Elamur Research⌕
☰
home / papers / 2026 / AFT-Tab: Adversarial Fine-Tuning for Tabular Data Synthesis with Long Text Columns

AFT-Tab: Adversarial Fine-Tuning for Tabular Data Synthesis with Long Text Columns

Paper · 2026 · Yuhao Zhang, Liang Yan, Shaoming Duan et al.

Summary

Many real tables mix numeric and categorical fields with lengthy free-text fields such as reviews or project descriptions. Synthetic data generators usually ignore this. Pure LLM generators keep text plausible but distort numeric distributions. Hybrid pipelines, where a GAN or diffusion model produces structured fields and an LLM then writes the text, lose the link between text and structured values. The authors build what they call the first benchmark for this setting, using three public datasets: Cloth, Kickstarter and Wine. They also introduce TCCF, a metric that checks whether text–attribute associations in synthetic data match those in real data. Their framework, AFT-Tab, trains two generators together against an LLM discriminator:

  • an LLM text generator, refined with PPO rewards from the discriminator;
  • a deep generative model for non-text columns, updated through experience replay of high-quality samples.

With Qwen2.5-7B as the backbone, AFT-Tab outperforms CLLM, GReaT and the CTGAN/TabDDPM+LLM hybrids on statistical fidelity, TCCF, diversity and downstream utility. This makes it a practical option for privacy-preserving sharing or augmentation of text-heavy tables.

Key points

  • Benchmark datasets (each has 3 text columns):
    • Cloth: 18,788 train / 4,698 test rows, regression scored by R²
    • Kickstarter: 86,502 / 21,626, binary classification scored by AUC
    • Wine: 84,123 / 21,031, multi-class classification scored by accuracy
  • Downstream utility (Train-on-Synthetic, Test-on-Real, all-text setting):
    • Wine accuracy: AFT-Tab 67.4 vs. 40.9 for the best baseline (82.4 with real data)
    • Cloth R²: 61.0 vs. 54.4
    • Kickstarter AUC: 66.3 vs. 59.1
  • Baseline failure modes: on Cloth, several baselines (e.g., GReaT, CTGAN+LLM) produce negative R² scores. In the t-SNE plots, CLLM shows mode collapse and GReaT produces diffuse, poorly structured samples.
  • Ablation on Cloth:
    • Full model: WD 4.112, TCCF 0.9116
    • Without PPO (SFT only): TCCF drops to 0.8932
    • Without experience replay: WD worsens to 6.109

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Harbin Institute of TechnologyHarbin · China185 works · 29,140 citations🔬Peng Cheng LaboratoryShenzhen · China47 works · 2,699 citations

Also on the byline: Inspur.

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksComputer Vision and Pattern Recognition · 3,486 works

Cite

Cite as Yuhao Zhang, Liang Yan, Shaoming Duan, Xinyu Zha, Jinhang Su, Peiyi Han and Chuanyi Liu. 2026. AFT-Tab: Adversarial Fine-Tuning for Tabular Data Synthesis with Long Text Columns. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.acl-long.209

@inproceedings{zhang2026aft,
  title = {{AFT-Tab: Adversarial Fine-Tuning for Tabular Data Synthesis with Long Text Columns}},
  author = {Yuhao Zhang, Liang Yan, Shaoming Duan, Xinyu Zha, Jinhang Su, Peiyi Han and Chuanyi Liu},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-long.209},
  url = {https://doi.org/10.18653/v1/2026.acl-long.209}
}

