Research map / Label-efficient and robust learning / Recognition and long-tail
Learning with Multiple Complementary Labels
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use upper-bound surrogate losses (LOG/EXP) in the unbiased risk estimator instead of raw losses | unbiased risk estimator for MCLs · UCI, MNIST, Kuzushiji, Fashion, 20Newsgroups, CIFAR-10 (linear and NN) | accuracy | Helped |
| Replace unbounded CCE loss with bounded losses (MAE, MSE, GCE, PHuber-CE) in the empirical risk estimator (ablation) | unbiased risk estimator for MCLs · MNIST, Fashion-MNIST, Kuzushiji, CIFAR-10 (test accuracy curves) | accuracy | Helped |
| Decompose MCLs into single CLs (wrappers) instead of treating each MCL set as a whole with ordinary CL methods | ordinary complementary-label learning methods · UCI, Table 2-4 benchmarks | accuracy | Hurt |
| Use the Forward CL method inside the decomposition wrapper | wrapper with Forward (Yu et al., 2018) · MNIST / Texture / Fashion (linear and NN) | accuracy | Mixed |
| Use MAE (bounded, no upper-bound surrogate) instead of LOG in the risk estimator (ablation) | MAE loss on 20Newsgroups MLP · 20Newsgroups | accuracy: MAE 49.83% vs LOG 79.46% | Hurt |
| Increase fixed complementary-label set size j from 1 to k-2 (ablation) | proposed approaches · Appendix E.3 experiments | accuracy | Helped |
Closest papers by meaning
- Learning from Complementary Labels (NeurIPS 2017 · similarity 0.92)
- Learning with Complementary Labels Revisited: The Selected-Completely-at-Random Setting Is More Practical (ICML 2024 · similarity 0.90)
- Learning with Partial-Label and Unlabeled Data: A Uniform Treatment for Supervision Redundancy and Insufficiency (ICML 2024 · similarity 0.87)
- Addressing Multi-Label Learning with Partial Labels: From Sample Selection to Label Selection (AAAI 2025 · similarity 0.87)
- Partial Label Learning with a Partner (AAAI 2024 · similarity 0.86)
- Rebalancing Multi-Label Class-Incremental Learning (AAAI 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.