Research map / Label-efficient and robust learning / Noisy labels
Multi-Label Iterated Learning for Image Classification With Label Ambiguity
Check 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 MILe instead of sigmoid BCE training on single labels | ResNet-50 · ImageNet ReaL, 100% | ReaL-F1: 78.61 -> 79.89 | Helped |
| Use MILe instead of sigmoid BCE training on single labels | ResNet-50 · ImageNet val, 100% | accuracy: 76.35 -> 77.12 | Helped |
| Use MILe instead of sigmoid BCE training on single labels | ResNet-18 · ImageNet ReaL, 100% | ReaL-F1: 73.85 -> 74.48 | Helped |
| Use MILe rather than the strongest listed non-MILe method | ResNet-50 · WebVision | Top-1: 73.15 -> 75.2 | Helped |
| Use MILe rather than self-distillation | ResNet-50 (1x+SK) · ImageNet val, 1% labels | accuracy: 70.15 -> 73.08 | Helped |
| Add the ColoredMNIST+ color-prediction task so teacher generations can transmit color labels alongside digit labels | MILe · ColoredMNIST+ | accuracy: 51.1 ± 0.1 -> 53.5 ± 0.6 | Helped |
| Use MILe rather than sigmoid training | Multi-label MNIST · Multi-label MNIST | F1@0.25: 29.10 -> 41.35 | Helped |
| Use MILe rather than CE-Sigmoid | CelebA · CelebA | F1-score: 80.14 -> 81.40 | Helped |
| Use MILe rather than sigmoid training to recover alternative labels | ResNet-50 · ImageNet ReaL alternative-label recovery, 100% | mean average precision: 0.2845 -> 0.3248 | Helped |
| Lower the teacher pseudo-label threshold (ablation) | MILe · ImageNet ReaL | ReaL-F1 | Helped |
| Increase the student training budget to extreme values (ablation) | MILe · ImageNet ReaL | ReaL-F1 | Hurt |
| Use the noisy-student iteration schedule instead of MILe's short-cycle schedule (ablation) | MILe · ImageNet ReaL | ReaL-F1 | Hurt |
Closest papers by meaning
- Knowledge Amalgamation for Multi-Label Classification via Label Dependency Transfer (AAAI 2023 · similarity 0.89)
- Multi-Label Learning From Single Positive Labels (CVPR 2021 · similarity 0.88)
- [CLS] is Not Enough: Multi-Label Recognition via Patch-Level Inference and Adaptive Aggregation (ICML 2026 · similarity 0.88)
- Re-Labeling ImageNet: From Single to Multi-Labels, From Global to Localized Labels (CVPR 2021 · similarity 0.88)
- Theory-Inspired Deep Multi-View Multi-Label Learning with Incomplete Views and Noisy Labels (CVPR 2025 · similarity 0.86)
- Rethinking Self-Distillation: Label Averaging and Enhanced Soft Label Refinement with Partial Labels (ICLR 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.