Speaker: Nikola Banovic, Carnegie Mellon University
Title: Human Data Driven Interfaces
Human data driven interfaces leverage machine learning to learn, reason about and act in response to people's behaviors, enabling applications that help people to be more productive, healthy, and safe. Traditional machine learning supports such interfaces by training models to closely match data (i.e., minimizing variance in the model by learning empirical relationships in the data). However, an alternative approach is to train models that capture real world mechanisms that generate behavior (i.e., minimizing bias in the model by grounding it in theory). The latter is more suitable to contexts where explanation is important, such as scholarly studies of large scale human behavior data, usable machine learning, and computer supported coaching. My work provides a computational model of human behavior that is grounded in behavioral theories and a holistic deﬁnition of human routine behavior. One advantage of this approach is its ability to equally support behavior prediction, detection, simulation, and explanation. I will illustrate my approach in the domain of driving safety and show how to automatically learn and explain aggressive driving behaviors, and coach drivers on how to be less aggressive and thus safer in traffic.
Nikola Banovic is a Ph.D. student at the Human-Computer Interaction Institute at Carnegie Mellon University. His research broadly focuses on enabling human data driven interfaces by creating models of human behavior ranging from low-level pointing and text entry models to high-level models of routine behavior. Nikola received his Honours Bachelor of Science and Master of Science degrees from the University of Toronto. Nikola is an NSERC Post-graduate Fellow and a Yahoo! Fellow, and has received three best paper awards.
For additional information contact Steve Easterbrook