Benchmarking Many-Objective Topology and Weight Evolution of Neural Networks: A Study with NEWS/D
Adham Salih, Amiram Moshaiov · 2021 IEEE Symposium Series on Computational Intelligence (SSCI) · 2021
This study aims to provide procedures and benchmark problems to test the optimization capabilities of algorithms for Many-Objective Topology and Weight Evolution of Artificial Neural Networks (MaO-TWEANNs). In particular, the proposed benchmarks are based on a combination of continuous functions that have commonly been used to test single-objective optimization algorithms. In addition, this paper applies the proposed procedures and benchmark problems to evaluate NEWS/D, which is such an algorithm that has recently been introduced. The results of this study validate the optimization capabilities of NEWS/D for MaO-TWEANN problems with continuous outputs.