Distance-based Kernels for Surrogate Model-based Neuroevolution

TL;DR

This paper explores the use of distance-based kernels in surrogate models to improve the efficiency of neuroevolution for neural network topology optimization, especially when evaluations are costly.

Contribution

It introduces various distance measures for surrogate models and compares their effectiveness in a numerical test scenario.

Findings

Distance measures impact surrogate model performance

Certain kernels outperform others in the test scenario

Distance-based kernels can reduce optimization cost

Abstract

The topology optimization of artificial neural networks can be particularly difficult if the fitness evaluations require expensive experiments or simulations. For that reason, the optimization methods may need to be supported by surrogate models. We propose different distances for a suitable surrogate model, and compare them in a simple numerical test scenario.