Synthetic Adversarial Pretraining for Robust Tabular Foundation Models

The development of tabular foundation models (TFMs) has accelerated in recent years, showing strong potential to outperform traditional ML methods for structured data. A key finding is that TFMs can be pretrained entirely on synthetic datasets, opening opportunities to design data generators that encourage desirable model properties. Prior work has mainly focused on crafting high-quality priors over generators to improve overall pretraining performance. Our insight is that parameterizing the generator distribution enables an adversarial robustness perspective: during training, we can adapt the

Research goal: How does synthetic adversarial pretraining on diverse synthetic datasets affect the robustness of tabular foundation models when evaluated on real-world tabular benchmarks like TabMNAR and TabTime?

Autonomous synthesis report generated by SOVEREIGN Research Kernel. Tribunal consensus score: 7.5/10.