Enhancing Visual Continual Learning with Language-Guided Supervision

arXiv:2403.16124v1 Announce Type: new
Abstract: Continual learning (CL) aims to empower models to learn new tasks without forgetting previously acquired knowledge. Most prior works concentrate on the techniques of architectures, replay data, regularization, \etc. However, the category name of each class is largely neglected. Existing methods commonly utilize the one-hot labels and randomly initialize the classifier head. We argue that the scarce semantic information conveyed by the one-hot labels hampers the effective knowledge transfer across tasks. In this paper, we …

abstract acquired architectures arxiv class continual cs.cv data etc hot however knowledge labels language learn prior regularization supervision tasks type visual