Behavioral Machine Learning
Learning systems have become ever more pervasive in our lives. By controlling the flow of information, they play an increasingly significant role in shaping decisions made by us, about us, and for us. But systems are still developed under paradigms in which human actors are abstracted away, limiting their effectiveness.
In this talk I will argue for the need to infuse into machine learning the tangible presence of humans, and present two projects addressing this need. In the first project, I consider data generated by people, specifically human choice behavior, and present a theoretically sound predictive learning framework that is capable of capturing realistic, context-dependent choice patterns. The framework can be efficiently trained end-to-end and achieves state-of-the-art performance on three large choice datasets.
In the second project, I present a learning framework for supporting human decision-makers. Using a novel human-in-the-loop training procedure, we are able to learn human-aligned representations of inputs optimized directly for human performance. We demonstrate how our method promotes good human decisions on several tasks and with various representations, including a large-scale mTurk experiment, while maintaining an inherent sense of interpretability.
Nir Rosenfeld is a Postdoctoral Fellow at Harvard's School of Engineering and Applied Sciences (SEAS), where he is part of the EconCS group, and is a fellow of the Center for Research on Computation and Society (CRCS) and of the Harvard Data Science Initiative (HDSI). Prior to this, he worked at Microsoft Research in Israel. He received his M.Sc. and Ph.D. in Computer Science from the Hebrew University in Jerusalem, and holds a B.Sc. in both Computer Science and Psychology, also from the Hebrew University.