Multi-view Embedding with Adaptive Shared Output and Similarity for unsupervised feature selection

作者：

Highlights：

• The method ME-ASOS incorporates the global and local structure preserving, embedding from multiple views, a shared output subspace and the adaptive similarity matrix into a framework.

• For the global structure, this method maps different views into a shared low-dimensional embedding subspace with view-wise multi-output regular projections.

• For the local structure, it learns a common similarity matrix by an improved algorithm to characterize the structures across different views.

• The experiments on public dataset show that the performance of ME-ASOS outperforms other state-of-the-art methods.

摘要

•The method ME-ASOS incorporates the global and local structure preserving, embedding from multiple views, a shared output subspace and the adaptive similarity matrix into a framework.•For the global structure, this method maps different views into a shared low-dimensional embedding subspace with view-wise multi-output regular projections.•For the local structure, it learns a common similarity matrix by an improved algorithm to characterize the structures across different views.•The experiments on public dataset show that the performance of ME-ASOS outperforms other state-of-the-art methods.

论文关键词：Multi-view embedding for feature selection,Shared multi-output subspace,Adaptive similarity,Regulation term

论文评审过程：Received 5 May 2018, Revised 16 September 2018, Accepted 13 November 2018, Available online 22 November 2018, Version of Record 7 January 2019.

论文官网地址：https://doi.org/10.1016/j.knosys.2018.11.017