Reproduce “Exploring Solver-Level Warmstarting for Neural Network Verification” 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
Annelot Bosman, Minghao Liu, Marta Kwiatkowska, Holger Hoos, Jan van Rijn
The authors test solver-level warmstarting for local robustness verification, reusing search trees from mixed-integer linear programming runs. On three small MNIST networks, the mnist-net comparison for certified-robust source and target instances across perturbation radii reduced target runtime from 538 ± 859 to 16 ± 75 seconds (97%). Warmstarting solved 300 of 474 timeouts, but SYMPHONY results do not establish gains for modern verifiers or large models.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.