Reproduce “Posterior Flow Matching” 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
Chenrui Ma, Xi Xiao, Lin Zhao, Xinru Jiang, Tianyang Wang, Yanning Shen, Ferdinando Fioretto
PFM predicts conditional means and componentwise variances of flow-matching velocity targets with a model trained using Gaussian negative log-likelihood. It converts these moments into approximate endpoint-posterior readouts and uses covariance to set noise in a stochastic sampler. On class-conditioned ImageNet-256, the sampler improves FID from 24.45 to 22.28 versus PFM ODE, but the covariance comparison does not match injected noise energy.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.