Hyperdimensional Computing with Spiking-Phasor Neurons

Abstract

A Vector Symbolic Architecture (VSA) is a powerful framework for representing compositional reasoning. It can be used to create neural networks that perform cognitive functions, like spatial reasoning, arithmetic, symbol binding, and logic. But the vectors involved can be quite large, hence the alternative label Hyperdimensional (HD) computing. In this paper, we describe a number of simple neuron models that implement all the operators of a Fourier Holographic Reduced Representation (FHRR, a type of VSA). These spiking phasor neurons represent the phase of a complex number using spike timing within a cycle. We demonstrate the capabilities of these models on a series of VSA queries, including a state transition model, and spatial memory. This spiking version of FHRR paves the way for efficiently running a variety of VSA networks on neuromorphic hardware.