Reproduce “Gated Target Propagation for Compositional Generalization in Continual Learning” 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
Abdel Mfougouon Njupoun, Colin Bredenberg, Blake Aaron Richards, Guillaume Lajoie
GaTaP tests whether continual learning can retain earlier tasks and recombine learned components for new class pairings. It uses shared network weights and infers sparse, task-specific gains from labeled support examples; evaluation keeps those weights frozen. On MNIST, it retains 97.8% average accuracy on seen tasks and reaches 95.6% query accuracy on unseen combinations, in controlled visual-classification experiments.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.