Temporal sequence learning in winner-take-all networks of spiking neurons demonstrated in a brain-based device

Jeffrey L Mckinstry,Gerald M Edelman

doi:10.3389/fnbot.2013.00010

Abstract

Animal behavior often involves a temporally ordered sequence of actions learned from experience. Here we describe simulations of interconnected networks of spiking neurons that learn to generate patterns of activity in correct temporal order. The simulation consists of large-scale networks of thousands of excitatory and inhibitory neurons that exhibit short-term synaptic plasticity and spike-timing dependent synaptic plasticity. The neural architecture within each area is arranged to evoke winner-take-all (WTA) patterns of neural activity that persist for tens of milliseconds. In order to generate and switch between consecutive firing patterns in correct temporal order, a reentrant exchange of signals between these areas was necessary. To demonstrate the capacity of this arrangement, we used the simulation to train a brain-based device responding to visual input by autonomously generating temporal sequences of motor actions.

Highlights

A growing body of neurophysiological evidence suggests that patterns of activity in vertebrate brains observed during movement are commonly composed of temporal sequences of periods with steady-state firing rates lasting several hundred milliseconds separated by sharp transitions (Tanji, 2001; Averbeck et al, 2002; Nakajima et al, 2009)
Network models composed of mean-firing-rate neurons have been used to model sequential neural activity (Rhodes et al, 2004; Salinas, 2009; Verduzco-Flores et al, 2012), biological networks are composed of spiking neurons
In this paper we describe how our previous models of WinnerTake-All (WTA) spiking networks (Chen et al, 2013) can be coupled together and trained to generate segmented and sequential neural activity

Summary

Introduction

A growing body of neurophysiological evidence suggests that patterns of activity in vertebrate brains observed during movement are commonly composed of temporal sequences of periods with steady-state firing rates lasting several hundred milliseconds separated by sharp transitions (Tanji, 2001; Averbeck et al, 2002; Nakajima et al, 2009). This pattern of activity is observed during sensory perception in gustatory cortex (Jones et al, 2007), and the operation of working memory (Seidemann et al, 1996). Spike-timing dependent plasticity (STDP) modeled long-term synaptic changes that allowed the system to learn temporal sequences

Methods

Results

Conclusion