{"title":"Redirected transfer learning for robust multi-layer subspace learning","authors":"","doi":"10.1007/s10044-024-01233-8","DOIUrl":null,"url":null,"abstract":"<h3>Abstract</h3> <p>Unsupervised transfer learning methods usually exploit the labeled source data to learn a classifier for unlabeled target data with a different but related distribution. However, most of the existing transfer learning methods leverage 0-1 matrix as labels which greatly narrows the flexibility of transfer learning. Another major limitation is that these methods are influenced by the redundant features and noises residing in cross-domain data. To cope with these two issues simultaneously, this paper proposes a redirected transfer learning (RTL) approach for unsupervised transfer learning with a multi-layer subspace learning structure. Specifically, in the first layer, we first learn a robust subspace where data from different domains can be well interlaced. This is made by reconstructing each target sample with the lowest-rank representation of source samples. Besides, imposing the <span> <span>\\(L_{2,1}\\)</span> </span>-norm sparsity on the regression term and regularization term brings robustness against noise and works for selecting informative features, respectively. In the second layer, we further introduce a redirected label strategy in which the strict binary labels are relaxed into continuous values for each datum. To handle effectively unknown labels of the target domain, we construct the pseudo-labels iteratively for unlabeled target samples to improve the discriminative ability in classification. The superiority of our method in classification tasks is confirmed on several cross-domain datasets.</p>","PeriodicalId":54639,"journal":{"name":"Pattern Analysis and Applications","volume":"50 1","pages":""},"PeriodicalIF":3.7000,"publicationDate":"2024-02-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Pattern Analysis and Applications","FirstCategoryId":"94","ListUrlMain":"https://doi.org/10.1007/s10044-024-01233-8","RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE","Score":null,"Total":0}
引用次数: 0
Abstract
Unsupervised transfer learning methods usually exploit the labeled source data to learn a classifier for unlabeled target data with a different but related distribution. However, most of the existing transfer learning methods leverage 0-1 matrix as labels which greatly narrows the flexibility of transfer learning. Another major limitation is that these methods are influenced by the redundant features and noises residing in cross-domain data. To cope with these two issues simultaneously, this paper proposes a redirected transfer learning (RTL) approach for unsupervised transfer learning with a multi-layer subspace learning structure. Specifically, in the first layer, we first learn a robust subspace where data from different domains can be well interlaced. This is made by reconstructing each target sample with the lowest-rank representation of source samples. Besides, imposing the \(L_{2,1}\)-norm sparsity on the regression term and regularization term brings robustness against noise and works for selecting informative features, respectively. In the second layer, we further introduce a redirected label strategy in which the strict binary labels are relaxed into continuous values for each datum. To handle effectively unknown labels of the target domain, we construct the pseudo-labels iteratively for unlabeled target samples to improve the discriminative ability in classification. The superiority of our method in classification tasks is confirmed on several cross-domain datasets.
期刊介绍：
The journal publishes high quality articles in areas of fundamental research in intelligent pattern analysis and applications in computer science and engineering. It aims to provide a forum for original research which describes novel pattern analysis techniques and industrial applications of the current technology. In addition, the journal will also publish articles on pattern analysis applications in medical imaging. The journal solicits articles that detail new technology and methods for pattern recognition and analysis in applied domains including, but not limited to, computer vision and image processing, speech analysis, robotics, multimedia, document analysis, character recognition, knowledge engineering for pattern recognition, fractal analysis, and intelligent control. The journal publishes articles on the use of advanced pattern recognition and analysis methods including statistical techniques, neural networks, genetic algorithms, fuzzy pattern recognition, machine learning, and hardware implementations which are either relevant to the development of pattern analysis as a research area or detail novel pattern analysis applications. Papers proposing new classifier systems or their development, pattern analysis systems for real-time applications, fuzzy and temporal pattern recognition and uncertainty management in applied pattern recognition are particularly solicited.
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号
