Domain Invariant Representation Learning with Domain Density Transformations

Domain generalization refers to the problem where we aim to train a model on\ndata from a set of source domains so that the model can generalize to unseen\ntarget domains. Naively training a model on the aggregate set of data (pooled\nfrom all source domains) has been shown to perform suboptimally, since the\ninformation learned by that model might be domain-specific and generalize\nimperfectly to target domains. To tackle this problem, a predominant approach\nis to find and learn some domain-invariant information in order to use it for\nthe prediction task. In this paper, we propose a theoretically grounded method\nto learn a domain-invariant representation by enforcing the representation\nnetwork to be invariant under all transformation functions among domains. We\nalso show how to use generative adversarial networks to learn such domain\ntransformations to implement our method in practice. We demonstrate the\neffectiveness of our method on several widely used datasets for the domain\ngeneralization problem, on all of which we achieve competitive results with\nstate-of-the-art models.\n