| Project/Area Number |
16K00317
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Multi-year Fund |
|---|
| Section | 一般 |
|---|
| Research Field |
Intelligent informatics
|
|---|
| Research Institution | Nara National College of Technology |
|---|
Principal Investigator |
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
高玉 圭樹 電気通信大学, 大学院情報理工学研究科, 教授 (20345367)
|
|---|
| Project Period (FY) |
2016-04-01 – 2019-03-31
|
| Project Status |
Completed (Fiscal Year 2018)
|
| Budget Amount *help |
¥4,550,000 (Direct Cost: ¥3,500,000、Indirect Cost: ¥1,050,000)
Fiscal Year 2018: ¥1,170,000 (Direct Cost: ¥900,000、Indirect Cost: ¥270,000)
Fiscal Year 2017: ¥1,560,000 (Direct Cost: ¥1,200,000、Indirect Cost: ¥360,000)
Fiscal Year 2016: ¥1,820,000 (Direct Cost: ¥1,400,000、Indirect Cost: ¥420,000)
|
|---|
| Keywords | 機械学習 / 学習過程 / 自律学習 / 逆強化学習 / 継続的学習 / 多目的強化学習 / 目標生成 / 報酬生起確率 / 強化学習 / 継続的学習支援 / 学習エージェント / 振り返り / 学習目標空間 / 学習目標生成 / 上達過程の可視化 / 冗長解 / 派生問題生成 / 学習目標の空白域 / 気づき支援 / 継続的強化学習 / 報酬獲得解 / 生起確率ベクトル空間 / 凸包 / 一括強化学習 / 上達過程 / 人工知能 / 自律学習システム / コーチ機能 |
|---|
| Outline of Final Research Achievements |
This research proposed the autonomously continuous learning system by visualizing the learning processes which is easy to understand for a human. The main objective of this research is the continuous improvement for a human learning skill and the visualization of the improvement process. In this research, we investigated the method for indirectly visualizing the gap area of undiscovered goals by visualizing the positional relation among derived goals in an automated way. The result of the comparative experiment by the human subjects suggested that it is important that the display condition which indicates the positional relation with the gap of learning (which is the distance between a new goal found by the learner and the area of known goals) as the learning feedback information during the improvement process. In other words, it is the condition to facilitate the awareness of the learner’s unknown sense of values.
|
| Academic Significance and Societal Importance of the Research Achievements |
近年，注目されている深層学習の主な弱点は(1)人が実現不能な学習手法と(2)内部の学習過程の理解困難さである．これに対し，本研究では深層学習の弱点を補うため，(1)様々な問題を生成し提供することで，人が学習の仕方を学べる機能，(2)学習結果の解釈を行い，人が理解しやすくなるように学習過程・上達過程を可視化する機能を考案した．
本研究によって学習目標となる報酬設計が難しかった強化学習法の幅広い分野への適用が可能になる．また，自律学習システムは問題領域ごとに初期問題を与えると様々な派生問題とその解を反復的に生成するため，問題や解のバリエーションを大量に必要とするタスクに応用できる．
|
