Stochastic Dynamics and Learning Rules in Layered Neural Networks
Hidetsugu Sakaguchi · Progress of Theoretical Physics · 1990
We propose two kinds of layered neural network models using stochastic dynamics. One is a feedforward type of the Boltzmann machine, i.e., the network has probabilistic input-output relation. The other model utilizes a stochastic dynamics for a learning rule to avoid the difficulty being trapped in the local minima of an error function.