Enhancing neural-network performance via assortativity

TL;DR

This paper investigates how assortativity, or degree-degree correlations, in neural network topologies affects their robustness to noise, revealing that positive correlations enhance performance especially when hub neurons encode information.

Contribution

It introduces an analytical and computational framework to study the impact of assortativity on neural network dynamics, extending previous topology-performance analyses.

Findings

Assortative networks show increased robustness to noise.

Hub neurons in assortative networks are crucial for information storage.

Positive degree correlations improve neural network stability.

Abstract

The performance of attractor neural networks has been shown to depend crucially on the heterogeneity of the underlying topology. We take this analysis a step further by examining the effect of degree-degree correlations -- or assortativity -- on neural-network behavior. We make use of a method recently put forward for studying correlated networks and dynamics thereon, both analytically and computationally, which is independent of how the topology may have evolved. We show how the robustness to noise is greatly enhanced in assortative (positively correlated) neural networks, especially if it is the hub neurons that store the information.