Amortized Linear-time Exact Shapley Value for Product-Kernel Methods
Researchers have introduced PKeX-Shapley, a novel algorithm designed to compute exact Shapley values for product-kernel methods in machine learning. While kernel methods are powerful, their black-box nature often limits use in high-stakes applications due to the intractability of exact explainability computations. Existing solutions like SHAP rely on approximations that introduce estimation errors. PKeX-Shapley exploits the multiplicative structure of product kernels, using a distribution-free removal operator that replaces feature kernel factors with the multiplicative identity. This approach creates a parameter-free value function requiring no sampling or density estimation. The algorithm computes exact Shapley values for all features in quadratic time relative to the number of features, achieving amortized linear time per feature through shared recursive formulations. This ensures numerical stability and efficiency. Furthermore, the framework extends beyond predictive modeling to kernel-based discrepancies like Maximum Mean Discrepancy (MMD) and Hilbert-Schmidt Independence Criterion (HSIC), offering new tools for interpretable statistical analysis and enhancing the transparency of complex machine learning models.
Wire timeline
Amortized Linear-time Exact Shapley Value for Product-Kernel Methods
Researchers have introduced PKeX-Shapley, a novel algorithm designed to compute exact Shapley values for product-kernel methods in machine learning. While kernel methods are powerful, their black-box nature often limits use in high-stakes applications due to the intractability of exact explainability computations. Existing solutions like SHAP rely on approximations that introduce estimation errors. PKeX-Shapley exploits the multiplicative structure of product kernels, using a distribution-free removal operator that replaces feature kernel factors with the multiplicative identity. This approach creates a parameter-free value function requiring no sampling or density estimation. The algorithm computes exact Shapley values for all features in quadratic time relative to the number of features, achieving amortized linear time per feature through shared recursive formulations. This ensures numerical stability and efficiency. Furthermore, the framework extends beyond predictive modeling to kernel-based discrepancies like Maximum Mean Discrepancy (MMD) and Hilbert-Schmidt Independence Criterion (HSIC), offering new tools for interpretable statistical analysis and enhancing the transparency of complex machine learning models.
cs.AI updates on arXiv.org