Research map / Learning theory and algorithms / Prediction and decision losses
RKHS-SHAP: Shapley Values for Kernel Methods
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 RKHS-ISV instead of KernelSHAP-ISV to estimate interventional Shapley values | RKHS-SHAP · Banana data (b in 1..100) | R^2 vs ground-truth ISVs | No clear effect |
| Use RKHS-OSV (CME-based, no density estimation) instead of Gaussian Monte Carlo OSV (gshap-osv) | RKHS-SHAP · Banana data (b in 1..100) | R^2 vs ground-truth OSVs | Helped |
| Aggregate feature maps of imputed data before evaluating f (RKHS-ISV) instead of averaging f over samples (KernelSHAP) | RKHS-SHAP · Banana data, n in 100..5000 | run time (log10 seconds) | Helped |
| Compute observational SVs (needs CME matrix inversion) instead of interventional SVs | RKHS-SHAP · Banana data, n in 100..5000 | run time (log10 seconds) | Hurt |
| Increase ISV-Reg strength lambda_s from 0 to 2.5 | Kernel ridge regression · Synthetic linear data with noise on X5 at test time (sigma' 0 to 1.5) | RMSE under covariate shift | Helped |
| Increase ISV-Reg strength lambda_s from 0 to 2.5 | Kernel ridge regression · Synthetic linear data, no noise (sigma'=0) | RMSE | Hurt |
| Increase OSV-Reg strength lambda_s | Kernel ridge regression · Synthetic data with sensitive feature X5 correlated with X4 | distribution of Shapley values of X5 and X4 | Helped |
| Regularise interventional SVs (ISV-Reg) instead of observational SVs (OSV-Reg) for fairness (ablation) | Shapley regulariser · Synthetic data with sensitive feature X5 correlated with X4 | distribution of Shapley values of X4 | Hurt |
| Replace Gaussian kernel with Matern kernel (nu=0.5) via non-product-kernel RKHS-SHAP (ablation) | Kernel ridge regression explanations · Diabetes and House price datasets | feature attribution ranking | Mixed |
Closest papers by meaning
- Provably Accurate Shapley Value Estimation via Leverage Score Sampling (ICLR 2025 · similarity 0.92)
- shapiq: Shapley Interactions for Machine Learning (NeurIPS 2024 · similarity 0.92)
- A Unified Framework for Provably Efficient Algorithms to Estimate Shapley Values (NeurIPS 2025 · similarity 0.90)
- Amortized Linear-time Exact Shapley Value for Product-Kernel Methods (NeurIPS 2026 · similarity 0.90)
- KernelSHAP-IQ: Weighted Least Square Optimization for Shapley Interactions (ICML 2024 · similarity 0.90)
- SHAP values via sparse Fourier representation (NeurIPS 2025 · 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.