Real-Time Explanations for Tabular Foundation Models

TL;DR

ShapPFN is a novel foundation model that provides real-time explanations for tabular data by integrating Shapley value regression, achieving high fidelity and speed over traditional methods.

Contribution

It introduces ShapPFN, a foundation model that combines prediction and explanation in one forward pass, enabling instant interpretability for tabular models.

Findings

ShapPFN achieves $R^2$=0.96 and cosine=0.99 in explanation fidelity.

It is over 1000 times faster than KernelSHAP, taking only 0.06 seconds.

The model maintains competitive predictive performance on standard benchmarks.

Abstract

Interpretability is central for scientific machine learning, as understanding \emph{why} models make predictions enables hypothesis generation and validation. While tabular foundation models show strong performance, existing explanation methods like SHAP are computationally expensive, limiting interactive exploration. We introduce ShapPFN, a foundation model that integrates Shapley value regression directly into its architecture, producing both predictions and explanations in a single forward pass. On standard benchmarks, ShapPFN achieves competitive performance while producing high-fidelity explanations ($R^2$=0.96, cosine=0.99) over 1000\times faster than KernelSHAP (0.06s vs 610s). Our code is available at https://github.com/kunumi/ShapPFN