A Unified Framework for Provably Efficient Algorithms to Estimate Shapley Values

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Chen, Tyler, Seshadri, Akshay, Villani, Mattia J., Niroula, Pradeep, Chakrabarti, Shouvanik, Ray, Archan, Deshpande, Pranav, Yalovetzky, Romina, Pistoia, Marco, Kumar, Niraj
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918208511934464
author Chen, Tyler
Seshadri, Akshay
Villani, Mattia J.
Niroula, Pradeep
Chakrabarti, Shouvanik
Ray, Archan
Deshpande, Pranav
Yalovetzky, Romina
Pistoia, Marco
Kumar, Niraj
author_facet Chen, Tyler
Seshadri, Akshay
Villani, Mattia J.
Niroula, Pradeep
Chakrabarti, Shouvanik
Ray, Archan
Deshpande, Pranav
Yalovetzky, Romina
Pistoia, Marco
Kumar, Niraj
contents Shapley values have emerged as a critical tool for explaining which features impact the decisions made by machine learning models. However, computing exact Shapley values is difficult, generally requiring an exponential (in the feature dimension) number of model evaluations. To address this, many model-agnostic randomized estimators have been developed, the most influential and widely used being the KernelSHAP method (Lundberg & Lee, 2017). While related estimators such as unbiased KernelSHAP (Covert & Lee, 2021) and LeverageSHAP (Musco & Witter, 2025) are known to satisfy theoretical guarantees, bounds for KernelSHAP have remained elusive. We describe a broad and unified framework that encompasses KernelSHAP and related estimators constructed using both with and without replacement sampling strategies. We then prove strong non-asymptotic theoretical guarantees that apply to all estimators from our framework. This provides, to the best of our knowledge, the first theoretical guarantees for KernelSHAP and sheds further light on tradeoffs between existing estimators. Through comprehensive benchmarking on small and medium dimensional datasets for Decision-Tree models, we validate our approach against exact Shapley values, consistently achieving low mean squared error with modest sample sizes. Furthermore, we make specific implementation improvements to enable scalability of our methods to high-dimensional datasets. Our methods, tested on datasets such MNIST and CIFAR10, provide consistently better results compared to the KernelSHAP library.
format Preprint
id arxiv_https___arxiv_org_abs_2506_05216
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle A Unified Framework for Provably Efficient Algorithms to Estimate Shapley Values
Chen, Tyler
Seshadri, Akshay
Villani, Mattia J.
Niroula, Pradeep
Chakrabarti, Shouvanik
Ray, Archan
Deshpande, Pranav
Yalovetzky, Romina
Pistoia, Marco
Kumar, Niraj
Machine Learning
Data Structures and Algorithms
Quantum Physics
Shapley values have emerged as a critical tool for explaining which features impact the decisions made by machine learning models. However, computing exact Shapley values is difficult, generally requiring an exponential (in the feature dimension) number of model evaluations. To address this, many model-agnostic randomized estimators have been developed, the most influential and widely used being the KernelSHAP method (Lundberg & Lee, 2017). While related estimators such as unbiased KernelSHAP (Covert & Lee, 2021) and LeverageSHAP (Musco & Witter, 2025) are known to satisfy theoretical guarantees, bounds for KernelSHAP have remained elusive. We describe a broad and unified framework that encompasses KernelSHAP and related estimators constructed using both with and without replacement sampling strategies. We then prove strong non-asymptotic theoretical guarantees that apply to all estimators from our framework. This provides, to the best of our knowledge, the first theoretical guarantees for KernelSHAP and sheds further light on tradeoffs between existing estimators. Through comprehensive benchmarking on small and medium dimensional datasets for Decision-Tree models, we validate our approach against exact Shapley values, consistently achieving low mean squared error with modest sample sizes. Furthermore, we make specific implementation improvements to enable scalability of our methods to high-dimensional datasets. Our methods, tested on datasets such MNIST and CIFAR10, provide consistently better results compared to the KernelSHAP library.
title A Unified Framework for Provably Efficient Algorithms to Estimate Shapley Values
topic Machine Learning
Data Structures and Algorithms
Quantum Physics
url https://arxiv.org/abs/2506.05216