Machine Teaching and its Applications
I describe a branch of machine learning known as machine teaching. It is an inverse problem of learning: Given a desired target model, find a training set from which a learning algorithm learns the model. "He already has the model, what's the fuss about?" the audience murmured. You see, new applications are enabled by machine teaching: (1) We start to debug the machine learning pipeline just like we debug software, enhancing the provenance and reproducibility of data science; (2) We reveal serendipitous adversarial attacks, going beyond tweaking test items; (3) We improve human learning, optimizing the lecture based on cognitive models of the student. Beyond applications, machine teaching poses challenging questions in optimization and theory. It’s computation involves combinatorial bi-level optimization, and the theory centers on a teaching dimension whose relation to VC-dimension remains open. I hope you will find machine teaching interesting, and invite you to join the research.
Jerry Zhu is the Sheldon & Marianne Lubar Professor in the Department of Computer Sciences at the University of Wisconsin-Madison. His research focuses on machine learning, in particular machine teaching and semi-supervised learning. Jerry received his Ph.D. from Carnegie Mellon University in 2005. He is a recipient of a National Science Foundation CAREER Award in 2010, ICML classic paper prize in 2013, and several best paper awards. He is co-chair for CogSci 2018 and AISTATS 2017.