河南理工大学学报（自然科学版）2026，Vol.45Issue(5)：48-59,12.DOI:10.16186/j.cnki.1673-9787.2026010020
基于增强数据对比的自监督学习综述
Data augmentation-based contrastive self-supervised learning:A review
摘要
Abstract
Significance Traditional supervised learning relies heavily on large-scale manually annotated da-tasets,which incur high labeling costs and introduce potential human bias,thereby limiting model general-ization.Meanwhile,the vast amount of unlabeled data remains underutilized.As a result,self-supervised learning has emerged as a major research focus.Among its paradigms,contrastive learning has become a representative discriminative approach.It constructs positive and negative sample pairs and learns represen-tations by pulling similar instances closer while pushing dissimilar ones apart in the feature space,en-abling models to capture high-level semantic features in a self-supervised manner.Compared with genera-tive methods,contrastive learning demonstrates stronger transferability and lower computational costs,pro-viding a promising direction for universal representation learning.Progress This paper systematically re-views the theoretical framework and core components of contrastive learning,including data augmentation strategies that enforce semantic invariance,sample construction mechanisms that determine feature discrim-inability,Siamese-network-based encoder architectures,and loss functions such as InfoNCE.Representa-tive methods,including MoCo,SimCLR,Barlow Twins,and SimSiam,are further summarized in terms of their technical designs.In addition,emerging applications are discussed such as representation learning un-der long-tailed distributions and multimodal vision-language alignment.key challenges are analyzed in con-trastive learning,including false negative sample issues in large-batch training and modality heterogeneity in cross-modal scenarios.Conclusions and Prospects Contrastive learning effectively alleviates the reliance on large-scale labeled data,providing a solid foundation for developing perception systems in artificial gen-eral intelligence.Further research should focus on designing lightweight and efficient sample matching mechanisms to reduce computational complexity,while also improving theoretical interpretability to better understand feature disentanglement and generalization mechanisms.关键词
对比学习/自监督学习/数据增强/长尾分布/多模态学习Key words
contrastive learning/self-supervised learning/data augmentation/long-tailed distribution/multi-modal learning分类
信息技术与安全科学引用本文复制引用
张嘉睿,曹哲骁,王田,王传云,滕婧,傅瑶..基于增强数据对比的自监督学习综述[J].河南理工大学学报（自然科学版）,2026,45(5):48-59,12.基金项目
国家自然科学基金资助项目(92467108) （92467108）
北京市昌平区科技副区长专项项目(202504005043) （202504005043）
