Reproduce “Diversity-enhancing Generative Network for 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
Ruijiang Dong, Feng Liu, Haoang Chi, Tongliang Liu, Mingming Gong, Gang Niu, Masashi Sugiyama, Bo Han
DEG-Net adapts a source classifier to a domain using few labeled target examples and generated data, without source data. It trains a class-conditional generator to preserve class meaning and target relevance while penalizing dependence among generated examples’ semantic features with HSIC. On MNIST→USPS, it reaches 90.1±0.6% accuracy with three examples per class, but does not outperform the no-diversity variant at all target-data quantities.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.