Research map / Label-efficient and robust learning / Recognition and long-tail
Every Error has Its Magnitude: Asymmetric Mistake Severity Training for Multiclass Multiple Instance Learning
Sungrae Hong, Jiwon Jeong, Jisu Shin, Donghee Han, Sol Lee, Kyungeun Kim, Mun Yong Yi
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| add mistake severity cross-entropy L_MSCE (ablation) | PAMS · BRACS (TransMIL) | AsMC: 47.10 -> 55.65 | Improved |
| add mistake severity cross-entropy L_MSCE (ablation) | PAMS · BRACS (DTFD-MIL-AFS) | AsMC: 54.45 -> 54.64 | No clear effect |
| add hierarchy alignment loss L_HA (ablation) | PAMS · BRACS (TransMIL) | ACC: 45.21 -> 47.59 | Improved |
| add Semantic Feature Remix (SFR) (ablation) | PAMS · BRACS (TransMIL) | AsMC: 47.92 -> 55.65 | Improved |
| add Semantic Feature Remix (SFR) (ablation) | PAMS · BRACS (TransMIL) | ACC: 47.51 -> 47.59 | No clear effect |
| SFR instead of random remix (ablation) | CE + TransMIL · In-house colon | AsMC: 54.34 -> 55.86 | Improved |
| random or pseudo remix on in-house data (ablation) | CE + TransMIL · In-house colon | ACC: 90.41 -> 89.25 | Hurt |
| add any remix strategy (ablation) | CE + TransMIL · In-house colon | AsMC: 44.16 -> 51.64 | Improved |
| PAMS vs prior priority learning (Hong et al.) on BRACS | TransMIL · BRACS | AsMC: 45.54 -> 55.65 | Improved |
| MSCE loss instead of CE on CIFAR-10 with class priority | ResNet-18 · CIFAR-10 with class priority | AsMC: 34.84 -> 35.70 | Improved |
| class-distance weighted CE instead of CE | ResNet-18 · CIFAR-10 with class priority | AsMC: 34.84 -> 34.63 | No clear effect |
| weighted CE (1:10) instead of CE | ResNet-18 · CIFAR-10 with class priority | ACC: 83.24 -> 73.69 | Hurt |
Closest papers by meaning
- CaMIL: Causal Multiple Instance Learning for Whole Slide Image Classification (AAAI 2024 · similarity 0.87)
- TaxoMIL: Taxonomy-Constrained Learning for Hierarchical Whole Slide Image Analysis (ECCV 2026 · similarity 0.87)
- Advancing Multiple Instance Learning with Continual Learning for Whole Slide Imaging (CVPR 2025 · similarity 0.86)
- Queryable Prototype Multiple Instance Learning with Vision-Language Models for Incremental Whole Slide Image Classification (AAAI 2025 · similarity 0.85)
- Dynamic Policy-Driven Adaptive Multi-Instance Learning for Whole Slide Image Classification (CVPR 2024 · similarity 0.84)
- How Effective Can Dropout Be in Multiple Instance Learning ? (ICML 2025 · similarity 0.84)
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.
