Variational Fair Clustering

We propose a general variational framework of fair clustering, which integrates an original Kullback-Leibler (KL) fairness term with large class clustering objectives, including prototype or graph based. Fundamentally different from the existing combinatorial and spectral solutions, our multi-term approach enables to control trade-off levels between objectives. derive tight upper bound based on concave-convex decomposition term, its Lipschitz-gradient property Pinsker’s inequality. Our can be jointly optimized various while yielding scalable solution, convergence guarantee. Interestingly, at each iteration, it performs independent update for assignment variable. Therefore, easily distributed large-scale datasets. This scalability is important as explore Unlike relaxation, formulation does not require computing eigenvalue decomposition. report comprehensive evaluations comparisons state-of-the-art methods over benchmarks, show that yield highly competitive solutions in terms