Reproduce “TOHAN: A One-step Approach towards Few-shot Hypothesis Adaptation” 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
Haoang Chi, Feng Liu, Wenjing Yang, Long Lan, Tongliang Liu, Bo Han, William K. Cheung, James T. Kwok
TOHAN adapts classifiers with few labeled target examples without accessing source images. It generates class-specific intermediate examples from noise, guided by the source classifier and target examples, then jointly trains target features and classifier. On MNIST-to-USPS with one labeled target example per class, it reports 87.7% accuracy versus SHOT’s 87.2%, a setting-specific result that does not guarantee privacy.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.