Discovering latent class labels for multi-label learning
- Submitting institution
-
University of Greenwich
- Unit of assessment
- 11 - Computer Science and Informatics
- Output identifier
- 29895
- Type
- E - Conference contribution
- DOI
-
10.24963/ijcai.2020/423
- Title of conference / published proceedings
- Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence
- First page
- 3058
- Volume
- 0
- Issue
- -
- ISSN
- -
- Open access status
- -
- Month of publication
- -
- Year of publication
- 2020
- URL
-
-
- Supplementary information
-
-
- Request cross-referral to
- -
- Output has been delayed by COVID-19
- No
- COVID-19 affected output statement
- -
- Forensic science
- No
- Criminology
- No
- Interdisciplinary
- No
- Number of additional authors
-
4
- Research group(s)
-
-
- Citation count
- -
- Proposed double-weighted
- No
- Reserve for an output with double weighting
- No
- Additional information
- Existing multi-label learning approaches are only able to classify datasets for which known labels are available. Labelling efforts are usually focused on a set of target labels (observed/known labels) and labels outside this set (unobserved/hidden labels) will not be considered. To address this issue, this paper proposes a novel approach to predicting new data instances simultaneously with hidden and known labels. The proposed approach not only discovers latent labels beyond what existing approaches can achieve, but also significantly improves the classification accuracy in comparison to the state-of-the-art. The acceptance rate for IJCAI 2020 was 12.6%.
- Author contribution statement
- -
- Non-English
- No
- English abstract
- -
