2A11 強誘電性液晶パネルを用いた光・電子ハイブリッド適応的ハミングネットプロセッサ

概要

論文の詳細を見る
Adaptive resonance theory 1(ART1) network originated by Carpenter and Grossberg for binary pattern classification through unsupervised learning is famous for overcoming stability-plasticity dilemma in adaptive self-organizing learning. But ART1 requires iterative processing or resonance in learning and classification because matching score μ and vigilance parameter ρ in ART1 are determined by matching number of elements of state 1 in binary patterns. In this paper we propose "adaptive hamming network" which does not require iterative processig in learning and classification, by introducing matching score in hamming network, which takes matching of both "1" and "0" in binary patterns into consideration, into ART1. Therefore, adaptive hamming network has advantages of very fast learning and classification. Moreover, adaptive hamming network has an excellent charasteristic of controlling stability-plasticity balance in category learning and recognition of superset templates by vigilance hamming threshold. We propose not only adaptive hamming network but also hybrid optoelectronic adaptive hamming net processor using two stacked ferroelectric liquid crystal(FLC) panels. In this novel processor large capacity and high density of optical adaptive haming layer can be achieved by high resolution and big area of FLC panels.
日本液晶学会の論文
1995-09-10