Approximate kernel extreme learning machine for large scale data classification

详细信息    查看全文

• 作者：Alexandros Iosifidis^a ; ^b ; ^{alexandros.iosifidis@tut.fi}Author Vitae ; Anastasios Tefas^a ; ^{tefas@aiia.csd.auth.gr}Author Vitae ; Ioannis Pitas^a ; ^c ; ^{pitas@aiia.csd.auth.gr}Author Vitae
• 关键词：Extreme Learning Machine ; Large Scale Learning ; Facial Image Classification
• 刊名：Neurocomputing
• 出版年：2017
• 出版时间：5 January 2017
• 年：2017
• 卷：219
• 期：Complete
• 页码：210-220
• 全文大小：866 K
• 卷排序：219

文摘

In this paper, we propose an approximation scheme of the Kernel Extreme Learning Machine algorithm for Single-hidden Layer Feedforward Neural network training that can be used for large scale classification problems. The Approximate Kernel Extreme Learning Machine is able to scale well in both computational cost and memory, while achieving good generalization performance. Regularized versions and extensions in order to exploit the total and within-class variance of the training data in the feature space are also proposed. Extensive experimental evaluation in medium-scale and large-scale classification problems denotes that the proposed approach is able to operate extremely fast in both the training and test phases and to provide satisfactory performance, outperforming relating classification schemes.