Machine Teaching: Supervised Learning and Beyond

I will give a gentle overview of machine teaching, the optimal design of training data to teach a learner. We start with version space learners in classification to introduce the teaching dimension, contrasting with learning from iid data and active learning. We then move on to convex risk minimization learners and connect teaching with optimal control. Finally, we discuss teaching in reinforcement learning and games. There will be many open problems along the way.