arXiv 2303.172352024-02-07cs.LG

Kaizen：实用的自监督持续学习与持续微调

Kaizen: Practical Self-supervised Continual Learning with Continual Fine-tuning

Nokia Bell Labs（诺基亚贝尔实验室）；University of Cambridge（剑桥大学）

Chi Ian Tang, Lorena Qendro, Dimitris Spathis, Fahim Kawsar, Cecilia Mascolo, Akhil Mathur

备注：: Presented at IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) 2024. The code for this work is available at https://github.com/dr-bell/kaizen
发表信息：: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2024, pp. 2841-2850

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

Kaizen提出一种实用的自监督持续学习框架，通过精心设计的损失函数缓解特征提取器和分类器的灾难性遗忘，在分割CIFAR-100上准确率提升16.5%，实现端到端持续学习。

摘要：

自监督学习（SSL）在离线训练的计算机视觉任务中表现出卓越的性能。然而，在持续学习（CL）场景中，当新数据逐步引入时，模型仍然遭受灾难性遗忘。从头开始重新训练模型以适应新生成的数据既耗时又低效。先前的方法建议重新利用自监督目标与知识蒸馏来缓解跨任务的遗忘，但假设在微调期间所有任务的标签都是可用的。在本文中，我们将自监督持续学习推广到一个实际场景，其中可用的标签可以在SSL过程的任何步骤中被利用。随着持续任务数量的增加，这为预训练和微调阶段提供了更大的灵活性。通过Kaizen，我们引入了一种训练架构，该架构通过精心设计的损失函数能够缓解特征提取器和分类器的灾难性遗忘。通过使用一组反映持续学习不同方面的综合评估指标，我们证明了Kaizen在竞争性视觉基准上显著优于先前的SSL模型，在分割CIFAR-100上准确率提升高达16.5%。Kaizen能够通过端到端模型平衡知识保留和学习新数据之间的权衡，为持续学习系统的实际部署铺平了道路。

英文摘要：

Self-supervised learning (SSL) has shown remarkable performance in computer vision tasks when trained offline. However, in a Continual Learning (CL) scenario where new data is introduced progressively, models still suffer from catastrophic forgetting. Retraining a model from scratch to adapt to newly generated data is time-consuming and inefficient. Previous approaches suggested re-purposing self-supervised objectives with knowledge distillation to mitigate forgetting across tasks, assuming that labels from all tasks are available during fine-tuning. In this paper, we generalize self-supervised continual learning in a practical setting where available labels can be leveraged in any step of the SSL process. With an increasing number of continual tasks, this offers more flexibility in the pre-training and fine-tuning phases. With Kaizen, we introduce a training architecture that is able to mitigate catastrophic forgetting for both the feature extractor and classifier with a carefully designed loss function. By using a set of comprehensive evaluation metrics reflecting different aspects of continual learning, we demonstrated that Kaizen significantly outperforms previous SSL models in competitive vision benchmarks, with up to 16.5% accuracy improvement on split CIFAR-100. Kaizen is able to balance the trade-off between knowledge retention and learning from new data with an end-to-end model, paving the way for practical deployment of continual learning systems.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。