Continuous Learning in Single-Incremental-Task Scenarios
This work addresses the challenge of preventing forgetting in deep models for incremental learning tasks, which is crucial for applications like online learning, but it appears incremental as it builds on prior strategies.
The paper tackles the problem of continuous learning in single-incremental-task scenarios, such as class-incremental learning, where existing methods like LWF, EWC, and SI are inadequate, and proposes AR1, a new approach combining architectural and regularization strategies that outperforms existing methods on CORe50 and iCIFAR-100 datasets.
It was recently shown that architectural, regularization and rehearsal strategies can be used to train deep models sequentially on a number of disjoint tasks without forgetting previously acquired knowledge. However, these strategies are still unsatisfactory if the tasks are not disjoint but constitute a single incremental task (e.g., class-incremental learning). In this paper we point out the differences between multi-task and single-incremental-task scenarios and show that well-known approaches such as LWF, EWC and SI are not ideal for incremental task scenarios. A new approach, denoted as AR1, combining architectural and regularization strategies is then specifically proposed. AR1 overhead (in term of memory and computation) is very small thus making it suitable for online learning. When tested on CORe50 and iCIFAR-100, AR1 outperformed existing regularization strategies by a good margin.