Intelligent Assistance: Active Learning of User Preferences and the Michigan Autonomous Guidance System

Speaker: Julie Weber

This talk will describe two projects in the design of intelligent assistants. The first part will focus on a new algorithm for active learning that is embedded within an interactive calendar management system to learn users' scheduling preferences. The algorithm uses a measure of entropy in the set of possible solutions to make online decisions about which scheduling alternatives to present to a user. It is designed to balance the goals of meeting the user's needs while at the same time ensuring efficient learning. The second part of the talk describes ongoing work on the the development of MAGS, the Michigan Autonomous Guidance System, an assistive system intended to increase the autonomy of people with cognitive disabilities. MAGS extends the Autominder system, and will eventually comprise a set of components for interfacing and interacting with users, dealing wisely with their schedules, learning their preferences (using the algorithm discussed in the first part of the talk), performing activity detection and providing appropriate guidance, accessing web-based information, and detecting and promoting social networking.


Julie S. Weber is a doctoral candidate in Computer Science and Engineering at the University of Michigan. She received her Masters degree at Tufts University in 2004 and her Bachelors degree at Wellesley College in 2003. Her research interests lie in adaptive interfaces and in assistive technology for the elderly and cognitively impaired. She interned at Google in New York in summer 2006, applying machine learning techniques to sentiment analysis on the web. She spent part of this summer working at SRI International and is looking forward to teaching computer science to middle school and high school students at Michigan's Camp CAEN.