Combinatorial Optimizer Based on Deep Reinforcement Learning

2022 Fiscal Year Final Research Report

• Project/Area Number: 20K11988
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 61040:Soft computing-related
• Research Institution: Kyoto Institute of Technology
• Principal Investigator: IIMA Hitoshi 京都工芸繊維大学, 情報工学・人間科学系, 准教授 (70273547)
• Project Period (FY): 2020-04-01 – 2023-03-31
• Keywords: 機械学習 / 深層学習 / 強化学習 / 深層強化学習 / 最適化 / 組合せ最適化

Outline of Final Research Achievements

We have developed optimizers based on deep reinforcement learning for solving combinatorial optimization problems, which are known to be challenging to find optimal solutions in a short time. These optimizers are classified into two types: Go and video games. We have also studied the basic framework of a method that can find a solution for a new instance in a short time after pre-training for another instance. These methods were applied to a delivery scheduling problem, and their performance was investigated experimentally.

Free Research Field

ソフトコンピューティング

Academic Significance and Societal Importance of the Research Achievements

深層強化学習が学習問題に対して高い性能を示すことが知られているが、これを組合せ最適化法として構成し直すことにより、従来の方法とは一線を画す解法を開発できた。この解法を既存の最適化法と比較することにより、さらに優れた最適化法を開発することが期待できる。また、優れた最適化法を開発することで、生産、物流、通信ネットワーク、金融、交通、土木、農業などの多くの分野のシステム開発に寄与できる。