Rethinking Class-Incremental Learning from a Dynamic Imbalanced Learning Perspective

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Wang, Leyuan, Xiang, Liuyu, Wang, Yunlong, Wu, Huijia, He, Zhaofeng
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916258970075136
author Wang, Leyuan
Xiang, Liuyu
Wang, Yunlong
Wu, Huijia
He, Zhaofeng
author_facet Wang, Leyuan
Xiang, Liuyu
Wang, Yunlong
Wu, Huijia
He, Zhaofeng
contents Deep neural networks suffer from catastrophic forgetting when continually learning new concepts. In this paper, we analyze this problem from a data imbalance point of view. We argue that the imbalance between old task and new task data contributes to forgetting of the old tasks. Moreover, the increasing imbalance ratio during incremental learning further aggravates the problem. To address the dynamic imbalance issue, we propose Uniform Prototype Contrastive Learning (UPCL), where uniform and compact features are learned. Specifically, we generate a set of non-learnable uniform prototypes before each task starts. Then we assign these uniform prototypes to each class and guide the feature learning through prototype contrastive learning. We also dynamically adjust the relative margin between old and new classes so that the feature distribution will be maintained balanced and compact. Finally, we demonstrate through extensive experiments that the proposed method achieves state-of-the-art performance on several benchmark datasets including CIFAR100, ImageNet100 and TinyImageNet.
format Preprint
id arxiv_https___arxiv_org_abs_2405_15157
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Rethinking Class-Incremental Learning from a Dynamic Imbalanced Learning Perspective
Wang, Leyuan
Xiang, Liuyu
Wang, Yunlong
Wu, Huijia
He, Zhaofeng
Computer Vision and Pattern Recognition
Deep neural networks suffer from catastrophic forgetting when continually learning new concepts. In this paper, we analyze this problem from a data imbalance point of view. We argue that the imbalance between old task and new task data contributes to forgetting of the old tasks. Moreover, the increasing imbalance ratio during incremental learning further aggravates the problem. To address the dynamic imbalance issue, we propose Uniform Prototype Contrastive Learning (UPCL), where uniform and compact features are learned. Specifically, we generate a set of non-learnable uniform prototypes before each task starts. Then we assign these uniform prototypes to each class and guide the feature learning through prototype contrastive learning. We also dynamically adjust the relative margin between old and new classes so that the feature distribution will be maintained balanced and compact. Finally, we demonstrate through extensive experiments that the proposed method achieves state-of-the-art performance on several benchmark datasets including CIFAR100, ImageNet100 and TinyImageNet.
title Rethinking Class-Incremental Learning from a Dynamic Imbalanced Learning Perspective
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2405.15157