Comparing DDPM and Conditional Flow Matching with identical U-Net backbones on MNIST.
-
Updated
May 9, 2026 - Jupyter Notebook
Comparing DDPM and Conditional Flow Matching with identical U-Net backbones on MNIST.
Source code for the BRACIS 2026 project: [ From Bias to Fairness: Can Synthetic Data Bridge the Gap in Tabular Machine Learning? ]
To associate your repository with the difussion-model topic, visit your repo's landing page and select "manage topics."