Domain Agnostic Internal Distributions for Unsupervised Model Adaptation

We develop an algorithm for sequential adaptation of a classifier that is trained for a source domain to generalize in a unannotated target domain. We consider that the model has been trained on the source domain annotated data and then it needs to be adapted using the target domain unannotated data when the source domain data is not accessible. We align the distributions of the source and the target domains in a discriminative embedding space via an intermediate internal distribution. This distribution is estimated using the source data representations in the embedding space. We provide theoretical analysis and conduct extensive experiments on several benchmarks to demonstrate the proposed method is effective.