基于动态域定界的循环分类模型  

A Cyclic Classification Model Based on Dynamic Domain Delimitation

作　　者：邓慧娜 叶阿勇[1,2] 张娇美 DENG Huina;YE Ayong;ZHANG Jiaomei(College of Computer and Cyber Security,Fujian Normal University,Fuzhou 350117,China;Fujian Provincial Key Laboratory of Network Security and Cryptology,Fuzhou 350117,China)

机构地区：[1]福建师范大学计算机与网络空间安全学院,福建福州350117 [2]福建省网络安全与密码技术重点实验室,福建福州350117

出　　处：《福建师范大学学报（自然科学版）》2022年第6期33-42,共10页Journal of Fujian Normal University：Natural Science Edition

基　　金：国家自然科学基金资助项目(61972096、61771140、 61872088、 61872090);福建省高校产学合作项目(2022H6025)。

摘　　要：在多源域迁移学习应用中,现有的分类模型大多对数据标签要求较高,难以适应数据集的动态变化和标签缺失情况.针对该问题,提出一种基于动态域定界的循环分类模型CAMDOT(cyclic classification model based on dynamic domain delimitation).首先,引入互信息量化各数据域间的相关性,提高分类模型适应异构用户和数据动态变化的能力.其次,针对数据标签缺失和不平衡问题,提出一种循环分类算法.最后,通过理论与实验分析,验证该模型在多源域数据训练中具有较高的分类准确率.In the application of multi-source domain transfer learning, most of the existing classification models have high requirements for data labels, which is difficult to adapt to the dynamic changes of datasets and the absence of labels. To address this problem, CAMDOT(cyclic classification model based on dynamic domain delimitation) is proposed. Firstly, mutual information is introduced to quantify the correlation between data domains and improve the ability of the classification model to adapt to heterogeneous users and dynamic changes of data. Secondly, a cyclic classification algorithm is proposed for the problem of missing and unbalanced data labels. Finally, through theoretical and experimental analysis, it is verified that the model in this paper has high classification accuracy in multi-source domain data training.

关 键 词：迁移学习 动态分类 数据相关性 卷积神经网络 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]