Multi-view Bipartite Graph Clustering

PROJECT TITLE :

Bipartite Graph based Multi-view Clustering

ABSTRACT:

When performing graph-based multi-view clustering, one of the most important challenges is to obtain consensus on the structures of the clusters using a two-stage learning scheme. To be more specific, initially acquire knowledge of similarity graph matrices representing multiple perspectives, and then combine this information into a single, superior graph matrix. The vast majority of currently available methods discover pairwise similarities between data points for each view independently, which is a strategy that is typically utilized in single view. However, the consensus information that is found in multiple views is disregarded, and as a result, an undesirable unified graph matrix is produced as a result of the biases involved. In order to accomplish this, we suggest using an approach known as bipartite graph based multi-view clustering (BIGMC). The information that constitutes the consensus can be summarized using a limited number of representative uniform anchor points for the various points of view. In order to illustrate the relationship between the data points and the anchor points, a bipartite graph is constructed. BIGMC generates a unified bipartite graph matrix by first constructing the bipartite graph matrices for each view and then fusing those matrices together. The improvement made by the unified bipartite graph matrix to the bipartite graph similarity matrix of each view also results in an update to the anchor points. The final unified graph matrix is responsible for directly forming the final clusters. An adaptive weight is added to each view in the BIGMC algorithm in order to eliminate outlier views. In order to construct a multi-component unified bipartite graph, a low-rank constraint is imposed on the Laplacian matrix of the unified matrix. The component number corresponds to the required cluster number. In this method of optimization, the objective function is worked on in an alternating fashion. The effectiveness and superiority of this method is demonstrated by experimental results on both synthetic and real-world data sets, in comparison to the baselines used by the current state of the art.

Did you like this research project?

To get this research project Guidelines, Training and Code... Click Here