Research map / Label-efficient and robust learning / Domain adaptation and generalization
Demystifying Disagreement-on-the-Line in High Dimensions
Donghwan Lee, Behrad Moniri, Xinmeng Huang, Edgar Dobriban, Hamed Hassani
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 shared-sample disagreement instead of risk as the source-domain signal | Random features regression · CIFAR-10-C, Tiny ImageNet-C, and Camelyon17 | source-target linear relationship | Improved |
| Use risk rather than disagreement as the source-domain signal | Random features regression · Tiny ImageNet and Camelyon17 | source-target linear relationship | Hurt |
| Use independent disagreement rather than shared-sample disagreement (ablation) | Random features regression · Ridgeless, overparameterized regime | source-target disagreement linearity | No clear effect |
| Share the training sample between models (ablation) | Random features regression · Ridgeless, overparameterized regime | source-target disagreement linearity | Improved |
| Share model weights rather than training samples (ablation) | Random features regression · Overparameterized regime | source-target disagreement linearity | Hurt |
| Add ridge regularization (ablation) | Independent and shared-sample disagreement · Overparameterized regime | exact source-target linearity | Hurt |
| Add ridge regularization (ablation) | Independent and shared-sample disagreement · Overparameterized regime | approximate source-target linearity | Improved |
| Move from the overparameterized to the underparameterized regime (ablation) | Independent and shared-sample disagreement · Ridgeless random features regression | source-target disagreement linearity | Hurt |
| Increase feature width toward infinite overparameterization (ablation) | Shared-sample disagreement · Ridgeless random features regression | disagreement | Improved |
| Increase feature width toward infinite overparameterization (ablation) | Independent and shared-weight disagreement · Ridgeless random features regression | disagreement | No clear effect |
| Allow concept shift between source and target (ablation) | Disagreement-on-the-line · Distribution shift | source-target disagreement linearity | No clear effect |
Closest papers by meaning
- On the Bayes Inconsistency of Disagreement Discrepancy Surrogates (ICLR 2026 · similarity 0.88)
- Agree to Disagree: Diversity through Disagreement for Better Transferability (ICLR 2023 · similarity 0.88)
- Forest vs Tree: The (N, K) Trade-off in Reproducible ML Evaluation (AAAI 2026 · similarity 0.86)
- Mitigating Spurious Correlations via Disagreement Probability (NeurIPS 2024 · similarity 0.85)
- How many classifiers do we need? (NeurIPS 2024 · similarity 0.85)
- "What is Different Between These Datasets?" A Framework for Explaining Data Distribution Shifts (NeurIPS 2026 · 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.