References

47 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Abien Fred Agarap. 2018. Statistical analysis on e-commerce reviews, with sentiment classification using bidirectional recurrent neural network (rnn). arXiv preprint arXiv:1805.03687. — [arXiv:1805.03687](https://arxiv.org/abs/1805.03687) find
  2. Abdallah Alshantti, Damiano Varagnolo, Adil Rasheed, Aria Rahmati, and Frank Westad. 2024. Castgan: Cascaded generative adversarial network for realistic tabular data synthesis. IEEE Access, 12:13213– 13232. find
  3. Martin Arjovsky, Soumith Chintala, and Léon Bottou. 2017. Wasserstein generative adversarial networks. In International conference on machine learning, pages 214–223, Sydney, Australia. PMLR. find
  4. Karim Armanious, Chenming Jiang, Marc Fischer, Thomas Küstner, Tobias Hepp, Konstantin Nikolaou, Sergios Gatidis, and Bin Yang. 2020. Medgan: Medical image translation using gans. Computerized medical imaging and graphics, 79:101684. find
  5. Nicolas Audebert, Catherine Herold, Kuider Slimani, and Cédric Vidal. 2019. Multimodal deep networks for text and image-based document classification. In Joint European conference on machine learning and knowledge discovery in databases, pages 427–443. Springer. find
  6. Vadim Borisov, Kathrin Sessler, Tobias Leemann, Martin Pawelczyk, and Gjergji Kasneci. 2023a. Language models are realistic tabular data generators. In The Eleventh International Conference on Learning Representations, Kigali, Rwanda. find
  7. Vadim Borisov, Kathrin Seßler, Tobias Leemann, Martin Pawelczyk, and Gjergji Kasneci. 2023b. Language models are realistic tabular data generators. In ICLR, Kigali, Rwanda. find
  8. Hanqun Cao, Cheng Tan, Zhangyang Gao, Yilun Xu, Guangyong Chen, Pheng-Ann Heng, and Stan Z Li. 2024. A survey on generative diffusion models. IEEE transactions on knowledge and data engineering, 36(7):2814–2830. find
  9. Wenhu Chen, Jianshu Chen, Yu Su, Zhiyu Chen, and William Yang Wang. 2020. Logical natural language generation from open-domain tables. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 7929–7942.
  10. Lingxi Cui, Huan Li, Ke Chen, Lidan Shou, and Gang Chen. 2024. Tabular data augmentation for machine learning: Progress and prospects of embracing generative ai. arXiv preprint arXiv:2407.21523. — [arXiv:2407.21523](https://arxiv.org/abs/2407.21523) find
  11. Liancheng Fang, Aiwei Liu, Hengrui Zhang, Henry Peng Zou, Weizhi Zhang, and Philip S Yu. 2025. Tabgen-icl: Residual-aware in-context example selection for tabular data generation. arXiv preprint arXiv:2502.16414. find
  12. Xi Fang, Weijie Xu, Fiona Anting Tan, Jiani Zhang, Ziqing Hu, Yanjun Qi, Scott Nickleach, Diego Socol-insky, Srinivasan Sengamedu, and Christos Faloutsos. 2024. Large language models (llms) on tabular data: Prediction, generation, and understanding–a survey. arXiv preprint arXiv:2402.17944. — [arXiv:2402.17944](https://arxiv.org/abs/2402.17944) find
  13. Mohamed Gueye, Yazid Attabi, and Maxime Dumas. 2023. RC-TGAN: Row conditional-tabular gan for generating synthetic relational databases. In Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP 2023), Rhodes, Greece. find
  14. Manbir Gulati and Paul Roysdon. 2023. Tabmt: Generating tabular data with masked transformers. In Advances in Neural Information Processing Systems, volume 36, pages 46245–46254, New Orleans, USA. find
  15. Stefan Hegselmann, Alejandro Buendia, Hunter Lang, Monica Agrawal, Xiaoyi Jiang, and David Sontag. 2023. Tabllm: Few-shot classification of tabular data with large language models. In International conference on artificial intelligence and statistics, pages 5549–5581. PMLR. find
