Distinguished Speaker Series 2014
Friday, February 7, 2014
Rice Hall, Rm. 130 (Auditorium)
3:30 PM (Light refreshments after the seminar Rice Hall 4th floor atrium)
Jerry ZhuAssociate Professor Department of Computer Science
University of Wisconsin, Madison
Host: Jane Qi
How to Make Machines Learn: Passive, Active, and Teaching
Machine learning plays a fundamental role in many systems. How does one effectively train a machine learner? I discuss three increasingly powerful paradigms. The first paradigm is passive learning, where the learner receives independent and identically-distributed training data from the world. The second paradigm is active learning, where the learner can query an oracle for the label of any item. The third and emerging paradigm is machine teaching, where there is a teacher who knows the learning goal and can design good training data for the machine learner. Machine teaching is appropriate for settings where the teacher wishes to maximally influence the learner but can only do so via a training set. One application of machine teaching is in education where the teacher wishes to design the best lesson for a human student. Another application is in computer security to defend against a malicious "teacher" who attacks a machine learner by manipulating its training data.
Xiaojin (Jerry) Zhu is an Associate Professor in the Department of Computer Sciences at the University of Wisconsin-Madison. Dr. Zhu received his B.S. and M.S. degrees in Computer Science from Shanghai Jiao Tong University in 1993 and 1996, respectively, and a Ph.D. degree in Language Technologies from Carnegie Mellon University in 2005. He was a research staff member at IBM China Research Laboratory from 1996 to 1998. Dr. Zhu received the National Science Foundation CAREER Award in 2010, and best paper awards at ICML, ECML/PKDD, and SIGSOFT. His research interest is in machine learning, with applications in natural language processing, cognitive science, and social media.