2022 Fiscal Year Final Research Report
Organizing Videos by Human-in-the-Loop Machine Learning for for Skill Transfer and Know-How Sharing
| Project/Area Number |
20K12115
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Multi-year Fund |
|---|
| Section | 一般 |
|---|
| Review Section |
Basic Section 62030:Learning support system-related
|
|---|
| Research Institution | Tokyo University of Science |
|---|
Principal Investigator |
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
古田 諒佑 東京大学, 生産技術研究所, 助教 (20843535)
|
|---|
| Project Period (FY) |
2020-04-01 – 2023-03-31
|
| Keywords | 人間参加型機械学習 / Human-in-the-loop / 弱教師あり学習 / 技能伝承 / 画像認識 |
|---|
| Outline of Final Research Achievements |
This research aims to promote skill transfer and know-how sharing by automatically organizing unedited work videos --- dividing them into semantic scenes and tagging them --- using image recognition. To reduce the annotation burden associated with the development of image recognition models, we investigated the following two approaches: (1) weakly supervised and transfer learning approach, which trains image recognition models by taking advantage of the small amount of incomplete annotation data provided by the user, and (2) a human-in-the-loop machine learning approach, which improves image recognition models by presenting the image recognition results to the user and obtaining the user feedback.
|
| Free Research Field |
知覚情報処理
|
| Academic Significance and Societal Importance of the Research Achievements |
製造業・看護など様々な業界において，熟練者から初心者への技能伝承，組織内でのノウハウ共有が重要な課題となっている．技能伝承・ノウハウは，企業やコミュニティに閉じたドメイン固有の概念を扱うため，学習データの整備，汎用的な画像認識モデルの構築が難しいことが問題となっている．本研究成果は，この問題点の解決に向けたもので，技能伝承・ノウハウ共有の促進に資するものである．なお，ここで開発した手法は，学習データ整備の困難を軽減するもので，当初想定した分野以外への展開も可能であることを示した．
|