  16. Xiaofeng Lin, Chenheng Xu, Matthew Yang, and Guang Cheng. 2024. Ctsyn: A foundational model for cross tabular data generation. arXiv preprint arXiv:2406.04619. — [arXiv:2406.04619](https://arxiv.org/abs/2406.04619) find
  17. Tongyu Liu, Ju Fan, Nan Tang, Guoliang Li, and Xiaoyong Du. 2024. Controllable tabular data synthesis using diffusion models. Proceedings of the ACM on Management of Data, 2(1):1–29. find
  18. María Luisa Menéndez, Julio Angel Pardo, Leandro Pardo, and María del C Pardo. 1997. The jensen-shannon divergence. Journal of the Franklin Institute, 334(2):307–318. find
  19. Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J Liu. 2020. Exploring the limits of transfer learning with a unified text-to-text transformer. Journal of machine learning research, 21(140):1–67.
  20. Amirarsalan Rajabi and Ozlem Ozmen Garibay. 2022. Tabfairgan: Fair tabular data generation with generative adversarial networks. Machine Learning and Knowledge Extraction (MAKE), 4(2):488–501. find
  21. Mikel Hernandez, Gorka Epelde, Ane Alberdi, Rodrigo Cilla, and Debbie Rankin. 2022. Synthetic data generation for tabular health records: A systematic review. Neurocomputing, 493:28–45. find
  22. Eduardo Reis, Mohamed Abdelaal, and Carsten Binnig. 2024. Generalizable data cleaning of tabular data in latent space. Proceedings of the VLDB Endowment, 17(13):4786–4798. find
  23. Ronghang Hu and Amanpreet Singh. 2021. Unit: Multimodal multitask learning with a unified transformer. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 1439– 1449. find
  24. Nabeel Seedat, Nicolas Huynh, Boris Van Breugel, and Mihaela Van Der Schaar. 2024. Curated llm: synergy of llms and data curation for tabular augmentation in low-data regimes. In Proceedings of the 41st International Conference on Machine Learning, pages 44060–44092, Vienna, Austria. find
  25. Madelon Hulsebos, Çagatay Demiralp, and Paul Groth. 2023. GitTables: A large-scale corpus of relational tables. Proceedings of the ACM on Management of Data, 1(1):1–17. find
  26. Lasse Skovgaard Jensen and Ali Gürcan Özkil. 2018. Identifying challenges in crowdfunded product development: a review of kickstarter projects. Design Science, 4:e18. find
  27. Jinhee Kim, Taesung Kim, and Jaegul Choo. 2024. Epic: Effective prompting for imbalanced-class data synthesis in tabular data classification via large language models. Advances in Neural Information Processing Systems, 37:31504–31542. find
  28. Akim Kotelnikov, Dmitry Baranchuk, Ivan Rubachev, and Artem Babenko. 2023. Tabddpm: Modelling tabular data with diffusion models. In International conference on machine learning, pages 17564–17579, Hawaii, USA. PMLR. find
  29. Chaejeong Lee, Jayoung Kim, and Noseong Park. 2023. Codi: Co-evolving contrastive diffusion models for mixed-type tabular synthesis. In International Conference on Machine Learning, pages 18940–18956, Hawaii, USA. PMLR. find
  30. Juntong Shi, Minkai Xu, Harper Hua, Hengrui Zhang, Stefano Ermon, and Jure Leskovec. 2024. Tabdiff: a unified diffusion model for multi-modal tabular data generation. In NeurIPS 2024 Third Table Representation Learning Workshop, Vancouver, Canada. find
  31. Xingjian Shi, Jonas Mueller, Nick Erickson, Mu Li, and Alexander J Smola. 2021. Benchmarking multimodal automl for tabular data with text fields. arXiv preprint arXiv:2111.02705. — [arXiv:2111.02705](https://arxiv.org/abs/2111.02705) find
