Probabilistic Boolean Operation in Networks of Spiking Neurons for Bayesian Inference of Hidden Markov Models

Abstract

<p>Spiking neural networks (SNN) have gained popularity among the researchers for low power consumption. An interesting problem on Bayesian inference of hidden Markov models (HMM) on SNN paradigm is timely. In this work we propose a novel approach to address this issue. A probabilistic temporal encoding scheme is adopted and it has been shown that a spiking neuron behaves as a probabilistic Boolean operator. Using this property the posterior of a hidden state is mapped to probability of firing a logic $1$ by a spiking neuron. It has been demonstrated that both binary state HMM as well as multi-state HMMs are implementable using the present approach. The proposed method has been demonstrated to be effective for reasonably accurate estimation of hidden states. The present method seems to be reliable, power efficient and easy to follow.</p>