Local voting based multi-view embedding.pdfVIP

Neurocomputing 171 (2016) 901–909 Contents lists available at ScienceDirect Neurocomputing journal homepage: /locate/neucom Local voting based multi-view embedding Xinjian Gao a,n, Tingting Mu b, Meng Wang a a School of Computer and Information, Hefei University of Technology, China b Department of Electrical Engineering and Electronics, The University of Liverpool, UK article info Article history: Received 26 February 2015 Received in revised form 1 May 2015 Accepted 12 July 2015 Communicated by Yongdong Zhang Available online 30 July 2015 Keywords: Multi-view learning Embeddings k-NN search Semi-supervised learning Pairwise constraints abstract This paper studies subspace based multi-view learning, investigating how to mine useful information within various kinds of data or features (views) and achieve the optimal cooperation between views. Unlike most existing methods focused on learning an optimal weighting scheme to linearly combine different types of view information, we propose to ?rst improve the original information provided by each view by designing a voting based scheme to model individual neighbor structures of the data. This leads to a set of re?ned local proximity matrices corresponding to different con?dence levels. Then, different schemes can be applied to further combine this re?ned set of composite local neighborhood representations. Also, we provide the semi-supervised version of the proposed algorithms to incorporate partially labeled objects. The experimental results demonstrate effectiveness and robustness of the proposed algorithms. 2015 Elsevier B.V. All rights reserved. 1. Introduction As the development of sensor and computer techniques, data that exhibits heterogeneous properties of the studied objects can be collected from various domains, feature collectors and extractors. These are often referred as multi-view representations of the objects and correspond to the multi-view learning task in machine learning, which has facilitated comple