Research map / Learning theory and algorithms / Prediction and decision losses
Incorporating Unlabeled Data into Distributionally Robust Learning
Charlie Frogner, Sebastian Claici, Edward Chien, Justin Solomon
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 |
|---|---|---|---|
| constrain adversary with unlabeled data instead of Wasserstein ball only | distributionally robust logistic regression · UCI binary datasets, N_l up to 1000 | performance bound (likelihood) | Improved |
| use strong prior on label probabilities instead of weak prior | DRL with unlabeled data · UCI binary datasets | performance bound | Improved |
| choose epsilon by cross-validation in traditional Wasserstein DRL | traditional DRL · six UCI datasets | out-of-sample likelihood | Mixed |
| vary radius epsilon in DRL with unlabeled data | DRL with unlabeled data · six UCI datasets | out-of-sample likelihood | No clear effect |
| use distributionally robust heuristic instead of EMC, Min MC, Max MC for active learning | DR active learning · 14 UCI datasets | area under likelihood curve: outperform on 8 of 14 datasets | Improved |
| use distributionally robust heuristic instead of random sampling | DR active learning · 14 UCI datasets | area under likelihood curve: underperformed random on 5 of 14 | Mixed |
| use existing model change heuristics instead of random sampling | EMC, Min MC, Max MC · 14 UCI datasets | area under likelihood curve: underperform on 10 of 14 | Hurt |
| use strong prior instead of weak prior in DR active learning | DR active learning · 14 UCI datasets | area under likelihood curve: 69.8 -> 69.8 | No clear effect |
| use DR (strong) instead of Random on Letter (U/V) | DR active learning · Letter (U/V) | area under likelihood curve: 77.8 -> 82.3 | Improved |
| use DR (strong) instead of Random on Magic | DR active learning · Magic | area under likelihood curve: 52.2 -> 43.1 | Hurt |
Closest papers by meaning
- Distributionally Robust Optimization with Bias and Variance Reduction (ICLR 2024 · similarity 0.88)
- Gradient Flow Provably Learns Robust Classifiers for Orthonormal GMMs (ICML 2025 · similarity 0.87)
- Outlier-Robust Distributionally Robust Optimization via Unbalanced Optimal Transport (NeurIPS 2024 · similarity 0.87)
- Structure-informed Risk Minimization for Robust Ensemble Learning (ICML 2025 · similarity 0.87)
- Achievable distributional robustness when the robust risk is only partially identified (NeurIPS 2024 · similarity 0.87)
- DRF: Improving Certified Robustness via Distributional Robustness Framework (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.
