Machine teaching provides a rigorous formalism for a number of real-world applications including personalized educational systems, adversarial attacks, and program synthesis by examples. In this talk, I will discuss research questions in the context of two societal applications: (i) teaching participants of citizen science projects for biodiversity monitoring, and (ii) training medical students via educational simulators to perform surgical operations. In these educational applications, we have a human learner as the student and the focus is on developing effective teaching algorithms. Most of the existing algorithmic results on machine teaching only consider simple, idealistic settings and are not rich enough for these applications. I will highlight the need for studying more realistic student models that are suitable for teaching a human learner. I will begin by discussing the well-studied case of teaching a version space learner and then focus on two important modeling aspects: teaching a student with limited computation capability, and teaching a student who has preferences over which hypotheses to select. I will demonstrate the importance of these two modeling aspects via extensive user studies.
March 7 @ 12:30
12:30 pm (1h)
Discovery Building, Orchard View Room