Bottleneck in Tabular Data Generation? Study Shows Probabilistic Circuits Beat Diffusion Models

tetraduzione · x · 2026-08-12

For tabular data generation, a recent study points out that while diffusion-based models appear to be state-of-the-art, this is largely due to flawed evaluation metrics. In contrast, probabilistic circuits—the generative counterpart of decision trees—can deliver comparable or superior performance at a fraction of the cost. The author emphasizes that there is still a significant gap to fill for realistic tabular data generation, even for foundation models.

Original post →

More from Research

Research channel →