Research map / Learning theory and algorithms / Prediction and decision losses
Large-Scale Methods for Distributionally Robust Optimization
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 |
|---|---|---|---|
| use mini-batch gradient estimator (n=50-5K) instead of full-batch subgradient | penalized chi2 DRO · ImageNet | minutes to within 2% of optimum: 380 -> 16.5 | Helped |
| use mini-batch gradient estimator (n=50-5K) instead of full-batch subgradient | penalized chi2 DRO · Digits | minutes to within 2% of optimum: 37.0 -> 4.4 | Helped |
| increase batch size from 5K to 50K (ablation) | mini-batch estimator · ImageNet | minutes to within 2% of optimum: 16.5 -> 50 | Hurt |
| reduce batch size from 50 to 10 (ablation) | mini-batch estimator · Digits | reaches 2% of optimum: 24 -> $\infty$ | Hurt |
| replace mini-batch estimator with MLMC estimator (ablation) | penalized chi2 DRO · ImageNet | reaches 2% of optimum: 16.5 -> $\infty$ | Hurt |
| use DRO instead of tuned-l2 ERM | digit classifier · Digits worst sub-group loss | worst-group log loss: 17.5% to 27% improvement | Helped |
| use DRO instead of ERM | ImageNet linear classifier · ImageNet worst 10 classes vs average | log loss / accuracy: 5--10% worst-class log loss gain; average loss up to 10% worse | Mixed |
Closest papers by meaning
- Stochastic Gradient Methods for Distributionally Robust Optimization with f-divergences (NeurIPS 2016 · similarity 0.91)
- Distributionally Robust Optimization with Bias and Variance Reduction (ICLR 2024 · similarity 0.91)
- Large-Scale Non-convex Stochastic Constrained Distributionally Robust Optimization (AAAI 2024 · similarity 0.89)
- Distributionally Robust Performative Optimization (NeurIPS 2025 · similarity 0.87)
- Drago: Primal-Dual Coupled Variance Reduction for Faster Distributionally Robust Optimization (NeurIPS 2024 · similarity 0.87)
- Stability and Sharper Risk Bounds with Convergence Rate $\tilde{O}(1/n^2)$ (NeurIPS 2025 · 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.