Evolutionary multi-objective optimization for multi-view clustering

An evolutionary many-objective approach to multiview clustering using feature and relational data

“…Recent research has reported some first steps towards exploiting the intrinsic multi-criterion nature of MvC [14,11,15,16,17,8]. In MvC, data views are available either in the form of multiple feature sets or as multiple dissimilarity matrices [5,6,18].…”

“…Although this method exhibited good performance in problems with two views, it requires high computational resources, and the extension to more than two data views was not analyzed. Jiang et al [11] used multiobjective evolutionary optimization to approximate the set of optimal trade-off solutions, seeing each view as an independent objective and describing the solutions in the form of cluster centroids. However, a suitable mechanism for a 1-1 mapping each of these centroids to a candidate partition, whilst preserving the multi-criterion nature of the problem, was not described.…”

“…Furthermore, it should make no assumptions about the importance of different views. This is particularly relevant in situations in which individual views are incommensurable, i.e., where merging all views to form a single dataset is inconvenient due to their unique properties, or where views might present varying levels of reliability [10,11]. Finally, a robust multiview clustering approach needs to be able to reliably scale to more than two views (a property we call "many-view", borrowing from the term "many-objective" first introduced in the multiobjective optimization field [12]).…”

Multi-View Data Analysis Techniques for Monitoring Smart Building Systems

“…There are many recent studies in the field of MV clustering. These include surveys [6,20,21] that summarize the work done or articles proposing novel algorithms [7,[22][23][24][25] to address the challenges in the field. While Fu et al [6] have compared the performance of the selected MV clustering algorithms on real-world data sets, Yang et al [21] and Chao et al [20] have categorized the MV clustering algorithms into different categories.…”

“…The authors of [22,23] have proposed algorithms that are capable of handling incomplete or missing data in the views. Jiang et al [24] in their work have considered MV clustering as a multi-objective optimization problem and compared how five multi-objective evolutionary algorithms work for the considered problem. In cite [25], non-negative matrix factorization is used to cluster data across the views.…”