Training Diffusion Models with Reinforcement Learning

This is a new paper from UC Berkeley. We will review the algorithm, performance and the new Github code.

Paper abstract:
Diffusion models are a class of flexible generative models trained with an approximation to the log-likelihood objective. However, most use cases of diffusion models are not concerned with likelihoods, but instead with downstream objectives such as human-perceived image quality or drug effectiveness. In this paper, we investigate reinforcement learning methods for directly optimizing diffusion models for such objectives. We describe how posing denoising as a multi-step decision-making problem enables a class of policy gradient algorithms, which we refer to as denoising diffusion policy optimization (DDPO), that are more effective than alternative reward-weighted likelihood approaches. Empirically, DDPO is able to adapt text-to-image diffusion models to objectives that are difficult to express via prompting, such as image compressibility, and those derived from human feedback, such as aesthetic quality. Finally, we show that DDPO can improve prompt-image alignment using feedback from a vision-language model without the need for additional data collection or human annotation.

Presenter: Junling Hu

Join this event here: https://us02web.zoom.us/meeting/register/tZMlc--urTMpGteRnTC6x2v5f8eQyoLIcF8J

Related links:
Paper:
Black, Kevin, Michael Janner, Yilun Du, Ilya Kostrikov, and Sergey Levine. "Training diffusion models with reinforcement learning." arXiv:2305.13301 (May 22, 2023). https://arxiv.org/abs/2305.13301

Blog of Carper.ai, Sept 27, 2023, https://carper.ai/enhancing-diffusion-models-with-reinforcement-learning/
Code: https://github.com/carperai/drlx

Agenda:
7pm-7:05pm Meet and Greet
7:05-8:05pm Presentation
8:05-8:30 pm Q&A and Discussions