[4L3-J-8] ソフトコンピューティング

[4L3-J-8-01] Connection-Aware Spectrum-Diversity for Neuroevolution

〇Danilo Vasconcellos Vargas¹, Yuta Inoue¹ (1. Kyushu University)

キーワード：Neuroevolution, Spectrum-diversity Unified Neuroevolution Architecture, General Artificial Intelligence, Reinforcement Learning

Spectrum-diverse Unified Neuroevolution Architecture (SUNA) is currently the most adaptive neuroevolution methods which is able to tackle different problems efficiently. This is possible by making use of (a) a unified neural model which allows a greater representation power together with (b) a new diversity metric called spectrum diversity which enables it to search in the huge search space created by the unified neural model. However, many questions remain unanswered regarding the feasibility of layers and other improved structures as well as improved diversity measures. Here we provide a study over a variation of the diversity measure. In other words, we create a connection-aware spectrum diversity. Experiments show that a connection-aware spectrum diversity allows for better results to arise over the course of evolution. This is justified by the fact that neural networks with a low number of connections are kept even when increasing the connections might improve slightly the results. Moreover, these networks themselves are easier to improve than ones with a high number of connections.