| Item type |
SIG Technical Reports(1) |
| 公開日 |
2022-05-05 |
| タイトル |
|
|
タイトル |
Bio-Medical Data Classification Approaches with Limited Annotation |
| タイトル |
|
|
言語 |
en |
|
タイトル |
Bio-Medical Data Classification Approaches with Limited Annotation |
| 言語 |
|
|
言語 |
eng |
| キーワード |
|
|
主題Scheme |
Other |
|
主題 |
D論セッション |
| 資源タイプ |
|
|
資源タイプ識別子 |
http://purl.org/coar/resource_type/c_18gh |
|
資源タイプ |
technical report |
| 著者所属 |
|
|
|
Kyushu University |
| 著者所属 |
|
|
|
Kyushu University |
| 著者所属(英) |
|
|
|
en |
|
|
Kyushu University |
| 著者所属(英) |
|
|
|
en |
|
|
Kyushu University |
| 著者名 |
Shota, Harada
Seiichi, Uchida
|
| 著者名(英) |
Shota, Harada
Seiichi, Uchida
|
| 論文抄録 |
|
|
内容記述タイプ |
Other |
|
内容記述 |
This paper aims to solve the problem of limited annotation in several bio-medical data analysis tasks. Specifically, group-based labeling utilizing constrained clustering and semi-supervised learning are proposed as the approaches. For group-based labeling utilizing constrained clustering, I proposed a new constrained clustering method, where a user attaches annotations to several sample pairs. Annotations are two types: cannot-link and must-link. The pair with cannot-link should not belong to the same cluster, whereas the pair with must-link should belong. These annotations are useful especially for medical data, because medical experts can have a more expected clustering result by a small number of annotations. Moreover, those annotations are treated as soft-constraints and therefore medical experts can attach them without extreme carefulness. For semi-supervised learning in bio-medical data classification tasks, I proposed order-guided disentangled representation learning. This method performs disentangled representation learning with prior knowledge that is effective for learning bio-medical data classification tasks. This method could improve classification performance even with limited annotation by effectively utilizing the prior knowledge through disentangled representation learning. |
| 論文抄録(英) |
|
|
内容記述タイプ |
Other |
|
内容記述 |
This paper aims to solve the problem of limited annotation in several bio-medical data analysis tasks. Specifically, group-based labeling utilizing constrained clustering and semi-supervised learning are proposed as the approaches. For group-based labeling utilizing constrained clustering, I proposed a new constrained clustering method, where a user attaches annotations to several sample pairs. Annotations are two types: cannot-link and must-link. The pair with cannot-link should not belong to the same cluster, whereas the pair with must-link should belong. These annotations are useful especially for medical data, because medical experts can have a more expected clustering result by a small number of annotations. Moreover, those annotations are treated as soft-constraints and therefore medical experts can attach them without extreme carefulness. For semi-supervised learning in bio-medical data classification tasks, I proposed order-guided disentangled representation learning. This method performs disentangled representation learning with prior knowledge that is effective for learning bio-medical data classification tasks. This method could improve classification performance even with limited annotation by effectively utilizing the prior knowledge through disentangled representation learning. |
| 書誌レコードID |
|
|
収録物識別子タイプ |
NCID |
|
収録物識別子 |
AA11131797 |
| 書誌情報 |
研究報告コンピュータビジョンとイメージメディア（CVIM）
巻 2022-CVIM-230,
号 1,
p. 1-16,
発行日 2022-05-05
|
| ISSN |
|
|
収録物識別子タイプ |
ISSN |
|
収録物識別子 |
2188-8701 |
| Notice |
|
|
|
SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc. |
| 出版者 |
|
|
言語 |
ja |
|
出版者 |
情報処理学会 |
IPSJ-CVIM22230001.pdf (12.0 MB)
