Research map / Label-efficient and robust learning / OOD and anomaly detection
Agree to Disagree: Diversity through Disagreement for Better Transferability
Matteo Pagliardini, Martin Jaggi, François Fleuret, Sai Praneeth Karimireddy
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 D-BAT disagreement term on unlabeled OOD data (ensemble of 2) vs ERM | LeNet on C-MNIST · C-MNIST | test accuracy: 12.3 -> 90.2 | Improved |
| Add D-BAT disagreement term on unlabeled OOD data vs ERM | LeNet on M/C-Dominoes · M/C-D | test accuracy: 50.0 -> 73.3 | Improved |
| Add D-BAT disagreement term when D_ood differs from test domain vs ERM | LeNet on M/C-Dominoes · M/C-D | test accuracy: 50.0 -> 58.0 | Improved |
| Add D-BAT disagreement term on unlabeled test-like OOD data vs ERM | ResNet-50 on Waterbirds (single model) · Waterbirds | test accuracy: 86.0 -> 88.7 | Improved |
| Add D-BAT disagreement term on Office-Home Art domain vs ERM ensemble | ResNet-18 on Office-Home · Office-Home | test accuracy: 50.4 -> 51.1 single; 53.9 -> 54.5 ensemble (D_ood≠test) | Mixed |
| Add D-BAT disagreement term on Camelyon17 unlabeled test data vs ERM | ResNet-50 on Camelyon17 · Camelyon17 | test accuracy: 80.3 -> 93.1 single; 80.9 -> 91.9 ensemble | Improved |
| Add D-BAT disagreement with out-of-domain unlabeled validation data (Camelyon17) | ResNet-50 on Camelyon17 · Camelyon17 (D_ood ≠ test) | test accuracy: 80.3 -> 88.8 single; 80.9 -> 85.9 ensemble | Improved |
| Add D-BAT disagreement on OOD samples with CIFAR-10 classes 5-9 (ensemble of 4) | modified ResNet-18 on CIFAR-10 uncertainty · CIFAR-10 vs CIFAR-100 OOD | average ensemble accuracy: 92% (deep ensembles) -> 91.2% (D-BAT) | Mixed |
| Add D-BAT disagreement on OOD data for uncertainty (predicted probabilities on OOD samples are lower) | CIFAR-10 ensembles · CIFAR-100 test as OOD | predicted probability on OOD samples | Improved |
| Sequential vs simultaneous training of D-BAT ensemble members on 2D toy task (ablation) | D-BAT 2D toy · 2D toy | qualitative solution simplicity | Improved |
| Add D-BAT disagreement term on ERM baseline with spurious feature perfectly correlated (C-MNIST/dominoes) | LeNet ERM baseline · C-MNIST, M/F-D, M/C-D | test accuracy | No clear effect |
| Using a larger ensemble size (5) vs 2 for D-BAT (ablation) | D-BAT on Waterbirds and Office-Home · Waterbirds, Office-Home | test accuracy vs ensemble size | No clear effect |
Closest papers by meaning
- Demystifying Disagreement-on-the-Line in High Dimensions (ICML 2023 · similarity 0.88)
- A Unified Theory of Diversity in Ensemble Learning (JMLR 2023 · similarity 0.87)
- Sharpness-diversity tradeoff: improving flat ensembles with SharpBalance (NeurIPS 2024 · similarity 0.86)
- Aggregation Hides Out-of-Distribution Generalization Failures from Spurious Correlations (NeurIPS 2025 · similarity 0.86)
- Unraveling the Key Components of OOD Generalization via Diversification (ICLR 2024 · similarity 0.85)
- Spurious Feature Diversification Improves Out-of-distribution Generalization (ICLR 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.
