Low-rank tensor based smooth representation learning for multi-view unsupervised feature selection

Wang, Changjia

doi:10.1016/j.knosys.2024.112902

Low-rank tensor based smooth representation learning for multi-view unsupervised feature selection

2025-01-01

1

PRE

AI

DOI：10.1016/j.knosys.2024.112902

Original

Original request for help

Save

Abstract

En 中文

As an effective dimension reduction method, multi-view unsupervised feature selection (MUFS) has attracted much attention in recent years. However, most existing MUFS methods fail to consider noise interference in the data fully and usually ignore higher-order information between different views. To address the above issues, this article proposes a novel MUFS method, named low-rank tensor based smooth representation learning (LTSRL). Specifically, LTSRL utilizes a low-pass filtering technique to remove the noise in the original data, thereby obtaining a smooth representation of each view. Meanwhile, it performs self-representation learning on the smooth representation matrix, which can get the underlying structure information of data. Then, LTSRL stacks a tensor on the self-representation matrix and imposes a weighted low-rank tensor constraint on the tensor, which can capture higher-order information and simultaneously consider the contributions of different views. Extensive experiments demonstrate the effectiveness of our proposed method in comparison with several state-of-the-art benchmark methods.

Keywords:

Multi-view learning

Smooth representation

Low-rank tensor

Feature selection

Low-rank tensor based smooth representation learning for multi-view unsupervised feature selection

Abstract

Journal

Organization