Reproduce “Conic-Lifting for Learning-rate-free Convex 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
Julien Grand-Clément, Joon Kwon, Francesco Orabona
The authors use conic lifting for online convex optimization, adding one scalar coordinate per decision and running regret minimization on the resulting cone. Orthogonal lifted feedback lets the analysis optimize comparator scale rather than tune a learning rate. For bounded Hilbert spaces, the regret bound is 2R times the square root of the squared-gradient sum, with leading constant 2 versus AdaGrad-Norm’s 2√2; a known radius is required.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.