Reproduce “Benign population landscapes do not transfer to empirical risk minimization in nonsmooth 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
Penghao Wang, Mateo Díaz, Liwei Jiang
The authors analyze the empirical landscape of robust phase retrieval with noiseless Gaussian measurements. Although the population loss has only the true signals as local minima, the nonsmooth empirical loss can have sharp spurious minima near its saddle sphere. For each fixed dimension d≥4, the asymptotic probability is at least 0.136; this is not a joint-growth result.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.