Experimental Evidence for Asymptotic Non-Optimality of Comb Adversary Strategy

This addresses a theoretical problem in online learning for researchers, but it is incremental as it focuses on disproving a specific conjecture with computational evidence.

The paper tackles the problem of prediction with expert advice in adversarial settings with finite stopping time, providing strong computer evidence that the comb strategy for k=5 experts is not asymptotically optimal, which contradicts a prior conjecture.