Guided teaching interactions with robots: embodied queries and teaching heuristics

We propose to use concepts from algorithmic teaching to measure and improve human teaching for machine learners. We first investigate input examples produced by human teachers in comparison to optimal or useful teaching sequences, and find that human teachers do not naturally generate the best learning examples. Then we provide humans with teaching guidance in the form of step-by-step teaching strategy or a generic teaching heuristic, to elicit better teaching. We present results for both experiments on three different problems, showing that everyday human teachers are not naturally optimal from a machine learning perspective, but teaching g idance significantly improves their i put. This provides promising evidence that human intelligence and flexibility can be leveraged to achieve better sample efficiency when input data to a learning algorithm comes from a human teacher.