Reproduce “Sparse Sum-of-Squares Layers for Certified 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
Ege Gursoy, Nicolas Mansard
The Sparse SOS layer trains learned objective coefficients through the scalar value of a fixed sparse sum-of-squares certificate, retaining local solver coordinates for its backward pass. In a design task, it recorded zero violations on training, test, and size-specific splits, including sizes 96 and 160, with 480 solves per seed versus 3,360 for finite differences. The result is limited to tested instances, sizes, and seeds.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.