2023 Fiscal Year Final Research Report
Research on the innovative evolution of deep reinforcement learning based on the profit sharing principle and its application to real problems
| Project/Area Number |
21K12024
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Multi-year Fund |
|---|
| Section | 一般 |
|---|
| Review Section |
Basic Section 61030:Intelligent informatics-related
|
|---|
| Research Institution | National Institution for Academic Degrees and Quality Enhancement of Higher Education |
|---|
Principal Investigator |
Miyazaki Kazuteru 独立行政法人大学改革支援・学位授与機構, 研究開発部, 教授 (20282866)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
山口 周 独立行政法人大学改革支援・学位授与機構, 研究開発部, 特任教授 (10182437)
原田 拓 東京理科大学, 創域理工学部経営システム工学科, 准教授 (70256668)
小玉 直樹 明治大学, 理工学部, 助教 (60908747)
|
|---|
| Project Period (FY) |
2021-04-01 – 2024-03-31
|
| Keywords | 深層強化学習 / 強化学習 / 深層学習 / 利益分配原理 / 経験強化型学習 |
|---|
| Outline of Final Research Achievements |
In this study, after completing the basic design of the Deep Profit Sharing method, which is “deep exploitation-oriented learning with reduced variability of learning results based on the profit sharing principle,” which was the original goal of this study, we expanded the application examples to real problems considering two sub-goals related to target problem classes. Specifically, we achieved the originally planned “application to smart energy systems” and also obtained certain results for “application to curriculum analysis support systems." In addition, as an example of an application not initially envisioned, after achieving a certain level of success with the application to road traffic signal control, we began applying the system to the suppression of negative tweets and the Angry Bird AI Competition.
|
| Free Research Field |
機械学習、人工知能
|
| Academic Significance and Societal Importance of the Research Achievements |
本研究では「利益分配原理に基づく学習結果のばらつきを抑えた深層経験強化型学習」であるDeep Profit Sharing method（DeePS）の有効性を主張できた。これは、動的計画法や政策の直接探索に基づく手法が主流を占める深層強化学習の世界に一石を投じるものであり、学術的意義が大きい。通常、それらの手法では、学習に多くの試行錯誤を要するが、DeePSは、より少ない経験でいかに学習するかを主眼に置いており、実問題への応用において、特に、威力を発揮するものと考える。実際、本研究では、複数の実問題に応用し、DeePSの有効性を示すことができたので、得られた成果の社会的意義は大きいと言える。
|
