Rainbow Memory: Continual Learning with a Memory of Diverse Samples

TLDR

Continual learning is a realistic AI scenario, yet most studies assume disjoint class sets, which is less realistic. This work investigates a more realistic blurry task boundary where tasks share classes. We propose Rainbow Memory, a memory management strategy that increases sample diversity using per‑sample classification uncertainty and data augmentation. Experiments on MNIST, CIFAR‑10, CIFAR‑100, and ImageNet show that Rainbow Memory significantly boosts accuracy in blurry continual learning, surpassing state‑of‑the‑art methods. Code and data splits are available at https://github.com/clovaai/rainbow-memory.

Abstract

Continual learning is a realistic learning scenario for AI models. Prevalent scenario of continual learning, however, assumes disjoint sets of classes as tasks and is less realistic rather artificial. Instead, we focus on 'blurry' task boundary; where tasks shares classes and is more realistic and practical. To address such task, we argue the importance of diversity of samples in an episodic memory. To enhance the sample diversity in the memory, we propose a novel memory management strategy based on per-sample classification uncertainty and data augmentation, named Rainbow Memory (RM). With extensive empirical validations on MNIST, CIFAR10, CIFAR100, and ImageNet datasets, we show that the proposed method significantly improves the accuracy in blurry continual learning setups, outperforming state of the arts by large margins despite its simplicity. Code and data splits will be available in https://github.com/clovaai/rainbow-memory.