Multi-view RBM with posterior consistency and domain adaptation

doi:10.1016/j.ins.2019.12.062

CORC > 计算技术研究所 > 中国科学院计算技术研究所 > 中国科学院计算技术研究所期刊论文 > 英文

	Multi-view RBM with posterior consistency and domain adaptation
	Zhang, Nan 1; Ding, Shifei 1,2; Sun, Tongfeng 1; Liao, Hongmei 1; Wang, Lijuan 1; Shi, Zhongzhi 3
刊名	INFORMATION SCIENCES
	2020-04-01
卷号	516 页码:142-157
关键词	Restricted Boltzmann machines Representational learning Multi-view learning Posterior consistency Domain adaptation
ISSN号	0020-0255
DOI	10.1016/j.ins.2019.12.062
英文摘要	The restricted Boltzmann machine (RBM) and extensions are rarely used in the field of multi-view learning. In this paper, we first present a multi-view RBM model for classification, which is named RBM with posterior consistency (PCRBM). PCRBM computes multiple representations by regularizing the marginal likelihood function with the consistency among representations from multiple views. However, PCRBM ignores the specific information of each view, where the learned representations just contain the consistency information among multiple views. To address this problem, we propose a novel multi-view RBM model and name it as RBM with posterior consistency and domain adaptation (PDRBM). PDRBM divides the hidden units of each view into two groups: the one is the view-consistency group that contains the consistency information among multiple views, and the other is the view-specific group that contains the information unique to this separate view. In addition, PDRBM balances the relationship among view-consistency hidden representations for multi-view classification. Contrasting with existing multi-view classification methods, PDRBM has achieved satisfactory results on two-class and multi-class classification multi-view datasets. (C) 2019 Elsevier Inc. All rights reserved.
资助项目	National Natural Science Foundation of China[61672522] ; National Natural Science Foundation of China[61976216] ; National Natural Science Foundation of China[61379101]
WOS研究方向	Computer Science
语种	英语
出版者	ELSEVIER SCIENCE INC
WOS记录号	WOS:000515432200009
内容类型	期刊论文
源URL	[http://119.78.100.204/handle/2XEOYT63/14432]
专题	中国科学院计算技术研究所期刊论文_英文
通讯作者	Ding, Shifei
作者单位	1.China Univ Min & Technol, Sch Comp Sci & Technol, Xuzhou 221116, Jiangsu, Peoples R China 2.Minist Educ Peoples Republ China, Mine Digitizat Engn Res Ctr, Xuzhou 221116, Jiangsu, Peoples R China 3.Chinese Acad Sci, Inst Comp Technol, Key Lab Intelligent Informat Proc, Beijing 100190, Peoples R China
推荐引用方式 GB/T 7714	Zhang, Nan,Ding, Shifei,Sun, Tongfeng,et al. Multi-view RBM with posterior consistency and domain adaptation[J]. INFORMATION SCIENCES,2020,516:142-157.
APA	Zhang, Nan,Ding, Shifei,Sun, Tongfeng,Liao, Hongmei,Wang, Lijuan,&Shi, Zhongzhi.(2020).Multi-view RBM with posterior consistency and domain adaptation.INFORMATION SCIENCES,516,142-157.
MLA	Zhang, Nan,et al."Multi-view RBM with posterior consistency and domain adaptation".INFORMATION SCIENCES 516(2020):142-157.