Online Adaptation to Label Distribution Shift
This addresses the issue of distribution shifts in real-world deployments for machine learning practitioners, though it is incremental as it builds on classical online learning techniques.
The paper tackles the problem of online adaptation to label distribution shift, where models must adjust to changing test-time label distributions without true labels, and shows that OGD is effective and robust in various challenging scenarios.
Machine learning models often encounter distribution shifts when deployed in the real world. In this paper, we focus on adaptation to label distribution shift in the online setting, where the test-time label distribution is continually changing and the model must dynamically adapt to it without observing the true label. Leveraging a novel analysis, we show that the lack of true label does not hinder estimation of the expected test loss, which enables the reduction of online label shift adaptation to conventional online learning. Informed by this observation, we propose adaptation algorithms inspired by classical online learning techniques such as Follow The Leader (FTL) and Online Gradient Descent (OGD) and derive their regret bounds. We empirically verify our findings under both simulated and real world label distribution shifts and show that OGD is particularly effective and robust to a variety of challenging label shift scenarios.