Optimising the topology of complex neural networks

TL;DR

This paper investigates how the topology of complex neural networks, specifically Self-Organizing Maps, affects their classification performance and robustness, finding that topology influences performance minimally but can be optimized through evolution.

Contribution

It demonstrates that evolving the topology of Self-Organizing Maps can improve classification accuracy by nearly 10%, revealing the impact of network structure on performance.

Findings

Topology has minimal impact on performance and robustness at long learning times.

Artificial evolution can enhance performance by approximately 10%.

Evolved networks are more random with heterogeneous degree distribution.

Abstract

In this paper, we study instances of complex neural networks, i.e. neural netwo rks with complex topologies. We use Self-Organizing Map neural networks whose n eighbourhood relationships are defined by a complex network, to classify handwr itten digits. We show that topology has a small impact on performance and robus tness to neuron failures, at least at long learning times. Performance may howe ver be increased (by almost 10%) by artificial evolution of the network topo logy. In our experimental conditions, the evolved networks are more random than their parents, but display a more heterogeneous degree distribution.