Latching dynamics instructed through an extra adaptation-related field in Potts neural network

Potts neural network has been widely studied as a neural basis for infinite recursion for the purpose of capturing global cortical dynamics that are related to higher order cognitive functions (speech, thought, confabulation, etc.) in the brain for decades. Those studies are largely limited to spontaneous latching with randomly, weakly, and strongly correlated associative memory patterns. Efforts to guide the spontaneous dynamics have been devoted to imposing explicit instruction on the network. With Potts’ neural network, there was an attempt to modify the Hebbian learning rule by adding an extra heteroassociative learning term that contains instructions to control dynamics. It has been found that providing instruction deteriorates the quality of retrieval of patterns and makes the latching longer. In this paper, instead of adding a heteroassociative part to the Hebbian learning synaptic weight, we study latching dynamics by adding an adaptation related term (σ−θ coupling) to the field that a Potts unit receives, reflecting the realistic and temporally gradual transitions between attractors. We found that the transitions are not faithful to following instructions, and the quality of retrieval of patterns is deteriorated when imposing such instructions.