While high-resolution pathology images lend themselves well to `data hungry'\ndeep learning algorithms, obtaining exhaustive annotations on these images is a\nmajor challenge. In this paper, we propose a self-supervised CNN approach to\nleverage unlabeled data for learning generalizable and domain invariant\nrepresentations in pathology images. The proposed approach, which we term as\nSelf-Path, is a multi-task learning approach where the main task is tissue\nclassification and pretext tasks are a variety of self-supervised tasks with\nlabels inherent to the input data. We introduce novel domain specific\nself-supervision tasks that leverage contextual, multi-resolution and semantic\nfeatures in pathology images for semi-supervised learning and domain\nadaptation. We investigate the effectiveness of Self-Path on 3 different\npathology datasets. Our results show that Self-Path with the domain-specific\npretext tasks achieves state-of-the-art performance for semi-supervised\nlearning when small amounts of labeled data are available. Further, we show\nthat Self-Path improves domain adaptation for classification of histology image\npatches when there is no labeled data available for the target domain. This\napproach can potentially be employed for other applications in computational\npathology, where annotation budget is often limited or large amount of\nunlabeled image data is available.\n