Cross-covariance regularized autoencoders for nonredundant sparse feature representation

doi:10.1016/j.neucom.2018.07.050

	Cross-covariance regularized autoencoders for nonredundant sparse feature representation
	Chen, Jie 1,2; Wu, ZhongCheng 2,3; Zhang, Jun 2; Li, Fang 2; Li, WenJing 2; Wu, ZiHeng 2
刊名	NEUROCOMPUTING
	2018-11-17
卷号	316 页码:49-58
关键词	Autoencoder Cross-covariance Deep learning Feature representation Receptive fields
ISSN号	0925-2312
DOI	10.1016/j.neucom.2018.07.050
通讯作者	Chen, Jie(cj2016@mail.ustc.edu.cn)
英文摘要	We propose a new feature representation algorithm using cross-covariance in the context of deep learning. Existing feature representation algorithms based on the sparse autoencoder and nonnegativity-constrained autoencoder tend to produce duplicative encoding and decoding receptive fields, which leads to feature redundancy and overfitting. We propose using the cross-covariance to regularize the feature weight vector to construct a new objective function to eliminate feature redundancy and reduce overfitting. The results from the MNIST handwritten digits dataset, the NORB normalized-uniform dataset and the Yale face dataset indicate that relative to other algorithms based on the conventional sparse autoencoder and nonnegativity-constrained autoencoder, our method can effectively eliminate feature redundancy, extract more distinctive features, and improve sparsity and reconstruction quality. Furthermore, this method improves the image classification performance and reduces the overfitting of conventional networks without adding more computational time. (C) 2018 Elsevier B.V. All rights reserved.
WOS关键词	FEATURE-EXTRACTION ; NEURAL-NETWORKS ; NONNEGATIVITY CONSTRAINTS ; DENOISING AUTOENCODERS ; DEEP ; RECOGNITION ; ALGORITHM
WOS研究方向	Computer Science
语种	英语
出版者	ELSEVIER SCIENCE BV
WOS记录号	WOS:000443971900006
内容类型	期刊论文
源URL	[http://ir.hfcas.ac.cn:8080/handle/334002/38791]
专题	中国科学院合肥物质科学研究院
通讯作者	Chen, Jie
作者单位	1.Univ Sci & Technol China, Grad Sch Comp Appl Technol, Hefei, Anhui, Peoples R China 2.Chinese Acad Sci, Hefei Inst Phys Sci, Hefei, Anhui, Peoples R China 3.Univ Sci & Technol China, Hefei, Anhui, Peoples R China
推荐引用方式 GB/T 7714	Chen, Jie,Wu, ZhongCheng,Zhang, Jun,et al. Cross-covariance regularized autoencoders for nonredundant sparse feature representation[J]. NEUROCOMPUTING,2018,316:49-58.
APA	Chen, Jie,Wu, ZhongCheng,Zhang, Jun,Li, Fang,Li, WenJing,&Wu, ZiHeng.(2018).Cross-covariance regularized autoencoders for nonredundant sparse feature representation.NEUROCOMPUTING,316,49-58.
MLA	Chen, Jie,et al."Cross-covariance regularized autoencoders for nonredundant sparse feature representation".NEUROCOMPUTING 316(2018):49-58.