Verified SHAP: Provable Bounds for Exact Shapley Values of Neural Networks
For practitioners needing exact SHAP explanations, this work provides a first scalable approach, though it is an incremental step towards full exact computation.
The paper introduces an algorithm that uses neural network verification to compute provable bounds on exact SHAP values, scaling to search spaces orders of magnitude larger than prior exact methods.
Shapley additive explanations (SHAP) are widely recognised as computationally intractable for neural networks, since they induce an exponential search space over the input features. In this work, we take a first step towards scaling exact SHAP computation to larger search spaces by introducing an algorithm that leverages recent advances in neural network verification to compute arbitrarily tight exact lower and upper bounds on SHAP values for neural networks, ultimately recovering the exact SHAP values. We demonstrate that our approach scales to orders of magnitude larger search spaces than state-of-the-art exact methods. This provides an important first step towards exact SHAP computation and establishes a principled cornerstone for evaluating statistical approximation methods on larger search spaces.