A Spiking Neural Network Model for Associative Memory Using Temporal Codes

Abstract

Associative memory is defined as the ability to map input patterns to output patterns. Understanding how human brain performs association between unrelated patterns and stores this knowledge is one of the most important goals in computational intelligence. Although this problem has been widely studied using conventional neural networks, increasing biological findings suggest that spiking neural network can be an alternative. The proposed model encodes different memories using different subsets of encoding neurons with temporal codes. A spike-timing based learning algorithm and spike-timing-dependent plasticity (STDP) are used to form associative memory. Simulation results show that hetero-associative memory and auto-associative memory are achievable by the synaptic modification of connections between input layer and hidden layers, and recurrent connections of hidden layers, respectively.