Xin Liu

1. Efficient deep learning for video understanding, e.g. video object detection/tracking, segmentation, and action recognition. 2. Optimization for efficient training of DNNs on large-volume temporal data. 3. Efficient diffusion models.