Reproduce “Robust Flow Matching under Target Corruption and Label Noise” 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
Mert Can Kurucu, Erik Englesson, Hossein Azizpour
The authors develop flow-matching methods for image generation with corrupted targets or noisy labels. Temporal Residual Score downweights examples using late-time prediction errors, while Comparative Conditional Score compares class-conditioned errors on the same trajectory. On CIFAR-10, these methods improve reported generation metrics in tested settings, though label-noise class alignment improves alongside worse unconditional FID; neither signal is universally reliable.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.