Computational study of co-coercivity extensions for online convex optimization
Smoothing nonsmooth losses via Moreau envelopes to restore the inequality.
Quadratic and logistic losses satisfy the inequality perfectly (0% violations), confirming theoretical predictions.
Hinge and absolute value losses violate the inequality at non-differentiable points where subgradient selection matters.
Moreau envelope smoothing restores the inequality for all nonsmooth losses with appropriate smoothing parameter.
Despite inequality violations, OCO gradient norm regret remains controlled, suggesting weaker conditions may suffice.