転移学習付き遺伝的ネットワークプログラミングとその応用に関する研究

点击翻译按钮获取中文摘要

The primary goal of the project is to integrate transfer learning (TL) into an evolutionary computation (EC) algorithm -- genetic network programming (GNP) -- to develop advanced optimization techniques, which is capable of significantly improving the evolution efficiency (problem solving) of traditional EC algorithms. With respect to the above target, the following topics have been realized:1. Developed advanced GNP, especially GNP-based data mining model, to realize the fundamental framework of transfer learning in EC.2. Developed reinforcement learning (RL)-based scheme into GNP to develop a new algorithm GNP with knowledge transfer (GNP-KT), which can transfer knowledge across different robot situations.3. Integrated the RL-based scheme into the learning classifier systems (LCSs), which realize the rule-based knowledge transfer.4. Basically, the transferable EC pointed out a new research direction of EC, which will be further promoted in the future.