Conformalized Online Learning

Adaptive conformal inference is a highly flexible framework for constructing uncertainty sets with a valid coverage guarantee in an online setting, in which the underlying data distribution can drastically -- and even adversarially -- shift over time. In this report, we propose an instantiation of ACI that can be easily integrated with any online learning algorithm, requiring minimal implementation effort and computational cost. Additionally, we provide approaches for constructing intervals that quickly adapt to new changes in the distribution. Using synthetic and real-world benchmark data sets, we demonstrate the improved performance of our proposal over existing techniques.