Towards a General Framework for Continual Learning with Pre-training
Liyuan Wang, Jingyi Xie, Xingxing Zhang, Hang Su, Jun Zhu
TL;DR
The paper addresses continual learning in the context of pre-trained backbones by introducing a hierarchical objective decomposed into within-task prediction, task-identity inference, and task-adaptive prediction. It couples this framework with parameter-efficient fine-tuning and representation-statistics preservation to optimize all three components explicitly, supported by theoretical loss-bounding insights. Empirically, it demonstrates superior performance on downstream continual learning benchmarks and shows meaningful gains in upstream few-shot adaptation, outperforming existing prompt-based methods. The work also links the approach to neuroscience, offering a potential bridge between biological memory mechanisms and open-world knowledge acquisition for lifelong AI systems.
Abstract
In this work, we present a general framework for continual learning of sequentially arrived tasks with the use of pre-training, which has emerged as a promising direction for artificial intelligence systems to accommodate real-world dynamics. From a theoretical perspective, we decompose its objective into three hierarchical components, including within-task prediction, task-identity inference, and task-adaptive prediction. Then we propose an innovative approach to explicitly optimize these components with parameter-efficient fine-tuning (PEFT) techniques and representation statistics. We empirically demonstrate the superiority and generality of our approach in downstream continual learning, and further explore the applicability of PEFT techniques in upstream continual learning. We also discuss the biological basis of the proposed framework with recent advances in neuroscience.