Online-LoRA: Task-free Online Continual Learning via Low Rank Adaptation

TL;DR

Online-LoRA introduces a real-time, task-free continual learning framework that fine-tunes pre-trained Vision Transformers using low-rank adaptation and dynamic regularization to mitigate catastrophic forgetting without task boundaries.

Contribution

It proposes a novel online weight regularization strategy and leverages loss dynamics to automatically detect data distribution shifts in continual learning.

Findings

Outperforms state-of-the-art methods across multiple datasets.

Effectively adapts to various ViT architectures.

Robustly handles non-stationary data streams without task boundaries.

Abstract

Catastrophic forgetting is a significant challenge in online continual learning (OCL), especially for non-stationary data streams that do not have well-defined task boundaries. This challenge is exacerbated by the memory constraints and privacy concerns inherent in rehearsal buffers. To tackle catastrophic forgetting, in this paper, we introduce Online-LoRA, a novel framework for task-free OCL. Online-LoRA allows to finetune pre-trained Vision Transformer (ViT) models in real-time to address the limitations of rehearsal buffers and leverage pre-trained models' performance benefits. As the main contribution, our approach features a novel online weight regularization strategy to identify and consolidate important model parameters. Moreover, Online-LoRA leverages the training dynamics of loss values to enable the automatic recognition of the data distribution shifts. Extensive experiments across many task-free OCL scenarios and benchmark datasets (including CIFAR-100, ImageNet-R, ImageNet-S, CUB-200 and CORe50) demonstrate that Online-LoRA can be robustly adapted to various ViT architectures, while achieving better performance compared to SOTA methods. Our code will be publicly available at: https://github.com/Christina200/Online-LoRA-official.git.