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

Xinpeng Lv, Yunxin Mao, Renzhe Xu, Chunyuan Zheng, Yikai Chen, Haoxuan Li, Jinxuan Yang, Yuanlong Chen, Kun Kuang, Mingyang Geng, Shixuan Liu, Wanrong Huang, Shaowu Yang, Wenjing Yang, Zhouchen Lin, Haotian Wang
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:83603-83630, 2026.

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 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 strategic tabular data. We show that strategic manipulation creates a fundamental mismatch between the non-strategic prior learned during pretraining and the post-manipulation strategic prior encountered at deployment, which leads to an irreducible structural prediction bias. To address this issue, we propose the Strategic Prior-data Fitted Network (SPN), an inference-time strategy-aware framework that adapts tabular foundation models to strategic environments without retraining or architectural modification. SPN constructs strategic in-context examples to approximate post-manipulation inputs and aligns PFN predictions with the induced strategic distribution via in-context learning, with theoretical guarantees on bias reduction. 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.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-lv26g, title = {When Tabular Foundation Models Meet Strategic Tabular Data: A Prior Alignment Approach}, author = {Lv, Xinpeng and Mao, Yunxin and Xu, Renzhe and Zheng, Chunyuan and Chen, Yikai and Li, Haoxuan and Yang, Jinxuan and Chen, Yuanlong and Kuang, Kun and Geng, Mingyang and Liu, Shixuan and Huang, Wanrong and Yang, Shaowu and Yang, Wenjing and Lin, Zhouchen and Wang, Haotian}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {83603--83630}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/lv26g/lv26g.pdf}, url = {https://proceedings.mlr.press/v306/lv26g.html}, 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 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 strategic tabular data. We show that strategic manipulation creates a fundamental mismatch between the non-strategic prior learned during pretraining and the post-manipulation strategic prior encountered at deployment, which leads to an irreducible structural prediction bias. To address this issue, we propose the Strategic Prior-data Fitted Network (SPN), an inference-time strategy-aware framework that adapts tabular foundation models to strategic environments without retraining or architectural modification. SPN constructs strategic in-context examples to approximate post-manipulation inputs and aligns PFN predictions with the induced strategic distribution via in-context learning, with theoretical guarantees on bias reduction. 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.} }
Endnote
%0 Conference Paper %T When Tabular Foundation Models Meet Strategic Tabular Data: A Prior Alignment Approach %A Xinpeng Lv %A Yunxin Mao %A Renzhe Xu %A Chunyuan Zheng %A Yikai Chen %A Haoxuan Li %A Jinxuan Yang %A Yuanlong Chen %A Kun Kuang %A Mingyang Geng %A Shixuan Liu %A Wanrong Huang %A Shaowu Yang %A Wenjing Yang %A Zhouchen Lin %A Haotian Wang %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-lv26g %I PMLR %P 83603--83630 %U https://proceedings.mlr.press/v306/lv26g.html %V 306 %X 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 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 strategic tabular data. We show that strategic manipulation creates a fundamental mismatch between the non-strategic prior learned during pretraining and the post-manipulation strategic prior encountered at deployment, which leads to an irreducible structural prediction bias. To address this issue, we propose the Strategic Prior-data Fitted Network (SPN), an inference-time strategy-aware framework that adapts tabular foundation models to strategic environments without retraining or architectural modification. SPN constructs strategic in-context examples to approximate post-manipulation inputs and aligns PFN predictions with the induced strategic distribution via in-context learning, with theoretical guarantees on bias reduction. 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.
APA
Lv, X., Mao, Y., Xu, R., Zheng, C., Chen, Y., Li, H., Yang, J., Chen, Y., Kuang, K., Geng, M., Liu, S., Huang, W., Yang, S., Yang, W., Lin, Z. & Wang, H.. (2026). When Tabular Foundation Models Meet Strategic Tabular Data: A Prior Alignment Approach. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:83603-83630 Available from https://proceedings.mlr.press/v306/lv26g.html.

Related Material