Reproduce “Randomized and Oracle-Based Feasible Directions Methods for Nonsmooth Nonconvex Constrained Optimization” 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
Dan Greenstein, Nadav Hallak
The authors extend derivative-free feasible-direction optimization to general convex sets using finitely many inward directions and separate random feasible exploration. Exact regularized line searches keep iterates feasible without derivatives or a known objective decomposition. Under smoothness and curvature assumptions, they obtain finite-time approximate stationarity only for directions permitting a specified minimum travel distance; exploration gives almost-sure stationarity at accumulation points.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.