Reproduce “Weighting Schedules Govern What and When Score-Based Generative Models Learn from Multimodal Data” 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
Jérémie Klinger, Raphaël Urfin, Giulio Biroli, Marylou Gabrié
The authors analyze how noise-level weighting shapes feature learning in score-based generative models. Their high-dimensional Gaussian-mixture theory shows that high signal-to-noise ratios let models learn mode directions without relative weights, while near speciation they learn weights too and weaker directions take longer. Tinted image experiments qualitatively support selective timing; the exact dynamics apply to Gaussian mixtures and tailored score models.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.