Abstract:

Semi-supervised and Transfer Learning are ways to seek out more data to help train the model of the current task at hand. In  Semi-supervised, one collects and utilizes unlabeled data in the same domain to improve the model; while in Transfer Learning one leverages the data from the other domains to improve the model. In this talk, I will give an overview of the Semi-supervised and Transfer Learning and introduce some interesting work. I will also present some of our previous work on semi-supervised clustering and unsupervised transfer learning, which published in NIPS'16 and ICML'15 respectively, and their applications.