My research focuses on dense self-supervised learning, which learns meaningful representations for each patch of an image rather than a single vector for the whole image. This approach has applications in unsupervised object detection and semantic segmentation. Additionally, I work on video self-supervised learning to develop better pretraining weights that capture the temporal aspects of video, aiding tasks like action recognition. I also work on ML safety, particularly in outlier detection, to ensure that AI systems can identify and handle unexpected inputs reliably and securely.