Competing synapses with two timescales: a basis for learning and forgetting

TL;DR

This paper presents a model of synaptic competition with two timescales that explains learning and forgetting, predicting optimal memory retention when synapses are strongly or weakly differentiated.

Contribution

It introduces a game theory-inspired model of synaptic interactions that links synaptic competition to learning and memory dynamics.

Findings

Memory is optimized when weak synapses are very weak and strong synapses are very strong.

The model predicts a natural timescale for forgetting that is inherent in synaptic competition.

The framework can be extended and tested experimentally, potentially complementing existing models of reaching.

Abstract

Competitive dynamics are thought to occur in many processes of learning involving synaptic plasticity. Here we show, in a game theory-inspired model of synaptic interactions, that the competition between synapses in their weak and strong states gives rise to a natural framework of learning, with the prediction of memory inherent in a timescale for `forgetting' a learned signal. Among our main results is the prediction that memory is optimized if the weak synapses are really weak, and the strong synapses are really strong. Our work admits of many extensions and possible experiments to test its validity, and in particular might complement an existing model of reaching, which has strong experimental support.