Proposal and Evaluation of Surrogate-Assisted Self-Adaptive MBEANN

Abstract

This paper proposes an extension for Mutation-Based Evolving Artificial Neural Networks (MBEANN) algorithm. The proposed method consists of two parts: a surrogate model and a self-adaptive mutation. Firstly, the surrogate-assisted mechanism is introduced to MBEANN for reducing the cost of fitness evaluations. This mechanism employs approximated fitness values predicted by a surrogate model instead of true fitness functions. Secondly, the self-adaptive mutation is applied to MBEANN for adjusting the exploring area in parameter space. The performance of the proposed method is compared with the normal MBEANN and NEAT algorithms by using the three benchmarks of OpenAI Gym. The experimental results showed that the proposed method outperformed other algorithms in all benchmarks.