Date: Friday, March 10, 2017 at 11:00 am in Rice 536
Committee: Kamin Whitehouse (Advisor), Jack Stankovic, Westley Weimer, A.J. Brush, Microsoft Research; Quanquan Gu (Minor Representative)
Solving Practical Challenges in Ambient Intelligence Environments to Meet Personal and Societal Needs
ABSTRACT: With the explosion of physically embedded, connected technology in the Internet of Things and cyber physical systems, the vision of Ambient Intelligence Environments (first introduced in the 1980’s) is closer to reality than ever. Ambient intelligence environments are computational systems embedded in the physical environment that sense, reason about, and act for the benefit of the people in that environment. The key factor, and challenge, of an ambient intelligence environment is that it serves its objectives invisibly and transparently, with no requirements on the user’s behavior or cognitive load. Because of this, both highly personal, individual services (i.e. those a person will participate for) as well as less personal societal benefits (i.e. where people may forget to participate) can be equally supported with this technology. Such systems can help to support the projected 83.7 million U.S. elderly in 2050 and work to reduce the 40% of the world’s energy consumed in buildings. Additionally, they can sense, learn, and act to meet preferences that feed into health, happiness and better productivity.
My research aims to explore some of the practical uses and challenges of sensing and reasoning in ambient intelligence environments. First, I propose a case study to show how ambient intelligence environments can be used to meet not just personal, but societal needs with large scale impact. Then, I proposed to explore solutions to two practical challenges in ambient intelligence environments: multi-user environments and non-stationary environments. Multi-user environments are challenging in ambient sensing since a person’s identity must be sensed from the environment rather than supplied by the users. I propose to explore the privacy vs. information trade-off of performing such sensing with cameras. Non-stationary environments are challenging because a person’s preferences can change over time and it is essential to learn these preferences fast enough to provide personal services before they change. In buildings, the interactions used to learn such preferences are often few and far between. This results in a “small data” problem for learning. I propose to explore approaches to combine reinforcement learning and collaborative filtering to solve this small data problem.