Abstract
The semi-supervised, density-based clustering algorithm SSDBSCAN extracts clusters of a given dataset from different density levels by using a small set of labeled objects. A critical assumption of SSDBSCAN is, however, that at least one labeled object for each natural cluster in the dataset is provided. This assumption may be unrealistic when only a very few labeled objects can be provided, for instance due to the cost associated with determining the class label of an object. In this paper, we introduce a novel active learning strategy to select "most representative" objects whose class label should be determined as input for SSDBSCAN. By incorporating a Laplacian Graph Regularizer into a Local Linear Reconstruction method, our proposed algorithm selects objects that can represent the whole data space well. Experiments on synthetic and real datasets show that using the proposed active learning strategy, SSDBSCAN is able to extract more meaningful clusters even when only very few labeled objects are provided.
| Original language | English |
|---|---|
| Title of host publication | Advances in Artificial Intelligence : Proceedings of the 27th Canadian Conference on Artificial Intelligence |
| Editors | M. Sokolova, P. van Beek |
| Publisher | Springer VS |
| Publication date | 2014 |
| Pages | 179-190 |
| ISBN (Print) | 978-3-319-06482-6 |
| ISBN (Electronic) | 978-3-319-06483-3 |
| DOIs | |
| Publication status | Published - 2014 |
| Externally published | Yes |
| Event | 27th Canadian Conference on Artificial Intelligence - Montreal, Canada Duration: 6. May 2014 → 9. May 2014 |
Conference
| Conference | 27th Canadian Conference on Artificial Intelligence |
|---|---|
| Country | Canada |
| City | Montreal |
| Period | 06/05/2014 → 09/05/2014 |
| Sponsor | 'Nana Traiteur par l'Assommoir', Canadian Artificial Intelligence Association (CAIAC), et al., GRAND (Graphics, Animation and New Media) Research Network, Grevin, Polytechnique Montreal |
| Series | Lecture Notes in Computer Science |
|---|---|
| Volume | 8436 |
| ISSN | 0302-9743 |
Fingerprint
Keywords
- Active learning
- Density-based clustering
- Semi-supervised clustering
Cite this
}
Active learning strategies for semi-supervised DBSCAN. / Li, Jundong; Sander, Jörg; Campello, Ricardo; Zimek, Arthur.
Advances in Artificial Intelligence: Proceedings of the 27th Canadian Conference on Artificial Intelligence. ed. / M. Sokolova; P. van Beek. Springer VS, 2014. p. 179-190 (Lecture Notes in Computer Science, Vol. 8436).Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review
TY - GEN
T1 - Active learning strategies for semi-supervised DBSCAN
AU - Li, Jundong
AU - Sander, Jörg
AU - Campello, Ricardo
AU - Zimek, Arthur
PY - 2014
Y1 - 2014
N2 - The semi-supervised, density-based clustering algorithm SSDBSCAN extracts clusters of a given dataset from different density levels by using a small set of labeled objects. A critical assumption of SSDBSCAN is, however, that at least one labeled object for each natural cluster in the dataset is provided. This assumption may be unrealistic when only a very few labeled objects can be provided, for instance due to the cost associated with determining the class label of an object. In this paper, we introduce a novel active learning strategy to select "most representative" objects whose class label should be determined as input for SSDBSCAN. By incorporating a Laplacian Graph Regularizer into a Local Linear Reconstruction method, our proposed algorithm selects objects that can represent the whole data space well. Experiments on synthetic and real datasets show that using the proposed active learning strategy, SSDBSCAN is able to extract more meaningful clusters even when only very few labeled objects are provided.
AB - The semi-supervised, density-based clustering algorithm SSDBSCAN extracts clusters of a given dataset from different density levels by using a small set of labeled objects. A critical assumption of SSDBSCAN is, however, that at least one labeled object for each natural cluster in the dataset is provided. This assumption may be unrealistic when only a very few labeled objects can be provided, for instance due to the cost associated with determining the class label of an object. In this paper, we introduce a novel active learning strategy to select "most representative" objects whose class label should be determined as input for SSDBSCAN. By incorporating a Laplacian Graph Regularizer into a Local Linear Reconstruction method, our proposed algorithm selects objects that can represent the whole data space well. Experiments on synthetic and real datasets show that using the proposed active learning strategy, SSDBSCAN is able to extract more meaningful clusters even when only very few labeled objects are provided.
KW - Active learning
KW - Density-based clustering
KW - Semi-supervised clustering
U2 - 10.1007/978-3-319-06483-3_16
DO - 10.1007/978-3-319-06483-3_16
M3 - Article in proceedings
AN - SCOPUS:84901657429
SN - 978-3-319-06482-6
T3 - Lecture Notes in Computer Science
SP - 179
EP - 190
BT - Advances in Artificial Intelligence
A2 - Sokolova, M.
A2 - van Beek, P.
PB - Springer VS
ER -