CeCR: Cross-entropy contrastive replay for online class-incremental continual learning

Download PDF
Open Webpage

Guanglu Sun, Baolun Ji, Lili Liang, Minghui Chen

Published: 01 Jan 2024, Last Modified: 08 Apr 2025Neural Networks 2024EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•Samples with balanced class distribution are selected to join the replay training.•In memory update, samples with core knowledge are preserved in the replay set.•The CeCL is used to achieve effective learning with obtaining solid knowledge.