Research map / Learning theory and algorithms / Prediction and decision losses
Large-Scale Robust Deep AUC Maximization: A New Surrogate Loss and Empirical Studies on Medical Image Classification
Zhuoning Yuan, Yan Yan, Milan Sonka, Tianbao Yang
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 |
|---|---|---|---|
| use AUC margin loss instead of AUC square loss | DAM · C10 (imratio 1%, DenseNet121) | test AUC: 0.745 -> 0.760 | Improved |
| use AUC margin loss instead of AUC square loss | DAM · S10 (imratio 1%, DenseNet121) | test AUC: 0.669 -> 0.703 | Improved |
| use AUC margin loss instead of cross-entropy | DenseNet121 · C2 (imratio 1%) | test AUC: 0.718 -> 0.809 | Improved |
| use AUC margin loss instead of cross-entropy at imratio 10% | ResNet20 · C10 (imratio 10%) | test AUC: 0.898 -> 0.898 | No clear effect |
| use AUC margin loss instead of cross-entropy at imratio 10% | DenseNet121 · C10 (imratio 10%) | test AUC: 0.898 -> 0.887 | Hurt |
| use AUC margin loss instead of AUC square loss | DAM · Melanoma private (w/ TTA 30) | test AUC: 0.9364 -> 0.9393 | Improved |
| add patient metadata network | DAM AUC-M · Melanoma private (w/ TTA 30) | test AUC: 0.9393 -> 0.9423 | Improved |
| use AUC margin loss instead of cross-entropy | EfficientNet-B5 · Melanoma private (w/ TTA 30) | test AUC: 0.9345 -> 0.9393 | Improved |
| use AUC margin loss instead of cross-entropy | DenseNet121 · PatchCamelyon (imratio 1%) | test AUC: 0.8394 -> 0.8896 | Improved |
| add batch score normalization (ablation) | DAM DenseNet121 · benchmark datasets (imratio 1% and 10%) | test AUC | Improved |
| use AUC margin loss instead of AUC square loss with noisy and easy data added (ablation) | DAM DenseNet121 · C2-IB with noisy and easy data | test AUC | Improved |
| add non-negative constraint on alpha (ablation) | AUC-M · C2-IB (imratio 1%, 40% easy and 1% noisy added) | test AUC | Improved |
Closest papers by meaning
- Compositional Training for End-to-End Deep AUC Maximization (ICLR 2022 · similarity 0.92)
- Provable Multi-instance Deep AUC Maximization with Stochastic Pooling (ICML 2023 · similarity 0.87)
- AUCSeg: AUC-oriented Pixel-level Long-tail Semantic Segmentation (NeurIPS 2024 · similarity 0.86)
- Towards Nonlinear Sparse AUC Maximization via Compositional Stochastic Hard Thresholding (AAAI 2026 · similarity 0.85)
- Federated Deep AUC Maximization for Hetergeneous Data with a Constant Communication Complexity (ICML 2021 · similarity 0.85)
- Pairwise AUC Optimization Needs Corrective Power: A Unified View (NeurIPS 2026 · similarity 0.83)
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.
