Fast Multi-View Discrete Clustering Via Spectral Embedding Fusion

Multi-view spectral clustering (MVSC) has garneredgrowing interest across various real-world applications, owingto its flexibility in managing diverse data space structures.Nevertheless, the fusion of multiple n × n similarity matricesand the separate post-discretization process hinder the utilizationof MVSC in large-scale tasks, where n denotes the number ofsamples. Moreover, noise in different similarity matrices, alongwith the two-stage mismatch caused by the post-discretization,results in a reduction in clustering effectiveness. To overcomethese challenges, we establish a novel fast multi-view discreteclustering (FMVDC) model via spectral embedding fusion, whichintegrates spectral embedding matrices (n × c, c ≪ n) to directlyobtain discrete sample categories, where c indicates the numberof clusters, bypassing the need for both similarity matrix fusionand post-discretization. To further enhance clustering efficiency,we employ an anchor-based spectral embedding strategy todecrease the computational complexity of spectral analysis fromcubic to linear. Since gradient descent methods are incapable ofdiscrete models, we propose a fast optimization strategy based onthe coordinate descent method to solve the FMVDC model effi-ciently. Extensive studies demonstrate that FMVDC significantlyimproves clustering performance compared to existing state-of-the-art methods, particularly in large-scale clustering tasks.