Adaptive resonance theory design in mixed memristive-fuzzy hardware

Abstract

Fuzzification of neural networks show great promise in improving system reliability and computational efficiency. In the present work we explore the possibility of combining fuzzy inference with Adaptive Resonance Theory (ART) neural networks implemented on massively parallel hardware architectures including memristive devices. Memristive hardware holds promise to greatly reduce power requirements of such neuromorphic applications by increasing synaptic memory storage capacity and decreasing wiring length between memory storage and computational modules. Storing and updating synaptic weight values based on synaptic plasticity rules is one of the most computationally demanding operations in biologically-inspired neural networks such as Adaptive Resonance Theory (ART). Our work indicates that Fuzzy Inference Systems (FIS) can significantly improve computational efficiency. In this chapter, we introduce a novel method, based on fuzzy inference, to reduce the computational burden of a class of recurrent networks named recurrent competitive fields (RCFs).A novel algorithmic scheme is presented to more efficiently perform the synaptic learning component of ART networks in memristive hardware. RCF networks using FIS are able to learn synaptic weightswith small absolute error rates, and classify correctly. Using the FIS methodology it is possible to significantly reduce the computational complexity of the proposed memristive hardware using computationally cheaper and more robust fuzzy operators.