| Title |
One-step Graph-based Clustering via Learning Robust and Structured Graph from Corrupted Single-view |
| Authors |
박지웅(Jiwoong Park) ; 최진영(Jin Young Choi) |
| DOI |
https://doi.org/10.5573/ieie.2022.59.5.50 |
| Keywords |
Node clustering; Unsupervised learning; Spectral clustering |
| Abstract |
There are lots of efforts to increase the performance of graph-based clustering, such as using multi-view methods and constructing data-adaptive graphs in an one-step clustering framework. In this paper, to get the stable and accurate clustering from corrupted single-view without multi-views or data matrix information, we propose a new one-step graph-based clustering algorithm which finds a robust and structured graph. The proposed algorithm is designed to find a block diagonal affinity matrix expressed by the outer product of a nonnegative and orthogonal cluster indicator matrix. The proposed algorithm can learn the affinity matrix and cluster indicator matrix simultaneously from corrupted single-view, and does not need the post-processing (e.g., k-means clustering) steps. Using extensive datasets, we experimentally validate the effectiveness and robustness of our method and demonstrate that our method outperforms the existing state-of-the-art multi-view/one-step graph-based clustering algorithms although our method uses only a limited information of single-view corrupted affinity matrix. |