Research map / Learning theory and algorithms / Sample complexity
Interpretable Distribution Features with Maximum Testing Power
Wittawat Jitkrittum, Zoltán Szabó, Kacper P Chwialkowski, 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 |
|---|---|---|---|
| Optimize test locations and kernel width instead of drawing locations randomly and tuning only kernel width (ablation) | ME test · NIPS category comparisons | test power: Bayes-Deep: .034 -> .954; Bayes-Learn: .774 -> .990; Bayes-Neuro: .300 -> 1.00; Learn-Deep: .052 -> .956; Learn-Neuro: . | Improved |
| Optimize test locations and kernel width instead of drawing locations randomly and tuning only kernel width (ablation) | SCF test · NIPS category comparisons | test power: Bayes-Deep: .180 -> .688; Bayes-Learn: .534 -> .836; Bayes-Neuro: .500 -> .828; Learn-Deep: .138 -> .656; Learn-Neuro: . | Improved |
| Optimize test locations and kernel width instead of drawing locations randomly and tuning only kernel width (ablation) | ME test · KDEF + vs. - | test power: .656 -> .998 | Improved |
| Optimize test locations and kernel width instead of drawing locations randomly and tuning only kernel width (ablation) | SCF test · KDEF + vs. - | test power: .750 -> 1.00 | Improved |
| Optimize test locations rather than use random locations (ablation) | ME test · Gaussian variance difference (GVD) | test power | Improved |
| Optimize test locations rather than use random locations (ablation) | SCF test · Gaussian variance difference (GVD) | test power | Improved |
| Optimize test locations rather than use random locations (ablation) | ME test · Blobs | test power | Improved |
| Use frequency-domain features rather than spatial features | Full-feature tests · Blobs | test power | Improved |
| Use spatial features rather than frequency features | Full-feature tests · NIPS category comparisons | test power | Improved |
| Use optimized ME locations rather than randomly drawn locations (ablation) | ME test · Gaussian mean difference as dimension increases | test power: ME-full achieves maximum test power of 1.0 throughout | Improved |
| Use a linear-time ME test with learned features instead of quadratic-time MMD | ME-full · Gaussian variance difference | test power | Improved |
| Use a linear-time SCF test with learned frequency features instead of quadratic-time MMD | SCF-full · Blobs | test power | Improved |
Closest papers by meaning
- An Adaptive Test of Independence with Analytic Kernel Embeddings (ICML 2017 · similarity 0.83)
- Testing and Learning on Distributions with Symmetric Noise Invariance (NeurIPS 2017 · similarity 0.83)
- Semiparametric Efficient Tests for Interpretable Distributional Treatment Effects (NeurIPS 2026 · similarity 0.81)
- Revealing Distribution Discrepancy by Sampling Transfer in Unlabeled Data (NeurIPS 2024 · similarity 0.80)
- Efficiently Learning Significant Fourier Feature Pairs for Statistical Independence Testing (NeurIPS 2024 · similarity 0.80)
- "What is Different Between These Datasets?" A Framework for Explaining Data Distribution Shifts (NeurIPS 2026 · similarity 0.80)
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.
