Research map / Learning theory and algorithms / Sample complexity
On-Demand Sampling: Learning Optimally from Multiple Distributions
Nika Haghtalab, Michael Jordan, Eric Zhao
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 |
|---|---|---|---|
| replace GDRO with R-MDL | ResNet-50 · CelebA, standard regularization | worst-group accuracy (%): 41.7 -> 88.9 | Improved |
| replace GDRO with R-MDL | ResNet-50 · Waterbirds, standard regularization | worst-group accuracy (%): 76.9 -> 86.4 | Improved |
| replace GDRO with R-MDL | BERT · MultiNLI, standard regularization | worst-group accuracy (%): 66.6 -> 70.3 | Improved |
| replace GDRO with R-MDL | ResNet-50 · Waterbirds, strong regularization | worst-group accuracy (%): 84.6 -> 89.4 | Improved |
| replace GDRO with R-MDL | ResNet-50 · CelebA, strong regularization | worst-group accuracy (%): 86.7 -> 88.8 | Improved |
| replace GDRO with R-MDL | BERT · MultiNLI, early stopping | worst-group accuracy (%): 77.7 -> 43.1 | Hurt |
| replace GDRO with R-MDL | ResNet-50 · CelebA, early stopping | worst-group accuracy (%): 88.3 -> 90.6 | Improved |
| remove strong weight-decay regularization | GDRO · Waterbirds | worst-group accuracy (%): 84.6 -> 76.9 | Hurt |
| remove strong weight-decay regularization | GDRO · CelebA | worst-group accuracy (%): 86.7 -> 41.7 | Hurt |
| remove strong weight-decay regularization | R-MDL · Waterbirds | worst-group accuracy (%): 89.4 -> 86.4 | No clear effect |
| remove strong weight-decay regularization | R-MDL · CelebA | worst-group accuracy (%): 88.8 -> 88.9 | No clear effect |
| replace GDRO with R-MDL | ResNet-50 · CelebA, standard regularization | gap between average and worst-group accuracy (%): 53 -> 3.4 | Improved |
Closest papers by meaning
- Sample-Adaptivity Tradeoff in On-Demand Sampling (NeurIPS 2025 · similarity 0.90)
- The Sample-Communication Complexity Trade-off in Federated Q-Learning (NeurIPS 2024 · similarity 0.87)
- Beyond Minimax Rates in Group Distributionally Robust Optimization via a Novel Notion of Sparsity (ICML 2025 · similarity 0.87)
- Group Distributionally Robust Optimization with Flexible Sample Queries (NeurIPS 2026 · similarity 0.86)
- Derandomizing Multi-Distribution Learning (NeurIPS 2024 · similarity 0.86)
- Streaming Federated Learning with Markovian Data (NeurIPS 2025 · 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.
