半直交連想記憶におけるニューロンポテンシャルの統計特性による偽安定点の検出

概要

論文の詳細を見る
Semi-orthogonal associative memory (SAM) consists of a neuron field and a characteristic field with N neurons and n characteristic notes, respectively. Like in other associstive memory (AM), however, there exist spurious stable states (SS), which are stable states but not the learned patterns. A stable state is called non-spurious stable state (N-SS) if it is a learned pattern. When SAM is applied to information processing systems, such as, pattern recognition systems, it is often expected that its recalling outputs converge to N-SS. Thus, it is significant giving a real-time condition to discriminate whether a recalling output is SS or N-SS in the recalling processes. In this paper, we shall try to propose this condition by using the statistical properties of the neuron potential (NP) and investigate its detecting efficiency.
社団法人電子情報通信学会の論文
1997-03-06