Experimental Evidence for Asymptotic Non-Optimality of Comb Adversary Strategy

TL;DR

This paper provides computational evidence that the comb strategy for five experts in adversarial prediction is not asymptotically optimal, challenging previous conjectures in the field.

Contribution

It offers the first strong computational evidence against the conjecture that the comb strategy is asymptotically optimal for five experts.

Findings

The comb strategy is not asymptotically optimal for k=5 experts.

Provides computational evidence challenging existing conjectures.

Supports the need for alternative strategies in adversarial prediction.

Abstract

For the problem of prediction with expert advice in the adversarial setting with finite stopping time, we give strong computer evidence that the comb strategy for $k=5$ experts is not asymptotically optimal, thereby giving strong evidence against a conjecture of Gravin, Peres, and Sivan.