Online Learning with Optimism and Delay
This addresses real-time forecasting challenges in climate and weather domains, offering a novel approach to mitigate delay effects.
The paper tackled the problem of online learning with delayed feedback by developing optimistic algorithms (DORM, DORM+, AdaHedgeD) that require no tuning and achieve optimal regret, demonstrating low regret on subseasonal climate forecasting tasks.
Inspired by the demands of real-time climate and weather forecasting, we develop optimistic online learning algorithms that require no parameter tuning and have optimal regret guarantees under delayed feedback. Our algorithms -- DORM, DORM+, and AdaHedgeD -- arise from a novel reduction of delayed online learning to optimistic online learning that reveals how optimistic hints can mitigate the regret penalty caused by delay. We pair this delay-as-optimism perspective with a new analysis of optimistic learning that exposes its robustness to hinting errors and a new meta-algorithm for learning effective hinting strategies in the presence of delay. We conclude by benchmarking our algorithms on four subseasonal climate forecasting tasks, demonstrating low regret relative to state-of-the-art forecasting models.