A Unified Approach to Interpreting Model Predictions.
Explainable AI
Surrogate Explainers
Paper
Description
Lundberg and Lee (2017) propose SHAP as a provably unified approach to additive feature attribution methods (including LIME) with certain desiderata. Contrary to LIME, this approach involves permuting through the feature space and checking how different features impact model predictions when they are included in the permutations.
References
Lundberg, Scott M, and Su-In Lee. 2017. “A Unified Approach to Interpreting Model Predictions.” In Proceedings of the 31st International Conference on Neural Information Processing Systems, 4768–77.