Continual learning (CL) algorithms strive to acquire new knowledge while preserving prior information. However, this stability-plasticity trade-off remains a central challenge. This paper introduces a framework that dissects this trade-off, offering valuable insights into CL algorithms. The Readout-Decomposition of Activation Change (RDAC) framework first addresses the stability-plasticity dilemma and its relation to catastrophic forgetting. It relates learning-induced activation changes in the range of prior readouts to the degree of stability and changes in the null space to the degree of plasticity. In deep non-linear networks tackling split-CIFAR-110 tasks, the framework clarifies the stability-plasticity trade-offs of the popular regularization algorithms Synaptic intelligence (SI), Elastic-weight consolidation (EWC), and learning without Forgetting (LwF), and replay-based algorithms Gradient episodic memory (GEM), and data replay. GEM and data replay preserved stability and plasticity, while SI, EWC, and LwF traded off plasticity for stability. The inability of the regularization algorithms to maintain plasticity was linked to them restricting the change of activations in the null space of the prior readout. Additionally, for one-hidden-layer linear neural networks, we derived a gradient decomposition algorithm to restrict activation change only in the range of the prior readouts, to maintain high stability while not further sacrificing plasticity. Results demonstrate that the algorithm maintained stability without significant plasticity loss. The RDAC framework informs the behavior of existing CL algorithms and paves the way for novel CL approaches. Finally, it sheds light on the connection between learning-induced activation/representation changes and the stability-plasticity dilemma, also offering insights into representational drift in biological systems.

翻译：持续学习（CL）算法致力于在保留已有知识的同时获取新信息。然而，稳定性与可塑性的权衡始终是该领域的核心挑战。本文提出一个剖析该权衡的框架，为CL算法提供了深刻洞见。激活变化的读出分解（RDAC）框架首先阐释了稳定性-可塑性困境及其与灾难性遗忘的关系。该框架将先前读出值范围内由学习引发的激活变化与稳定性程度相关联，将零空间中的变化与可塑性程度相关联。在应对split-CIFAR-110任务的深度非线性网络中，该框架阐明了主流正则化算法——突触智能（SI）、弹性权重巩固（EWC）和遗忘学习（LwF）——以及基于回放的算法——梯度情景记忆（GEM）和数据回放——的稳定性-可塑性权衡。GEM和数据回放保持了稳定性与可塑性，而SI、EWC和LwF则以牺牲可塑性换取稳定性。正则化算法无法维持可塑性的原因，在于它们限制了对先前读出值零空间中激活变化的约束。此外，对于单隐层线性神经网络，我们推导出一种梯度分解算法，仅将激活变化限制在先前读出值的范围内，从而在保持高稳定性的同时不进一步牺牲可塑性。结果表明，该算法在几乎不损失可塑性的情况下保持了稳定性。RDAC框架揭示了现有CL算法的行为规律，并开辟了新型CL方法的研究路径。最后，该框架阐明了学习引发的激活/表征变化与稳定性-可塑性困境之间的关联，也为生物系统中的表征漂移提供了见解。