Research map / Optimization / Nonconvex and smooth optimization
Scalable Distributional Robustness in a Class of Non-Convex Optimization with Guarantees
Avinandan Bose, Arunesh Sinha, Tien Anh Mai
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 stratified sampling instead of k-means clustering for abstraction | MISOCP DRO · Synthetic SSG, S=24 | % of OPT: 94.03 -> 99.25 | Improved |
| Use stratified sampling (2 per strata) instead of clustering at small sample size | MISOCP DRO · Synthetic SSG, S=8 | % of OPT: 80.17 -> 91.78 | Improved |
| Increase samples S from 8 to 24 with clustering (ablation) | Clustering abstraction · Synthetic SSG | % of OPT: 80.17 -> 94.03 | Improved |
| Increase samples per stratum from 2 to 4 at S=8 (ablation) | Stratified sampling · Synthetic SSG, S=8 | % of OPT: 91.78 -> 78.89 | Hurt |
| Use clustering instead of PGA at high regularization xi=1E6 | MISOCP DRO · Synthetic SSG, N=500, M=10 | % of OPT: 49.2 -> 99.6 | Improved |
| Use clustering instead of TT-GAD at xi=1E6 | MISOCP DRO · Synthetic SSG, N=500, M=10 | % of OPT: 82.6 -> 99.6 | Improved |
| Use clustering instead of gradient baselines at low regularization xi=1E3 | MISOCP DRO · Synthetic SSG | % of OPT: 99.8 -> 99.9 | No clear effect |
| Increase regularization xi from 1E3 to 1E6 for PGA (ablation) | PGA baseline · Synthetic SSG | % of OPT: 98.9 -> 49.2 | Hurt |
| Increase regularization xi from 1E3 to 1E6 for clustering (ablation) | Clustering abstraction · Synthetic SSG | % of OPT: 99.9 -> 99.6 | No clear effect |
| Increase number of alternatives M from 10 to 50 for PGA (ablation) | PGA baseline · Synthetic SSG | % of OPT: 49.2 -> 93.5 | Improved |
| Use sampling instead of PGA at M=50 | MISOCP DRO · Synthetic SSG | % of OPT: 93.5 -> 99.5 | Improved |
| Use robust DRO (xi=1E2) instead of ERM for lowest-5% clients | MC-FCP m=10 · P&R-NYC test | avg choice probability (H): 0.150 -> 0.417 | Improved |
Closest papers by meaning
- Large-Scale Non-convex Stochastic Constrained Distributionally Robust Optimization (AAAI 2024 · similarity 0.91)
- Distributionally Robust Optimization with Bias and Variance Reduction (ICLR 2024 · similarity 0.89)
- Wasserstein Distributionally Robust Optimization through the Lens of Structural Causal Models and Individual Fairness (NeurIPS 2024 · similarity 0.89)
- Outlier-Robust Distributionally Robust Optimization via Unbalanced Optimal Transport (NeurIPS 2024 · similarity 0.88)
- Efficient Stochastic Approximation of Minimax Excess Risk Optimization (ICML 2024 · similarity 0.88)
- Drago: Primal-Dual Coupled Variance Reduction for Faster Distributionally Robust Optimization (NeurIPS 2024 · similarity 0.88)
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.
