Performance comparison of TFMs pretrained on mixed synthetic-real versus real data in TabMNAR benchmark with varying missingness

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 the performance of TFMs pretrained on mixed synthetic-real data compare to those pretrained on real data when evaluated on the TabMNAR benchmark with varying levels of random missing data (e.g., 10%, 50%), and does the gap persist across different missingness mechanisms?

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