基于双模型的半监督流形混合流量分类方法OA
A semi-supervised manifold mixup traffic classification method based on Mean-Teacher
深度学习技术在网络流量分类领域中得到广泛应用,但存在对大量数据的依赖以及过拟合问题.为解决该问题,提出了一种结合双模型协作与流形混合的半监督深度学习方法.该方法使用教师-学生架构,通过移动指数平均辅助模型学习过程,从而提升模型的泛化性能,并于模型的特征空间中进行数据的流形混合,能够有效改善模型的决策边界,进一步增强模型的鲁棒性.实验结果表明,在不同数据类别,数据量为 1 000的条件下,方法在三种网络流量数据集上都能达到90%以上的准确率,并在更少量数据的条件下保持较高的分类精度.
Deep Learning techniques have been widely applied in the field of network traffic classification.However,there still exist various challenges,including dependency on large scale data and overfitting.To address these issues,a semi-supervised deep learning method combi-ning mean teacher and manifold mixup is proposed.This method employs a teacher-student architecture,utilizing Exponential Moving Average(EMA)to assist the model learning process and to enhance the generalization capability of model.Additionally,manifold mixup in the feature space effectively refines the model's decision boundary,strengthening robustness.Experimental results demonstrate that with only 1 000 sam-ples per class,the method achieves over 90%accuracy across three network traffic datasets while maintaining outstanding performance under few-shot condition.
马可;何明枢;蔡晶晶;王小娟
北京邮电大学 电子工程学院,北京 100876北京邮电大学 电子工程学院,北京 100876永信至诚科技集团股份有限公司,北京 100089北京邮电大学 电子工程学院,北京 100876
信息技术与安全科学
流量分类半监督学习流形混合教师-学生模型
traffic classificationsemi-supervised learningmanifold mixupteacher-student model
《网络安全与数据治理》 2026 (1)
1-8,8
国家自然科学基金(62402053,62227805)中央高校基本科研业务费专项资金(2025KYQD17(BUPT))
评论