Reproduce “Contextual Precision in Tropical Abstract Domains: Sharp Generator Budgets and ReLU Certificates” 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
Nassim Arifette
The paper studies normalized max-plus joins of signed hidden-state summaries, which can introduce states absent from incoming paths. It derives a sharp worst-case observation-loss law for diagonal summaries in the unit cube: 1/[2(M−1)] for dimensions at least two and M≥3. In a ReLU example, four diagonal generators recover the margin, not a performance result for trained networks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.