Unsupervised domain adaptation (uDA) models focus on pairwise adaptation\nsettings where there is a single, labeled, source and a single target domain.\nHowever, in many real-world settings one seeks to adapt to multiple, but\nsomewhat similar, target domains. Applying pairwise adaptation approaches to\nthis setting may be suboptimal, as they fail to leverage shared information\namong multiple domains. In this work we propose an information theoretic\napproach for domain adaptation in the novel context of multiple target domains\nwith unlabeled instances and one source domain with labeled instances. Our\nmodel aims to find a shared latent space common to all domains, while\nsimultaneously accounting for the remaining private, domain-specific factors.\nDisentanglement of shared and private information is accomplished using a\nunified information-theoretic approach, which also serves to establish a\nstronger link between the latent representations and the observed data. The\nresulting model, accompanied by an efficient optimization algorithm, allows\nsimultaneous adaptation from a single source to multiple target domains. We\ntest our approach on three challenging publicly-available datasets, showing\nthat it outperforms several popular domain adaptation methods.\n