Reproduce “Persistently Trained, Diffusion-assisted Energy-based Models” 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
Xinwei Zhang, Zhiqiang Tan, Zhijian Ou
DA-EBM trains a persistent energy-based model jointly on images and progressively noised versions, using a sampler that moves across noise levels. On grayscale image tasks, it supports stable sampling from real data and generation from noise. For FashionMNIST versus MNIST, it achieves an AUROC of 0.93 versus 0.83 for persistent EBM, but complex images and general OOD detection remain untested.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.