Cognitive Learning and Memory Systems Using Spiking Neural Networks

Abstract

Neural networks have been studied for many years to mimic biological neural systems. However, humans can thoroughly defeat artificial intelligence with little difficulty when facing with cognitive tasks. The goal of this thesis is to investigate aspects of theories of spiking neural networks in an attempt to develop cognitive learning and memory models. This thesis starts off with a spike-timing-based integrated model that uses temporal codes and spike-timing based learning for pattern recognition. Inspired by the hippocampal CA3, a computationally efficient auto-associative memory model is proposed, in which memory items are encoded by different subsets of neurons with a recurrent network structure. To investigate the organizing mechanisms of memory, a hierarchically organized memory model using temporal population codes is proposed. Both associative memory as well as episodic memory can be stored in the model. Moreover, memories coding for features from simple to complex are organized in a hierarchical manner.