Real-Time Explanations for Tabular Foundation Models

Luan Borges Teodoro Reis Sena; Francisco Galuppo Azevedo

cs.LGMar 31, 2026

Real-Time Explanations for Tabular Foundation Models

Authors: Luan Borges Teodoro Reis Sena, Francisco Galuppo Azevedo

Organizations: Kunumi Institute, Belo Horizonte, Minas Gerais, Brazil · Universidade Federal de Minas Gerais, Belo Horizonte, Minas Gerais, Brazil

Abstract

Interpretability is central for scientific machine learning, as understanding 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

Figures & tables

Panel of Figure 1 — Figure 1: ShapPFN architecture. The model encodes features and targets into embeddings, processes them through transformer layers with alternating attention over features and datapoints, then decodes into a global baseline and per-feature contributions. The final prediction is explicitly additive over features, enabling SHAP-like decomposition directly from the architecture.

Appendix figures & tables1 asset

Supplementary material from the paper’s appendix.

Appendix

Parameter Optimized	Loss Wt.	Warmup Steps	Num Subsets	Bkg Samples	SHAP Cos	ROC AUC
SHAP Loss Weight	1	2000	4	4	0.979	0.829
	2	2000	4	4	0.977	0.826
	10	2000	4	4	0.979	0.798
	50	2000	4	4	0.890	0.777
Warmup Steps	1	0	4	4	0.977	0.830
	1	400	4	4	0.979	0.828

Appendix

Table 3: Sequential hyperparameter optimization. Each block optimizes one parameter (highlighted in gray) while keeping others fixed. Best ROC AUC in each block shown in bold.

Neural Network Interpretability

Neural network interpretability studies how models represent information and make predictions, and how to explain, test, or control their internal computations.

See all 329 papers

ProToMEx: Rapid, Interpretable Explanations via Structured Representations
Sep 2, 2026Athina Georgara, Adarsh Valoor, Sarvapali D. RamchurnExplainable Artificial Intelligence Interpretable ML
University of Southampton, Hampshire, England, UK
TabSHAP
Apr 22, 2026Aryan Chaudhary, Prateek Agarwal, Tejasvi AlladiFeature Attribution SHAP Feature Attribution
Department of Computer Science Birla Institute of Technology and Science India
TabPFN-3.5: Technical Report
Sep 15, 2026Benjamin Jäger, Nick Erickson, Léo Grinsztajn +44Tabular Foundation Models Inference-Time Optimization

Sep 2, 2026J/K move · Enter open · S save

ProToMEx: Rapid, Interpretable Explanations via Structured Representations

Athina Georgara, Adarsh Valoor, Sarvapali D. Ramchurn

University of Southampton, Hampshire, England, UK

Explainable Artificial Intelligence Interpretable ML

Open paper AskPDF arXiv

Existing post-hoc explainers for machine learning classifiers primarily focus on feature attribution, assigning importance scores to individual features. While valuable, this approach struggles to articulate the complex, combinatorial patterns that often drive a model's decision-making process. To overcome this limitation, we introduce ProToMEx, a new paradigm for explainability that leverages Probabilistic Topic Models (PTMs). Our model-agnostic framework learns latent ''topics'' that represent distinct, high-level reasons for a classification, moving beyond simple feature importance to reveal underlying semantic structures. ProToMEx naturally provides both global explanations of a model's overall behaviour and local explanations that can disentangle multiple co-existing reasons for a specific prediction. We demonstrate empirically that ProToMEx not only produces explanations of comparable fidelity to popular methods like SHAP and LIME but also drastically reduces the amortised computational cost of generating local explanations, making it highly suitable for real-time applications. Specifically, we show that ProToMEx is ~30-40x faster than SHAP and LIME over standardised tabular datasets and synthetic datasets.

Abstract

Neural Network Interpretability

Explore similar work

ProToMEx: Rapid, Interpretable Explanations via Structured Representations

TabSHAP

TabPFN-3.5: Technical Report