  32. Aivin V Solatorio and Olivier Dupriez. 2023. Realtabformer: Generating realistic relational and tabular data using transformers. arXiv preprint arXiv:2302.02041. — [arXiv:2302.02041](https://arxiv.org/abs/2302.02041) find
  33. Yuan Sui, Mengyu Zhou, Mingjie Zhou, Shi Han, and Dongmei Zhang. 2024. Table meets llm: Can large language models understand structured table data? a benchmark and empirical study. In Proceedings of the 17th ACM International Conference on Web Search and Data Mining, pages 645–654.
  34. Yuxin Wang, Duanyu Feng, Yongfu Dai, Zhengyu Chen, Jimin Huang, Sophia Ananiadou, Qianqian Xie, and Hao Wang. 2024. Harmonic: Harnessing llms for tabular data synthesis and privacy protection. find
  35. In Advances in Neural Information Processing Systems, volume 37, pages 100196–100212, Vancouver, Canada. find
  36. Xiaofeng Wu, Alan Ritter, and Wei Xu. 2025. Tabular data understanding with LLMs: A survey of recent advances and challenges. arXiv preprint arXiv:2508.00217. — [arXiv:2508.00217](https://arxiv.org/abs/2508.00217) find
  37. Lei Xu, Maria Skoularidou, Alfredo Cuesta-Infante, and Kalyan Veeramachaneni. 2019. Modeling tabular data using conditional gan. Advances in neural information processing systems, 32. find
  38. Lei Xu and Kalyan Veeramachaneni. 2018. Synthesizing tabular data using generative adversarial networks. arXiv preprint arXiv:1811.11264. — [arXiv:1811.11264](https://arxiv.org/abs/1811.11264) find
  39. June Yong Yang, Geondo Park, Joowon Kim, Hyeongwon Jang, and Eunho Yang. 2024a. Language-interfaced tabular oversampling via progressive imputation and self-authentication. In The Twelfth International Conference on Learning Representations, Vienna, Austria. find
  40. Ling Yang, Zhilong Zhang, Yang Song, Shenda Hong, Runsheng Xu, Yue Zhao, Wentao Zhang, Bin Cui, and Ming-Hsuan Yang. 2023. Diffusion models: A comprehensive survey of methods and applications. ACM computing surveys, 56(4):1–39. find
  41. Shuo Yang, Chenchen Yuan, Yao Rong, Felix Stein-bauer, and Gjergji Kasneci. 2024b. P-ta: Using proximal policy optimization to enhance tabular data augmentation via large language models. In Findings of the Association for Computational Linguistics ACL, pages 248–264, Bangkok, Thailand. find
  42. Haowei Zhang, Shengyun Si, Yilun Zhao, and 1 others. 2024. OpenT2T: An open-source toolkit for table-to-text generation. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing: System Demonstrations, pages 259–269. find
  43. Zilong Zhao, Robert Birke, and Lydia Y. Chen. 2022. Fct-gan: Enhancing table synthesis via fourier transform. arXiv preprint arXiv:2210.06239. ArXiv:2210.06239. — [arXiv:2210.06239](https://arxiv.org/abs/2210.06239) find
  44. Zilong Zhao, Robert Birke, and Lydia Y. Chen. 2025. Tabula: Harnessing language models for tabular data synthesis. In Pacific-Asia Conference on Knowledge Discovery and Data Mining, pages 247–259, Sydney, Australia. Springer Nature Singapore. find
  45. Zilong Zhao, Aditya Kunar, Robert Birke, and Lydia Y. Chen. 2021. CTAB-GAN: Effective table data synthesizing. In Proceedings of the 13th Asian Conference on Machine Learning (ACML 2021), volume 157 of Proceedings of Machine Learning Research, pages 97–112, Virtual. find
  46. Yaowei Zheng, Richong Zhang, Junhao Zhang, Yanhan Ye, Zheyan Luo, Zhangchi Feng, and Yongqiang Ma. 2024. Llamafactory: Unified efficient fine-tuning of 100+ language models. arXiv preprint arXiv:2403.13372.
  47. Zynicide. 2017. Wine reviews. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zhang-2026-aft-tab-adversarial-fine-tuning/