Schedule-Robust Online Continual Learning
This addresses the challenge of unpredictable performance in real-world scenarios where data schedules are unknown and dynamic, though it is incremental in improving robustness within the online class-incremental setting.
The paper tackles the problem of designing continual learning methods robust against arbitrary data presentation schedules, introducing a novel approach that outperforms existing methods on image classification benchmarks by a large margin.
A continual learning (CL) algorithm learns from a non-stationary data stream. The non-stationarity is modeled by some schedule that determines how data is presented over time. Most current methods make strong assumptions on the schedule and have unpredictable performance when such requirements are not met. A key challenge in CL is thus to design methods robust against arbitrary schedules over the same underlying data, since in real-world scenarios schedules are often unknown and dynamic. In this work, we introduce the notion of schedule-robustness for CL and a novel approach satisfying this desirable property in the challenging online class-incremental setting. We also present a new perspective on CL, as the process of learning a schedule-robust predictor, followed by adapting the predictor using only replay data. Empirically, we demonstrate that our approach outperforms existing methods on CL benchmarks for image classification by a large margin.