When Tabular Foundation Models Meet Strategic Tabular Data: A Prior Alignment Approach

2026-05-19

AI Summary

This work addresses a critical limitation of existing Prior-Function Network (PFN)-based tabular foundation models, which perform well on non-strategic data but suffer from deployment-time distribution shifts when individuals strategically manipulate their features, leading to biased predictions. To mitigate this, the authors propose the Strategic Prior Network (SPN) framework—the first approach to explicitly model strategic behavior within tabular foundation models. SPN aligns pretrained models with the strategic data distribution at inference time by generating strategic contextual examples, eliminating the need for retraining. Integrating game-theoretic modeling of strategic responses with the PFN architecture, SPN demonstrates significantly improved robustness and prediction accuracy under strategic manipulation, outperforming current methods on both real-world and synthetic datasets.

Abstract

Tabular foundation models based on pretrained prior-data fitted networks~(PFNs) have shown strong generalization on diverse tabular tasks, but they are typically designed for \emph{non-strategic} settings where data distributions are independent of deployed classifiers. In many real-world decision scenarios, however, individuals may strategically modify their features after deployment to obtain favorable outcomes, inducing a post-deployment distribution shift. This paper studies whether PFN-style tabular foundation models can generalize to such \emph{strategic} tabular data. We show that strategic manipulation creates a mismatch between the non-strategic prior learned during pretraining and the post-manipulation strategic prior, which leads to systematic prediction bias. To address this issue, we propose \textbf{Strategic Prior-data Fitted Network}~\textit{(SPN)}, an inference-time strategy-aware framework that adapts tabular foundation models to strategic environments without retraining. SPN constructs strategic in-context examples to approximate post-manipulation inputs and aligns PFN predictions with the induced strategic distribution. Experiments on real-world and synthetic tabular datasets show that SPN consistently improves robustness and predictive performance under strategic manipulation compared with both tabular foundation models and classical tabular methods.

Problem

• strategic manipulation
• tabular foundation models
• distribution shift
• prior mismatch
• post-deployment adaptation

Innovation

• Strategic Manipulation
• Tabular Foundation Models
• Prior Alignment
• In-context Learning
• Distribution Shift