Research map / Learning theory and algorithms / Kernels and regression theory
Integral Probability Metrics Meet Neural Networks: The Radon-Kolmogorov-Smirnov Test
Seunghoon Paik, Michael Celentano, Alden Green, Ryan J. Tibshirani
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 RKS test instead of Gaussian kernel MMD and energy distance when P and Q differ in one direction | RKS test · first four synthetic settings (pancake-shift, ball-shift, t-coord, var-one), d=2,4,8,16 | ROC / power | Improved |
| use RKS test instead of Gaussian kernel MMD and energy distance when P and Q differ in all directions | RKS test · var-all, larger d | ROC / power | Hurt |
| use smoothness degree k=0 in RKS (ablation) | RKS test · pancake-shift | ROC / power | Improved |
| increase smoothness degree to k=1 (ablation) | RKS test · t-coord | ROC / power | Improved |
| increase smoothness degree to k=2 or 3 (ablation) | RKS test · var-one | ROC / power | Improved |
| vary smoothness degree k=0,1,2 (ablation) | RKS test · ball-shift | ROC / power | No clear effect |
| increase dimension d from 2 to 16 (ablation) | all tests · pancake-shift, var-one | ROC / power | Hurt |
| aggregate RKS p-values over degrees k with min-rule or Fisher-rule (ablation) | RKS test · five synthetic settings, d=2,4,8,16 | ROC / power | Improved |
| increase number of neurons N (overparametrization) (ablation) | RKS optimization · RKS distance computation | optimization stability | Improved |
Closest papers by meaning
- Hierarchical Integral Probability Metrics: A distance on random probability measures with low sample complexity (ICML 2024 · similarity 0.83)
- Learning from higher-order correlations, efficiently: hypothesis tests, random features, and neural networks (NeurIPS 2024 · similarity 0.83)
- Sharp Generalization for Nonparametric Regression by Over-Parameterized Neural Networks: A Distribution-Free Analysis in Spherical Covariate (ICML 2025 · similarity 0.83)
- Kernel conditional tests from learning-theoretic bounds (NeurIPS 2025 · similarity 0.82)
- The $\varphi$ Curve: The Shape of Generalization through the Lens of Norm-based Capacity Control (NeurIPS 2025 · similarity 0.82)
- Kernel Quantile Embeddings and Associated Probability Metrics (ICML 2025 · similarity 0.82)
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.
