Machine Teaching: Supervised Learning and Beyond

This lecture by Jerry Zhu from the University of Wisconsin-Madison provides a gentle overview of machine teaching, focusing on the optimal design of training data to teach a learner. Begin with an exploration of version space learners in classification to understand the teaching dimension, comparing it with learning from iid data and active learning. Continue with convex risk minimization learners and discover the connection between teaching and optimal control. Finally, examine teaching applications in reinforcement learning and games. Throughout the 58-minute talk, part of the Theoretical Aspects of Trustworthy AI series at the Simons Institute, encounter numerous open problems in this evolving field.