Abstract
We propose a general transfer learning framework for clustering given a maindataset and an auxiliary one about the same subjects. The two datasets mayreflect similar but different latent grouping structures of the subjects. Wepropose an adaptive transfer clustering (ATC) algorithm that automaticallyleverages the commonality in the presence of unknown discrepancy, by optimizingan estimated bias-variance decomposition. It applies to a broad class ofstatistical models including Gaussian mixture models, stochastic block models,and latent class models. A theoretical analysis proves the optimality of ATCunder the Gaussian mixture model and explicitly quantifies the benefit oftransfer. Extensive simulations and real data experiments confirm our method'seffectiveness in various scenarios.