Reproduce “Learning to Generate Novel Domains for Domain Generalization” 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

Kaiyang Zhou, Yongxin Yang, Timothy Hospedales, Tao Xiang

L2A-OT addresses domain generalization by training a generator to create pseudo-novel, label-preserving source images, then training the model on original and synthetic data. On Digits-DG, its mean top-1 accuracy across four held-out-domain tests was 78.1%, versus 75.8% for CrossGrad and 73.7% for vanilla, averaged over three seeds. It requires source data and generator training, and does not guarantee unseen-domain performance.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.