I'm interested in computer vision and all the learning problems that are associated with it. In particular, I'm interested self-supervised learning and transfer. In computer vision, the standard labels we use (e.g. bounding boxes, keypoint annotations) are not only expensive to collect, but they also tend to be a poor approximation to what we actually know about images. My work aims to learn better representations from real-world image and video structure like motion and context relationships on huge datasets, as well as synthetic data where ground truth can be obtained easily. This naturally leads to questions about how to transfer representations from self-supervised tasks to tasks of interest, as well as transferring knowledge from simulation to reality.