Research map / Label-efficient and robust learning / Recognition and long-tail
Leveraged Weighted Loss for Partial Label Learning
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 LW loss instead of PRODEN | MLP · MNIST, q=0.1 | accuracy: 98.57 ± 0.07% -> 98.89 ± 0.06% | Helped |
| Use LW loss instead of PRODEN | MLP · Fashion-MNIST, q=0.1 | accuracy: 89.62 ± 0.13% -> 90.52 ± 0.19% | Helped |
| Use LW loss instead of PRODEN | MLP · Kuzushiji-MNIST, q=0.1 | accuracy: 92.20 ± 0.43% -> 94.14 ± 0.12% | Helped |
| Use LW loss instead of PRODEN | ConvNet · CIFAR-10, q=0.5 | accuracy: 85.95 ± 0.14% -> 88.31 ± 0.09% | Helped |
| Use cross-entropy rather than Sigmoid loss in LW | LW on MNIST, Fashion-MNIST, and Kuzushiji-MNIST · Benchmark datasets, q=0.1 | accuracy | Helped |
| Choose β=1 or β=2 rather than other tested values | LW-Sigmoid · Four benchmark datasets across varying q | accuracy | Helped |
| Set β=0 instead of a positive leverage parameter (ablation) | LW loss · Partial-label learning | classification risk | Hurt |
| Use losses on partial labels only, omitting losses on non-partial labels (ablation) | LW loss · Three benchmark datasets | accuracy | Hurt |
| Use losses on non-partial labels only, omitting losses on partial labels (ablation) | LW loss · Three benchmark datasets | accuracy | Hurt |
| Combine losses on partial and non-partial labels (ablation) | LW loss · Three benchmark datasets | accuracy | Helped |
| Simply rescale the losses rather than combine both loss components (ablation) | LW loss · Three benchmark datasets | accuracy | No clear effect |
| Increase ambiguity from Case 1 to Case 3 | All compared methods · Benchmark datasets, alternative generation cases | accuracy | Hurt |
Closest papers by meaning
- Partial-Label Regression (AAAI 2023 · similarity 0.91)
- Losses over Labels: Weakly Supervised Learning via Direct Loss Construction (AAAI 2023 · similarity 0.90)
- Forming Auxiliary High-confident Instance-level Loss to Promote Learning from Label Proportions (CVPR 2025 · similarity 0.88)
- Multi-Label Learning with Stronger Consistency Guarantees (NeurIPS 2024 · similarity 0.87)
- Learning with Partial-Label and Unlabeled Data: A Uniform Treatment for Supervision Redundancy and Insufficiency (ICML 2024 · similarity 0.86)
- Learning with Complementary Labels Revisited: The Selected-Completely-at-Random Setting Is More Practical (ICML 2024 · similarity 0.85)
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.