Reproduce “Learning Energy Networks with Generalized Fenchel-Young Losses” 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
Mathieu Blondel, Felipe Llinares-López, Robert Dadashi, Léonard Hussenot, Matthieu Geist
The generalized Fenchel–Young loss extends regularized prediction losses to general energy functions and uses the envelope theorem to compute gradients without differentiating through the output optimizer. It is nonnegative and zero at the label under uniqueness conditions. Pairwise models improve returns on three of four locomotion tasks and accuracy on four of six multilabel datasets, but results depend on task.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.