ReTabSyn: Realistic Tabular Data Synthesis via Reinforcement Learning

Xiaofeng Lin, Seungbae Kim, Zhuoya Li, Zachary Desoto, Charles Fleming, Guang Cheng
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:74460-74478, 2026.

Abstract

Deep generative models can help with data scarcity and privacy by producing synthetic training data, but they struggle in low-data, imbalanced tabular settings to fully learn the complex data distribution. We argue that striving for the full joint distribution could be overkill; for greater data efficiency, models should prioritize learning the conditional distribution $P(y\mid \mathbf{X})$, as suggested by recent theoretical analysis. Therefore, we overcome this limitation with ReTabSyn, a Reinforced Tabular Synthesis pipeline that provides direct feedback on feature correlation preservation during synthesizer training. This objective encourages the generator to prioritize the most useful predictive signals when training data is limited, thereby strengthening downstream model utility. We empirically fine-tune a language model-based generator using this approach, and across benchmarks with small sample sizes, class imbalance, and distribution shift, ReTabSyn consistently outperforms state-of-the-art baselines. Moreover, our approach can be readily extended to control various aspects of synthetic tabular data, such as applying expert-specified constraints on generated observations.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-lin26p, title = {{R}e{T}ab{S}yn: Realistic Tabular Data Synthesis via Reinforcement Learning}, author = {Lin, Xiaofeng and Kim, Seungbae and Li, Zhuoya and Desoto, Zachary and Fleming, Charles and Cheng, Guang}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {74460--74478}, 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/lin26p/lin26p.pdf}, url = {https://proceedings.mlr.press/v306/lin26p.html}, abstract = {Deep generative models can help with data scarcity and privacy by producing synthetic training data, but they struggle in low-data, imbalanced tabular settings to fully learn the complex data distribution. We argue that striving for the full joint distribution could be overkill; for greater data efficiency, models should prioritize learning the conditional distribution $P(y\mid \mathbf{X})$, as suggested by recent theoretical analysis. Therefore, we overcome this limitation with ReTabSyn, a Reinforced Tabular Synthesis pipeline that provides direct feedback on feature correlation preservation during synthesizer training. This objective encourages the generator to prioritize the most useful predictive signals when training data is limited, thereby strengthening downstream model utility. We empirically fine-tune a language model-based generator using this approach, and across benchmarks with small sample sizes, class imbalance, and distribution shift, ReTabSyn consistently outperforms state-of-the-art baselines. Moreover, our approach can be readily extended to control various aspects of synthetic tabular data, such as applying expert-specified constraints on generated observations.} }
Endnote
%0 Conference Paper %T ReTabSyn: Realistic Tabular Data Synthesis via Reinforcement Learning %A Xiaofeng Lin %A Seungbae Kim %A Zhuoya Li %A Zachary Desoto %A Charles Fleming %A Guang Cheng %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-lin26p %I PMLR %P 74460--74478 %U https://proceedings.mlr.press/v306/lin26p.html %V 306 %X Deep generative models can help with data scarcity and privacy by producing synthetic training data, but they struggle in low-data, imbalanced tabular settings to fully learn the complex data distribution. We argue that striving for the full joint distribution could be overkill; for greater data efficiency, models should prioritize learning the conditional distribution $P(y\mid \mathbf{X})$, as suggested by recent theoretical analysis. Therefore, we overcome this limitation with ReTabSyn, a Reinforced Tabular Synthesis pipeline that provides direct feedback on feature correlation preservation during synthesizer training. This objective encourages the generator to prioritize the most useful predictive signals when training data is limited, thereby strengthening downstream model utility. We empirically fine-tune a language model-based generator using this approach, and across benchmarks with small sample sizes, class imbalance, and distribution shift, ReTabSyn consistently outperforms state-of-the-art baselines. Moreover, our approach can be readily extended to control various aspects of synthetic tabular data, such as applying expert-specified constraints on generated observations.
APA
Lin, X., Kim, S., Li, Z., Desoto, Z., Fleming, C. & Cheng, G.. (2026). ReTabSyn: Realistic Tabular Data Synthesis via Reinforcement Learning. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:74460-74478 Available from https://proceedings.mlr.press/v306/lin26p.html.

Related Material