REU Supplement: Participant Support: Energy Management for Smart Residential Environments through Human-in-the-loop Algorithm Design

This REU supplement request is for the support of one undergraduate student in the Computer Science Department at University of Kentucky (UK). The student will be advised by Dr. Silvestri. The REU student will be involved in research training and experience for the semesters Spring 2021, Fall 2021, and Summer 2022. The student is expected to write scientific papers based on her research experience. REU Student Research Plan The REU student will be focusing on the application of Machine Learning (ML) to predict the energy consumption of a smart home. While the problem of predicting the energy consumption of group of homes has been widely investigated, the energy prediction of an individual home is still an open problem. The main challenge resides in the variability of user behavior, which makes the prediction non trivial and difficult to generalize. The REU student’ research will be structured in 3 phases, each lasting approximately 3-4 months. First phase: The REU student will receive exposure and training on the research issues related to the CPS CAREER funded project. The student will be given several hours of lectures introducing topics such as machine learning, energy management, and user behavioral modeling, etc. The student will also be expected to read research papers on the topic to build necessary background, and discuss further with the PI and graduate students. Second phase: The REU student will begin the development of the research project. The student will first work on real dataset of energy consumption data and identify relevant metrics for prediction (i.e., daily energy consumption, maximum/minimum/average energy consumption, next-hour energy consumption, etc.). Subsequently, the student will test standard ML algorithm such as neural networks, support vector machines, and decision trees to test accuracy with respect to these metrics. Following this initial experience to familiarize with the problem, the student will design a smart phone app to allow users to tag activities their are performing during the day. The objective is to build a more informative dataset which includes the user activities over time, in addition to the energy consumption. The final step will be to design new ML algorithms which exploit the knowledge and pattern of activities to predict the metrics listed above. During this time, the student will meet with the PI and graduate students on a regular basis (more details are discussed in the REU mentoring section) to discuss research progress and associated problems. Third phase: The REU student will make presentations of the research and implementation findings in group meetings, and make necessary revisions of the work based on the feedback, according to a spiral design approach. They will also be mentored to write a scientific paper for eventual publication. PI’s Experience PI Silvestri has supervised more than ten undergraduate students during his previous employment at the Missouri University of Science and Technology and currently at the University of Kentucky. The research experiences of these students have been tailored to their potential and skills. As an example, students with strong coding skills have been focusing on implementing complex simulators/algorithms, while high creative students have worked on designing innovative algorithmic solutions. Among the achievements of the students supervised by PI Silvestri, it is worth mentioning the experience of Brian Luciano and Eura Shin, both undergraduates from the University of Kentucky. Brian worked closely with Vijay Shah, a former graduate student of PI Silvestri, currently at Virginia Tech. Brian’s work was published at the premier conference IEEE INFOCOM 2019 [5] and the journal IEEE Transactions on Network Science and Management. 1 In addition, Brian is currently employed at CISCO, a leading company in communication networks. Eura Shin has worked for two years in PI Silvestri’s lab, and immediately showed her brilliant creativity. She led a new research direction focusing on the design of stream based active learning approaches for appliance identification in a smart home. The results of her work have been accepted at IEEE COMSNETS [4] and are also currently under review at ACM Transactions on Cyber-Physical Systems. Thanks to her research experience, Eura received the Goldwater Scholarship in 2019, the CRA Honorable Mention in 2019, and the NSF Graduate Research Fellowship in 2020. She is now a PhD student at Harvard University.