Efficient Memristive Circuit Design of Neural Network-Based Associative Memory for Pavlovian Conditional Reflex

Abstract

The brain’s learning and adaptation processes heavily rely on the concept of associative memory. One of the most basic associative learning processes is classical conditioning. This work presents a memristive neural network-based associative memory system. The system can emulate Pavlovian conditioning principles including acquisition, extension, generalization, differentiation, and spontaneous recovery that have not been considered in most of the previous counterparts. The proposed circuit can emulate these principles thanks to the resistance-changing characteristics of the memristor. Generalization has been achieved by providing both unconditional and neutral stimuli to the network to reduce the memristance of the memristor. Differentiation has been attained by employing unconditional and conditional stimuli in a training scheme to obtain a certain memristance that causes the network to respond differently to both stimuli. A revival of an exterminated stimuli is also done by increasing the synaptic weight of the system. Compared to previous designs, the proposed memristive circuit can implement all the functions of conditional reflex. Our rigorous simulations demonstrated that the proposed memristive system can condition neutral stimuli, show generalization between similar stimuli, distinguish dissimilarities between the generalized stimuli, and recover faded stimuli.