Reproduce “Semi-supervised Sequential Generative 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
Michael Teng, Tuan Anh Le, Adam Scibior, Frank Wood
CWS trains generative sequence models with sparse labels by combining observed and sampled unlabeled labels in one importance-weighted inference-network update. It uses a biased gradient estimator; SSWS retains separate supervised and unsupervised terms. On handwriting data, both methods generate phrase-like samples at 12.5% supervision. CWS outperforms a fully supervised baseline trained on partial data without the remaining unlabeled examples.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.