A chaos memory model
Masahiro Marshall Nakagawa · 2003
In this paper we explore that a chaos neural network (CNN) model with a sinusoidal activation function may realise /spl sim/10-20 times larger memory capacity than CNN with conventionally proposed sigmoidal activation function proposed by K. Aihara et al (1990). In fact the complete association is found to be promoted up to /spl sim/100% loading rate even though a relevant noise is involved in an input vector.