PolySHAP: Extending KernelSHAP with Interaction-Informed Polynomial Regression
#Shapley values #KernelSHAP #PolySHAP #Polynomial regression #Feature interactions #XAI #Model explanations
📌 Key Takeaways
- Shapley values are central to explainable AI but are computationally expensive to compute exactly. KernelSHAP approximates Shapley values through a linear surrogate model. PolySHAP extends KernelSHAP by fitting a polynomial regression that encodes interaction terms. The new approach aims to improve explanation accuracy while keeping evaluation costs low. The method is proposed in an arXiv preprint (2601.18608v2).
📖 Full Retelling
On the arXiv preprint *PolySHAP: Extending KernelSHAP with Interaction‑Informed Polynomial Regression*, a group of researchers present a novel approach to computing explanations for machine‑learning models. The paper introduces PolySHAP, a method that builds on Lundberg and Lee’s KernelSHAP algorithm but incorporates polynomial‑regression terms to capture feature interactions. The authors argue that by modelling the SHAP game more richly than a simple linear function, PolySHAP can deliver more accurate explanations while still requiring only a modest number of model evaluations.
Key aspects of the method and its context are explored in the abstract and the subsequent sections, though the full technical details are available only in the complete manuscript (arXiv:2601.18608v2). The motivation is to address the computational barrier inherent in exact Shapley value calculation, which scales exponentially with the number of features, by providing a flexible approximation framework that remains tractable.
**Key Points**
🏷️ Themes
Explainable AI, Game‑theoretic feature attribution, Machine‑learning interpretability, Computational efficiency, Feature interactions
Entity Intersection Graph
No entity connections available yet for this article.
Original Source
arXiv:2601.18608v2 Announce Type: replace
Abstract: Shapley values have emerged as a central game-theoretic tool in explainable AI (XAI). However, computing Shapley values exactly requires $2^d$ game evaluations for a model with $d$ features. Lundberg and Lee's KernelSHAP algorithm has emerged as a leading method for avoiding this exponential cost. KernelSHAP approximates Shapley values by approximating the game as a linear function, which is fit using a small number of game evaluations for ran
Read full article at source