ForestDiffusion: XGBoost-based tabular data diffusion model favors CPU parallelization

jm_alexia · x · 2026-08-30

The author discusses ForestDiffusion, a tool for generating and imputing tabular data using diffusion models and XGBoost. Key insight: because diffusion models require many trees, parallelizing with CPUs is often faster than using GPUs. The author tested a single-tree multiple-output approach but found it slightly slower and less effective.

Original post →

More from Research

Research channel →