Reproduce “Follow the Regularized Leader Does Not Converge in 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
Ioannis Anagnostides, Ioannis Panageas, Nikolas Patris, Tuomas Sandholm
The paper shows that continuous-time follow the regularized leader (FTRL) with Euclidean regularization can improve an objective without reaching first-order stationarity on a constructed smooth constrained problem. Its cumulative gradients can keep a coordinate fixed even when its instantaneous gradient favors movement. The KKT gap stays positive at every time, but the result concerns a nonanalytic objective, not typical problems.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.