# Self-Improving Tabular Language Models via Iterative Reward-Guided Post-Training

arXiv: https://arxiv.org/abs/2604.18966 · HTML: https://mlsanity.com/paper/2604.18966/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in LLM preference optimization methods, one of 10 directions in LLM alignment / RLHF / preference optimisation.
- Builds on 21 of its 23 references that are in the map.
- First posted on arXiv 21 Apr 2026.

## Placed in

- LLM alignment / RLHF / preference optimisation / [LLM preference optimization methods](https://mlsanity.com/alignment/llm-preference-optimization-methods/)
- Synthetic data and self-improvement for LLMs / [Privacy-preserving tabular data synthesis](https://mlsanity.com/synth/privacy-preserving-tabular-data-synthesis/)

## Builds on (most cited first, 21 of 21)

- [Proximal Policy Optimization Algorithms](https://mlsanity.com/paper/1707.06347/) (2017)
- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [XGBoost: A Scalable Tree Boosting System](https://mlsanity.com/paper/1603.02754/) (2016)
- [Scaling Laws for Reward Model Overoptimization](https://mlsanity.com/paper/2210.10760/) (2022)
- [Training Diffusion Models with Reinforcement Learning](https://mlsanity.com/paper/2305.13301/) (2023)
- [Diffusion Model Alignment Using Direct Preference Optimization](https://mlsanity.com/paper/2311.12908/) (2023)
- [Modeling Tabular data using Conditional GAN](https://mlsanity.com/paper/1907.00503/) (2019)
- [Negative Preference Optimization: From Catastrophic Collapse to Effective Unlearning](https://mlsanity.com/paper/2404.05868/) (2024)
- [TabDDPM: Modelling Tabular Data with Diffusion Models](https://mlsanity.com/paper/2209.15421/) (2022)
- [Language Models are Realistic Tabular Data Generators](https://mlsanity.com/paper/2210.06280/) (2022)
- [Revisiting Classifier Two-Sample Tests](https://mlsanity.com/paper/1610.06545/) (2016)
- [How Faithful is your Synthetic Data? Sample-level Metrics for Evaluating and Auditing Generative Models](https://mlsanity.com/paper/2102.08921/) (2021)
- [Mixed-Type Tabular Data Synthesis with Score-based Diffusion in Latent Space](https://mlsanity.com/paper/2310.09656/) (2023)
- [Fine-Tuning of Continuous-Time Diffusion Models as Entropy-Regularized Control](https://mlsanity.com/paper/2402.15194/) (2024)
- [TabDiff: a Mixed-type Diffusion Model for Tabular Data Generation](https://mlsanity.com/paper/2410.20626/) (2024)
- [Goodhart's Law in Reinforcement Learning](https://mlsanity.com/paper/2310.09144/) (2023)
- [SynthEval: A Framework for Detailed Utility and Privacy Evaluation of Tabular Synthetic Data](https://mlsanity.com/paper/2404.15821/) (2024)
- [The Inadequacy of Similarity-based Privacy Metrics: Privacy Attacks against "Truly Anonymous" Synthetic Datasets](https://mlsanity.com/paper/2312.05114/) (2023)
- [Avoiding mode collapse in diffusion models fine-tuned with reinforcement learning](https://mlsanity.com/paper/2410.08315/) (2024)
- [Preference Alignment on Diffusion Model: A Comprehensive Survey for Image Generation and Editing](https://mlsanity.com/paper/2502.07829/) (2025)

## Abstract (arXiv)

Tabular language models can generate synthetic tables by modeling rows as token sequences, but they are typically trained once with supervised fine-tuning and then used as static synthesizers. This is limiting because next-token likelihood does not directly optimize the distributional, utility, and indistinguishability properties used to evaluate synthetic data. We study iterative reward-guided post-training for tabular language models through a generate--score--align protocol, where a generator samples synthetic rows, a task-specified reward ranks them, and the model is updated relative to a fixed supervised reference. Within this protocol, we propose TabGRAA (Tabular Group-Relative Advantage Alignment), a group-relative alignment method that compares high- and low-reward generated groups using group-averaged policy/reference log-ratios rather than one-to-one preference pairs. Across five mixed-type benchmarks, TabGRAA improves a GReaT backbone beyond additional supervised fine-tuning and achieves the strongest average trade-off among adapted DPO, KTO, and NPO baselines on fidelity and downstream utility, while maintaining empirical privacy diagnostics near the supervised baseline. Ablations show that the gains depend on meaningful reward ranking and stable group-level updates rather than extra training alone. Reward-substitution and scorer-separation studies further show that the post-training loop can use both classifier-based and classifier-free rewards, and that proper scorer separation is important for preserving the fidelity--utility--privacy trade-off. These results position TabGRAA as a self-improving post-training method for tabular language-model generators, complementary to strong static tabular synthesizers.