Learning and stabilization of winner-take-all dynamics through interacting excitatory and inhibitory plasticity

Abstract

Winner-Take-All (WTA) networks are recurrently connected populations of excitatory and inhibitory neurons that represent promising candidate microcircuits for implementing cortical computation. WTAs can perform powerful computations, ranging from signal-restoration to state-dependent processing. However, such networks require fine-tuned connectivity parameters to keep the network dynamics within stable operating regimes. In this article, we show how such stability can emerge autonomously through an interaction of biologically plausible plasticity mechanisms that operate simultaneously on all excitatory and inhibitory synapses of the network. A weight-dependent plasticity rule is derived from the triplet spike-timing dependent plasticity model, and its stabilization properties in the mean-field case are analyzed using contraction theory. Our main result provides simple constraints on the plasticity rule parameters, rather than on the weights themselves, which guarantee stable WTA behavior. The plastic network we present is able to adapt to changing input conditions, and to dynamically adjust its gain, therefore exhibiting self-stabilization mechanisms that are crucial for maintaining stable operation in large networks of interconnected subunits. We show how distributed neural assemblies can adjust their parameters for stable WTA function autonomously while respecting anatomical constraints on neural wiring.

Thalamic Inhibition: Diverse Sources, Diverse Scales

“…Closed loop organization may allow more restricted spatial scale of inhibition whereas open loop connections (depending on their degree of divergence) may involve larger part of the thalamus. In addition open loop organization would allow exerting lateral inhibition within thalamic circuits and ascending sensory signals therefore, likely to show a winner-take-all feature [88]. It is intriguing to speculate that activating subsets of TRN neurons by top-down control can result in biasing thalamic processing towards particular sensory features based on this open-loop organization.…”

Local structure helps learning optimized automata in recurrent neural networks

“…Attractor networks have been investigated extensively and are hypothesized to implement important dynamical elements in biological brains [16]. Furthermore, recent results show that stable WTA dynamics, which form the basis of the types of attractor networks used in this work, can emerge in a self organized way through an interaction of biologically realistic learning rules [14]. The simple structure of the networks allows us to derive an update rule for the connections between neural populations, the effects of which can be intuitively understood.…”

“…the connections from the gating populations to the state populations. All other connections remain fixed, although they could in principle also be learned with biologi cally plausible rules [14]. Due to the competition mechanism, only one of the state populations will be active after an input symbol is provided and a gating population has been activated, namely the one receiving the greatest input from that gating population.…”

A hippocampal model for behavioral time acquisition and fast bidirectional replay of spatio-temporal memory sequences

“…The winner-take-all dynamics implemented using the competitive inhibition are commonplace in the theoretical neuroscience literature [64,65]. Furthermore, anatomical studies have shown that cortical networks contain essential features which can be implemented via winnertake-all circuits [66,67].…”