Reproduce “Towards General Preference Alignment: Diffusion Models at Nash Equilibrium” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jiaming Hu, Jiamu Bai, Haoyu Wang, Debarghya Mukherjee, Ioannis Ch. Paschalidis
Diff.-NPO trains image models through comparisons among evolving outputs, anchored to pretrained models. It forms training pairs by ranking generated images with five automatic scoring models. On SDXL and Parti-Prompts, it achieves a 77.02% average scorer-based win rate against original SDXL, versus 72.01% for Online-DPO. CLIP scores are an exception, and the results do not establish human preference cycles.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.