Can we globally optimize cross-validation loss? Quasiconvexity in ridge regression

Open in new window