Reproduce “Neural Simpletrons - Minimalistic Directed Generative Networks for Learning with Few Labels” 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
Dennis Forster, Abdul-Saboor Sheikh, Jörg Lücke
Neural Simpletrons (NeSi) train a classifier from labeled and unlabeled examples with one generative likelihood objective. The model uses confident predictions on unlabeled examples as class-layer training labels. On permutation-invariant MNIST with 100 balanced labels, feedforward self-labeling NeSi reports 4.96 ± 0.08% test error, versus 19.08 ± 0.26% without self-labeling; published hybrid methods report lower errors.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.