Greg Canal, “Enhancing Human-computer Interfacing in Artificial Intelligence Systems”
Please join us on Tuesday, November 7, 2023 at 12:00pm in Clark Hall Room 110 and on ZOOM for the CIS & MINDS Seminar Series:
Guest: Greg Canal
Senior AI Research Scientist
Johns Hopkins University, Applied Physics Laboratory
Topic: “Enhancing Human-computer Interfacing in Artificial Intelligence Systems”
Join Zoom Meeting: https://wse.zoom.us/j/95386212146
Passcode: cis&minds
If you would like to meet with Greg Canal, please sign up at this link: https://docs.google.com/spreadsheets/d/1CeFz-KbXQm0J7ZgauSmsp2qdYwBeAoKE-mlhpG4UHUY/edit#gid=1654784549
Greg Canal
Senior AI Research Scientist
Johns Hopkins University, Applied Physics Laboratory
“Enhancing Human-computer Interfacing in Artificial Intelligence Systems”
Abstract: Tremendous progress has been made in training highly accurate machine learning models on large, static datasets in benchmark settings. However, in order to fully integrate AI systems into a wide range of high-impact, real-world scenarios that benefit humans, significant improvements must be made at the human-AI interface. In particular, in human-AI teams it may be the case that data labels and interactions with humans are limited, human responses are noisy, and decision making is shared between the human and AI agent. In this talk, I will discuss three research thrusts aimed at improving each of these challenges in human-AI interfacing. First, I will discuss new information-theoretic insights on active learning, where an AI model is trained with humans “in the loop” over a limited number of interactions. Next, I will present a recommender system model based on paired comparisons, a low-noise query type that is highly compatible with human users. Finally, I will present a method to “explain” an AI’s predictions in order to better facilitate joint human-AI decision making, and will discuss further visions for human-AI interfacing in the next generation of AI systems.
Biography: Greg Canal is a Senior AI Research Scientist in the Intelligent Systems Center at the Johns Hopkins Applied Physics Laboratory. He recently completed a postdoc with Rob Nowak at the University of Wisconsin-Madison, before which he obtained his PhD in ECE at Georgia Tech with Chris Rozell. His primary research interests relate to problems in interactive machine learning, including topics such as active learning, preference learning, and explainable AI. As he begins at APL, he is seeking new research collaborations both within APL and at JHU.