Research map / Label-efficient and robust learning / Test-time adaptation
MMD Aggregated Two-Sample Test
Antonin Schrab, Ilmun Kim, Mélisande Albert, Béatrice Laurent, Benjamin Guedj, Arthur Gretton
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 |
|---|---|---|---|
| Aggregate tests across a wider bandwidth collection while keeping the well-calibrated central bandwidth highly weighted (ablation) | MMDAgg centred · Perturbed uniform distributions and MNIST | power | No clear effect |
| Use nine bandwidths instead of a narrower centred collection (ablation) | MMDAgg centred · MNIST | power | Improved |
| Use uniform weights across a wider bandwidth collection (ablation) | MMDAgg uniform · Perturbed uniform distributions and MNIST | power | Hurt |
| Use a wild bootstrap rather than permutations (ablation) | MMDAgg · Synthetic and real-world experiments | power: at most roughly 0.02 | No clear effect |
| Increase one sample while holding the other sample small (ablation) | MMDAgg and comparison tests · Perturbed uniform distributions and MNIST | power: 100 or 250; up to 5000; 0.25 | No clear effect |
| Use balanced sample sizes rather than a small sample paired with a very large sample (ablation) | MMDAgg and comparison tests · Perturbed uniform distributions and MNIST | power: 5100 instead of 1000; m=n=500 or m=n=2000 | Improved |
| Increase the sample size used by the linear-time ost test (ablation) | ost · Perturbed uniform d=1 | power: 75\,000 and 50\,000; m=n=500 | Improved |
| Increase the sample size used by the linear-time ost test (ablation) | ost · Perturbed uniform d=2 | power: one million, and half a million; 2000 samples | Improved |
| Increase the sample size used by the linear-time ost test (ablation) | ost · MNIST | power: 250\,000; 500 samples | Improved |
Closest papers by meaning
- A permutation-free kernel two-sample test (NeurIPS 2022 · similarity 0.81)
- MMD-Fuse: Learning and Combining Kernels for Two-Sample Testing Without Data Splitting (NeurIPS 2023 · similarity 0.81)
- Two Sample Testing in High Dimension via Maximum Mean Discrepancy (JMLR 2023 · similarity 0.80)
- Minimax Optimal Kernel Two-sample Testing in Sub-quadratic Time (NeurIPS 2026 · similarity 0.80)
- Kernel Selection is Model Selection: A Unified Complexity-Penalised Approach for MMD Two-Sample Tests (NeurIPS 2026 · similarity 0.79)
- DUAL: Learning Diverse Kernels for Aggregated Two-sample and Independence Testing (NeurIPS 2025 · similarity 0.76)
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.
