Adaptive weighted noise constraint-based low-rank representation learning for robust multi-view subspace clustering

Low-rank representation is popular and effective in multi-view subspace clustering. It explores the comprehensive correlation relationships to characterize the intrinsic structure of multi-view data. However, most existing methods are limited by the performance of rank approximation and ignore the adaptability of noise, resulting in unstable clustering performance and weak generalization ability in practical applications. To tackle the mentioned issues, we propose a robust multi-view subspace clustering algorithm based on credible rank approximation and adaptive noise penalty matrix (RLRMVSC). Specifically, RLRMVSC explores the consistency and diversity of different views and proposes a higher-order rank approximation function to optimize low rank constraints. At the same time, the adaptive penalty matrix is introduced to describe the impact of noise on multi-view data. It can be done without the prior knowledge of noise and therefore reduces the dependence on noise types. The augmented Lagrange multiplier method is utilized to solve the proposed RLRMVSC model. The experiments on six real-world datasets have demonstrated the effectiveness of RLRMVSC.