S$^3$VAADA: Submodular Subset Selection for Virtual Adversarial Active Domain Adaptation

Unsupervised domain adaptation (DA) methods have focused on achieving maximal\nperformance through aligning features from source and target domains without\nusing labeled data in the target domain. Whereas, in the real-world scenario's\nit might be feasible to get labels for a small proportion of target data. In\nthese scenarios, it is important to select maximally-informative samples to\nlabel and find an effective way to combine them with the existing knowledge\nfrom source data. Towards achieving this, we propose S$^3$VAADA which i)\nintroduces a novel submodular criterion to select a maximally informative\nsubset to label and ii) enhances a cluster-based DA procedure through novel\nimprovements to effectively utilize all the available data for improving\ngeneralization on target. Our approach consistently outperforms the competing\nstate-of-the-art approaches on datasets with varying degrees of domain shifts.\n