A Super Memory Retrieval with Chaos Associative Model
Masahiro Marshall Nakagawa · Journal of the Physical Society of Japan · 1999
We find that a chaos neural network ( CNN ) model with a sinusoidal activation function shows ∼10–20 times larger memory capacity than CNN with the sigmoidal activation function previously proposed by Aihara et al. Our network exhibits complete association even though a considerable noise is involved in an input vector.