基于模糊规则学习的无监督异构领域自适应  

Unsupervised Heterogeneous Domain Adaptation with Fuzzy Rule Learning

作　　者：孙武 邓赵红 娄琼丹 顾鑫[4] 王士同 SUN Wu;DENG Zhaohong;LOU Qiongdan;GU Xin;WANG Shitong(School of Artificial Intelligence and Computer Science,Jiangnan University,Wuxi,Jiangsu 214122,China;Key Laboratory of Computational Neuroscience and Brain-Inspired Intelligence,Ministry of Education,Fudan University,Shanghai 200433,China;Zhangjiang Lab,Shanghai 200120,China;Jiangsu North Huguang Opto-Electronics Co.,Ltd.,Wuxi,Jiangsu 214000,China)

机构地区：[1]江南大学人工智能与计算机学院,江苏无锡214122 [2]复旦大学计算神经科学与类脑智能教育部重点实验室,上海200433 [3]张江实验室,上海200120 [4]江苏北方湖光光电有限公司,江苏无锡214000

出　　处：《计算机科学与探索》2022年第2期403-412,共10页Journal of Frontiers of Computer Science and Technology

基　　金：国家自然科学基金面上项目(61772239);上海市“脑与类脑智能基础转化应用研究”市级重大科技专项(2018SHZDZX01)。

摘　　要：异构领域自适应是一种借助源域知识为语义相关但特征空间不同的目标域建模的技术。现有的异构领域自适应方法大多属于半监督方法,这些方法要求目标域中存在一部分已标记样本,然而这种数据集在很多异构领域自适应任务中是稀缺的。为了解决上述问题,提出了一种新的基于模糊规则学习的无监督异构领域自适应算法。一方面,该方法基于TSK模糊系统的规则学习分别对源域和目标域进行特征学习,通过学习两个特征变换矩阵将源域和目标域投影到一个公共特征子空间;另一方面,为了减少因特征变换所造成的信息损失,该算法采取了多种信息保持策略,并且最大化公共特征子空间中源域数据和目标域数据之间的相关性。通过在几个真实领域自适应数据集上进行实验,验证了所提算法相对于现有的异构领域自适应方法具有一定的优越性。Heterogeneous domain adaptation is a technique that uses the knowledge of source domain to model the target domain.The source domain and the target domain are semantically related,but their feature spaces are different.Among existing heterogeneous domain adaptive methods,most of them belong to semi-supervised methods,which require some labeled samples in the target domain.However,this kind of dataset is rare in many heterogeneous adaptive tasks.In order to solve the above problem,this paper proposes a new unsupervised heterogeneous domain adaptive algorithm based on fuzzy rule learning.On the one hand,by introducing the TSK fuzzy system,the proposed method learns two feature transformation matrices corresponding to the source domain and the target domain respectively.By learning two feature transformation matrices,the source domain and the target domain are projected into a common feature subspace.On the other,in order to reduce the information loss caused by feature transformation,the proposed algorithm adopts a variety of information preservation strategies and maximizes the correlation between the transformed source domain data and target domain data.Through experiments on domain adaptive datasets,the results show that the proposed algorithm has certain advantages over the existing heterogeneous domain adaptive methods.

关 键 词：模糊规则学习 TSK模糊系统 信息保持 异构领域自适应 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]