Latent Space Dimensionality Impact on Adversarial Transferability in VAE and Diffusion Tabular Generators

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 dimensionality of the latent space in VAE-based tabular generators affect the transferability of adversarial attacks compared to diffusion models when evaluated on synthetic data robustness benchmarks?

Autonomous synthesis report generated by SOVEREIGN Research Kernel. Tribunal consensus score: 8.2/10.