RETRIEVAL PROPERTIES OF ANALOG NEURAL NETWORKS AND THE NONMONOTONICITY OF TRANSFER-FUNCTIONS

Characteristic properties of associative memory networks with continuo us-time dynamics are extensively studied for a certain class of nonmon otonic transfer functions by means of the self-consistent signal-to-no ise analysis (SCSNA) and numerical simulations. The conventional Hebb- type symmetric synaptic connections with unbiased random pasterns are assumed. Although the occurrence of instability, including an oscillat ory one, makes the storage capacity fall below the upper bound for sto rage ratio obtained by the SCSNA, the storage capacity remains as larg e as 0.4 in the optimal cases. It is also noted that noise in the loca l fields (i.e., the inputs to neurons) can vanish for certain cases of nonmonotonic transfer functions even with an extensive number of stor ed patterns. Implication of the present results is the possibility of improving the network performances by the achievement of errorless ret rieval and enhancement of storage capacity with the use of nonmonotoni c transfer functions.