Score-Based Generative Modeling through Reverse Diffusion Monte Carlo

by Ethan Vertal, Mudita Sai, and Supriya Shingade

This was our final project for Generative AI at CMU. We implemented a new way of looking at the reverse diffusion process from Huang et. al., 2024. To our knowledge, no one had implemented this before... so we had the great idea that it would be fun to try. It turns out that implementing a new algorithm from just the paper comes with more challenges than we possibly could have accounted for. That being said, it was fun working with these two and I thought I'd make this work publicly available because I'm somewhat proud of how we presented it. Our results were below what we were hoping for but Google Colab is not exactly friendly for 12+ hour coding sessions. Enjoy!