On the optimality of the Hedge algorithm in the stochastic regime
This work addresses the theoretical understanding of online learning algorithms for researchers, showing that a simple, non-adaptive method can be optimal, which is incremental but clarifies prior assumptions about algorithm adaptability.
The paper tackles the problem of analyzing the Hedge algorithm in online stochastic settings, proving that anytime Hedge with decreasing learning rate achieves optimal regret guarantees in both worst-case and stochastic scenarios, matching the performance of newer adaptive algorithms.
In this paper, we study the behavior of the Hedge algorithm in the online stochastic setting. We prove that anytime Hedge with decreasing learning rate, which is one of the simplest algorithm for the problem of prediction with expert advice, is surprisingly both worst-case optimal and adaptive to the easier stochastic and adversarial with a gap problems. This shows that, in spite of its small, non-adaptive learning rate, Hedge possesses the same optimal regret guarantee in the stochastic case as recently introduced adaptive algorithms. Moreover, our analysis exhibits qualitative differences with other variants of the Hedge algorithm, such as the fixed-horizon version (with constant learning rate) and the one based on the so-called "doubling trick", both of which fail to adapt to the easier stochastic setting. Finally, we discuss the limitations of anytime Hedge and the improvements provided by second-order regret bounds in the stochastic case.