Reproduce “From Global to Tail Convergence: Dynamics of $α$-Stable Score-Based Generative 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
Ianis Hammani, Claire Boyer
The authors analyze continuous alpha-stable score-based reverse dynamics with a Wasserstein error bound. They show initialization error decays exponentially for both stable and Brownian dynamics, while exact-score stable outputs inherit tail index alpha and Brownian outputs remain light-tailed. Against power-law targets, extreme-quantile scaling depends on matching tail indices; matching tail constants is also needed for a limit of one.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.