Dynamics and Thermodynamics of Layered Neural Network
Daoyun Ji, Hu Bei-Lai, Chen Tian-Lun · Communications in Theoretical Physics · 1997
We consider the dynamics of overlaps in a layered network with as its threshold transfer function. The analysis of dynamical stability shows that when the external stimulus is close enough to one of the stored patterns the retrieval state becomes globally stable. The statistical mechanics of the same model, in which the parameter is introduced as thermal noise, is also studied. We find that the thermodynamical stability is equivalent to the dynamical stability.