While biological intelligence grows organically as new knowledge is gathered throughout life, Artificial Neural Networks forget catastrophically whenever they face a changing training data distribution. Rehearsal-based Continual Learning (CL) approaches have been established as a versatile and reliable solution to overcome this limitation; however, sudden input disruptions and memory constraints are known to alter the consistency of their predictions. We study this phenomenon by investigating the geometric characteristics of the learner's latent space and find that replayed data points of different classes increasingly mix up, interfering with classification. Hence, we propose a geometric regularizer that enforces weak requirements on the Laplacian spectrum of the latent space, promoting a partitioning behavior. We show that our proposal, called Continual Spectral Regularizer (CaSpeR), can be easily combined with any rehearsal-based CL approach and improves the performance of SOTA methods on standard benchmarks. Finally, we conduct additional analysis to provide insights into CaSpeR's effects and applicability.

翻译：尽管生物智能能够在整个生命周期中积累新知识并实现有机增长，人工神经网络在面临训练数据分布变化时仍会遭遇灾难性遗忘。基于重演的持续学习方法已被证明是克服这一限制的通用可靠方案，然而输入分布的突变和存储约束会破坏其预测的一致性。我们通过研究学习者潜在空间的几何特性发现，不同类别的重演数据点会逐渐混合，干扰分类过程。为此，我们提出一种几何正则化方法，通过对潜在空间拉普拉斯谱施加弱约束来促进有效的分区行为。实验表明，我们提出的持续光谱正则化器（CaSpeR）可轻松集成到任何基于重演的持续学习框架中，显著提升现有最优方法在标准基准上的性能。最后，我们通过进一步分析揭示了CaSpeR的作用机理与适用范围。