Reliability of neuronal information conveyed by unreliable neuristor-based leaky integrate-and-fire neurons: a model study.

Hyungkwang Lim,Vladimir Kornijcuk,Jun Yeong Seok,Inho Kim,Seong Keun Kim,Doo Seok Jeong,Cheol Seong Hwang

doi:10.1038/srep09776

Hyungkwang Lim, Vladimir Kornijcuk + Show 5 more

Open Access

https://doi.org/10.1038/srep09776

Copy DOI

Abstract

We conducted simulations on the neuronal behavior of neuristor-based leaky integrate-and-fire (NLIF) neurons. The phase-plane analysis on the NLIF neuron highlights its spiking dynamics – determined by two nullclines conditional on the variables on the plane. Particular emphasis was placed on the operational noise arising from the variability of the threshold switching behavior in the neuron on each switching event. As a consequence, we found that the NLIF neuron exhibits a Poisson-like noise in spiking, delimiting the reliability of the information conveyed by individual NLIF neurons. To highlight neuronal information coding at a higher level, a population of noisy NLIF neurons was analyzed in regard to probability of successful information decoding given the Poisson-like noise of each neuron. The result demonstrates highly probable success in decoding in spite of large variability – due to the variability of the threshold switching behavior – of individual neurons.

Highlights

To date, a great deal of efforts have been made to realize the different types of artificial hardware neurons, including the LIF neuron, using conventional complementary metal-oxide-semiconductor (CMOS) technologies[16,17]
A question arising from the analysis on individual neuristor-based LIF (NLIF) neurons is “Can conveying information, such as encoding and decoding, be achieved in a reliable manner by a population of these individual NLIF neurons?” This question is related to neuronal behavior at a higher dimension, i.e., the group, rather than at the individual neuronal level
In a single NLIF neuron circuit, standard circuit elements such as resistors (R1, R2, and RL), capacitors (C1 and C2), and threshold switch (TS) (S1 and S2) are in use, as shown in Fig. 1a. 18 When it comes to a network of NLIF neurons, V2 in Fig. 1a is relayed to a neighboring neuron through a synapse, so that V2 works as the output voltage, corresponding to membrane potential

Summary

Introduction

A great deal of efforts have been made to realize the different types of artificial hardware neurons, including the LIF neuron, using conventional complementary metal-oxide-semiconductor (CMOS) technologies[16,17]. The behavior of the TS is described by four parameters: Ron, Roff, Von, and Voff, which denote the on- and off-state resistances and threshold voltages for off-to-on and on-to-off switching, respectively.

Results

Conclusion