Denoising Diffusion Policy Optimization

Researchers propose iADD, a method for reinforcement-learning post-training of diffusion models that aims to improve alignment with reward functions without sacrificing output diversity, addressing a …