Research map / Label-efficient and robust learning / Recognition and long-tail
Task-Specific Fine-Tuning via Variational Information Bottleneck for Weakly-Supervised Pathology Whole Slide Image Classification
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 |
|---|---|---|---|
| Fine-tune the backbone with the proposed method | IN-1K features with max-pooling · Camelyon-16 | F1: 0.805 0.012 -> 0.927 0.011 | Helped |
| Fine-tune the backbone with the proposed method | IN-1K features with DTFD-MIL · TCGA-BRCA | AUC: 0.895 0.042 -> 0.951 0.016 | Helped |
| Fine-tune the backbone with the proposed method | IN-1K features with DTFD-MIL · LBP-CECA | AUC: 0.847 0.003 -> 0.922 0.005 | Helped |
| Fine-tune the backbone | MoCo-pretrained features with CLAM-SB · TCGA-BRCA | AUC: 0.904 0.030 -> 0.948 0.026 | Helped |
| Fine-tune the backbone | DINO-pretrained features with CLAM-SB · TCGA-BRCA | AUC: 0.891 0.043 -> 0.944 0.036 | Helped |
| Fine-tune the backbone | IN-1K features with max-pooling · Camelyon-16-C | F1: 0.689 -> 0.816 | Helped |
| Fine-tune the backbone | IN-1K features with CLAM-SB · Camelyon-17 | AUC: 0.702 -> 0.725 | Helped |
| Set backbone learning rate to 1e-5 (ablation) | WSI fine-tuning · Camelyon-16 | F1: 0.682 -> 0.899 | Helped |
| Set backbone learning rate to 1e-3 (ablation) | WSI fine-tuning · Camelyon-16 | F1: N/A | Hurt |
| Use Top-K 2048 instances rather than Top-K 512 (ablation) | IB-distilled WSI bag · Camelyon-16 | F1: 0.843 0.005 -> 0.846 0.004 | Helped |
| Use all instances rather than Top-K 2048 (ablation) | IB-distilled WSI bag · Camelyon-16 | F1: 0.846 0.004 -> 0.839 0.018 | Hurt |
| Set IB Lagrange multiplier beta to 1e-1 (ablation) | IB module · Camelyon-16 | F1: 0.839 0.018 -> 0.849 0.010 | Helped |
Closest papers by meaning
- CaMIL: Causal Multiple Instance Learning for Whole Slide Image Classification (AAAI 2024 · similarity 0.89)
- Generalizable Whole Slide Image Classification with Fine-Grained Visual-Semantic Interaction (CVPR 2024 · similarity 0.89)
- Dynamic Policy-Driven Adaptive Multi-Instance Learning for Whole Slide Image Classification (CVPR 2024 · similarity 0.88)
- PathVQ: Reforming Computational Pathology Foundation Model for Whole Slide Image Analysis via Vector Quantization (NeurIPS 2025 · similarity 0.88)
- Queryable Prototype Multiple Instance Learning with Vision-Language Models for Incremental Whole Slide Image Classification (AAAI 2025 · similarity 0.87)
- Transformer-Based Video-Structure Multi-Instance Learning for Whole Slide Image Classification (AAAI 2024 · similarity 0.87)
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.