Online Learning with Optimism and Delay

TL;DR

This paper introduces new optimistic online learning algorithms designed for real-time climate forecasting that handle delayed feedback without parameter tuning and achieve optimal regret guarantees.

Contribution

It presents a novel reduction of delayed online learning to optimistic learning, along with new algorithms and analysis that improve robustness and performance in delayed feedback scenarios.

Findings

Algorithms DORM, DORM+, and AdaHedgeD perform well on climate forecasting tasks.

The proposed methods achieve low regret compared to existing models.

Delay can be effectively mitigated using optimism-based strategies.

Abstract

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.