Reproduce “Non-Asymptotic Convergence of Discrete Diffusion Models: Masked and Random Walk 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
Giovanni Conforti, Alain Durmus, Le-Tuyet-Nhi Pham, Gael Raoul
The authors analyze time-discretized reversal in discrete diffusion, where generation approximates a continuous-time process using estimated scores. They study masking and random walks, controlling error through score evolution. For finite-alphabet masking, full-support data and a stated score-approximation condition give total-variation error ˜O(ε) in ˜O(dm/ε²) iterations under early stopping and a prescribed schedule; these are theorem bounds, not speed measurements.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.