Storage capacity of phase-coded patterns in sparse neural networks

TL;DR

This paper investigates how sparse recurrent neural networks can store multiple phase-coded patterns as stable attractors, using an STDP-inspired learning rule, and finds that small-world topologies optimize storage capacity.

Contribution

It introduces a novel learning rule for storing phase-coded patterns in sparse networks and analyzes how network topology affects storage capacity.

Findings

Networks can reliably replay stored phase-coded patterns after learning.

Small-world topology optimizes the trade-off between wiring cost and capacity.

Capacity depends on network topology and connectivity sparsity.

Abstract

We study the storage of multiple phase-coded patterns as stable dynamical attractors in recurrent neural networks with sparse connectivity. To determine the synaptic strength of existent connections and store the phase-coded patterns, we introduce a learning rule inspired to the spike-timing dependent plasticity (STDP). We find that, after learning, the spontaneous dynamics of the network replay one of the stored dynamical patterns, depending on the network initialization. We study the network capacity as a function of topology, and find that a small- world-like topology may be optimal, as a compromise between the high wiring cost of long range connections and the capacity increase.