Ai+ Training

Large-scale labeled training datasets have enabled deep neural networks to excel across a wide range of benchmark machine learning tasks. However, in many problems, it is prohibitively expensive and time-consuming to obtain large quantities of labeled data. This talk introduces my recent research on learning with less labels. I develop domain adaptation, low-shot learning, and self-supervised learning algorithms to transfer information through multiple domains and recognize novel categories with few-shot samples. My research enables the learning system to automatically adapt to real-world variations and new environmental conditions. Specifically, I will talk about adversarial multiple source domain adaptation, multi-source distilling domain adaptation, learning invariant risks and representations for domain transfer, compositional few-shot recognition with primitive discovery and enhancing, distant-domain few-shot recognition with mid-level patterns, and generalized zero-shot learning with dual adversarial networks.