A Deep Structure-Enforced Nonnegative Matrix Factorization for Data Representation

摘要：

　　In this paper,we focus on a deep structure-enforced nonnegative matrix factorization(DSeNMF)which represents a large class of deep learning models appearing in many applications.We present a unified algorithm framework,based on the classic alternating direction method of multipliers(ADMM).For updating subproblems,we derive an efficient updating rule according to its KKT conditions.We conduct numerical experiments to compare the proposed algorithm with stateof-the-art deep semi-NMF.Results show that our algorithm performs better and our deep model with different sparsity imposed indeed results in better clustering accuracy than single-layer model.Our DSeNMF can be flexibly applicable for data representation.

关键词： Deep matrix fatorization Alternating direction method Data representation

作者: Yijia Zhou Lijun Xu

作者单位: Dalian Neusoft University of Information,Dalian 116023,Liaoning,Peoples Republic of China Dalian Maritime University,Dalian 116026,Liaoning,Peoples Republic of China

会议类型: 国际会议

会议名称: 中国模式识别与计算机视觉大会(PRCV2018)

会议地点: 广州

会议语种:英文

页码: 340-350

在线出版日期: 2018-11-23（万方平台首次上网日期，不代表论文的发表时间）