Reproduce “Learning Discriminative Geometry for Drifting 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
Doudou Zhang, Wenwen Hou, Yilin Chen, Qi Chen
Drifting Models learn a sample-similarity representation during training to refine the drift used by a one-step image generator. The encoder persists across batches and distinguishes real from generated samples using kernel density estimates, while each batch supplies new density estimates. On SVHN, CIFAR-10, and CIFAR-100, the method reports 82–95% lower FID than pixel-space drifting, but larger-scale performance remains unestablished.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.