Reproduce “The Role of Global Labels in Few-Shot Classification and How to Infer Them” 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
Ruohan Wang, Massimiliano Pontil, Carlo Ciliberto
Meta Label Learning (MeLa) addresses few-shot image classification when tasks have local labels without a shared class map. It learns an initial image representation, then uses within-task labels to infer shared clusters for pretraining. On tieredImageNet, MeLa reached 96.4% cluster accuracy and 84.1% 5-shot accuracy, versus 28.2% and 78.8% for K-means; results cover benchmarks with disjoint classes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.