced and studied from the behavioral standpoint, i.e., model following or continuous trajectory learning. A nonlinearly parameterized (NLP) adaptive control is proposed as the training scheme. The solution contains a time-varying designer-specified vector parameter, and thus some degrees of freedom, which determines the control input (or the teaching signal) and the rate of adaptation (or learning). Its global stability and convergence are proven.