A Chaos Associative Memory with a Skew-Tent Activation Function
Masahiro Marshall Nakagawa · Journal of the Physical Society of Japan · 2000
We find that a chaos neural network (CNN) model with a time-dependent skew-tent activation function shows ∼10–20 times larger memory capacity than CNN with the sigmoidal activation function. Our model exhibits complete association with a certain noise involved in an input key vector beyond the time-invariant tent model.