• Search Research Projects
  • Search Researchers
  • How to Use
  1. Back to project page

2022 Fiscal Year Final Research Report

Segmentation of Time and Space in a Fully Online Reinforcement Learning System

Research Project

  • PDF
Project/Area Number 18K11473
Research Category

Grant-in-Aid for Scientific Research (C)

Allocation TypeMulti-year Fund
Section一般
Review Section Basic Section 61040:Soft computing-related
Research InstitutionOsaka Metropolitan University (2022)
Osaka Prefecture University (2018-2021)

Principal Investigator

Notsu Akira  大阪公立大学, 大学院現代システム科学研究科, 教授 (40405345)

Co-Investigator(Kenkyū-buntansha) 生方 誠希  大阪公立大学, 大学院情報学研究科, 准教授 (10755698)
本多 克宏  大阪公立大学, 大学院情報学研究科, 教授 (80332964)
Project Period (FY) 2018-04-01 – 2023-03-31
Keywords強化学習 / クラスタリング / 最適化アルゴリズム / 転移学習 / 学習と進化
Outline of Final Research Achievements

We modified the growing self-organizing map for reinforcement learning and devised a method for unsupervised learning of state space and state transitions while maintaining learning efficiency, and demonstrated the usefulness of this method. We also showed that the method can adapt to the environment by adaptively changing the hyperparameter settings significantly. Furthermore, we proposed a method for switching methods while estimating several local environments for differential evolution, which is one of the best optimization algorithm methods, and were able to improve the performance. In addition, we were able to apply our findings to deep reinforcement learning, which had not been considered much at first, and propose a completely new deep reinforcement learning system.

Free Research Field

強化学習

Academic Significance and Societal Importance of the Research Achievements

本研究は強化学習が必要とする空間を統計学的に大量のデータを用いて獲得するのでは無く,幾何学的なミクロな観点から獲得したという意味で学術的な意義があると考えている.また,機械学習にとってハイパーパラメータの設定は大きな問題であるが,その適応的変化や並列学習で対応できることを明らかにしたことは,学術的にも産業応用を考えた上でも意義がある.さらに,ブラックボックス最適化アルゴリズムを発展させることは複雑化する社会問題など,ありとあらゆる最適化に貢献できることを意味しているので,社会的にも大きな意義がある.

URL: 

Published: 2024-01-30  

Information User Guide FAQ News Terms of Use Attribution of KAKENHI

Powered by NII kakenhi