Reproduce “Annealed Lévy Diffusion via Gaussian Surrogate Reverse Dynamics” 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
Yuling Jiao, Lican Kang, Jerry Zhijian Yang, Cheng Yuan, Ruizhe Zhang
ALD schedules diffusion noise from heavy-tailed to Gaussian for exploration and refinement. At each time, it represents stable noise as Gaussian noise multiplied by a random scalar, allowing Gaussian noise prediction conditioned on that scale. On a clipped Holtsmark benchmark, ALD-posterior reaches sliced Wasserstein-2 distance 0.4701, below DDPM’s 0.4919 and DLPM’s 1.4924; this result is specific to that constructed benchmark.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.