Laghate, Shantanu: SmartBox Final Abstract

View video presentation

Title: SmartBox Final Abstract

Name: Shantanu Laghate

Major: Computer Engineering and Computer Science

School affiliation: School of Engineering

Programs: James J. Slade Scholars Program

Other contributors: not listed

Abstract: The undeniable potential for in-situ sensing for applications such as HVAC efficiency and health monitoring is dependent on scalable data collection from useful sensors. Further, with the prevalence of machine learning to conduct these applications, the need for accurate labeled data is unprecedented. To fill this need, we propose a data collection platform tightly integrated into an active learning environment where the user’s knowledge alongside machine learning will be used to accurately label incoming data. In this paper, we demonstrate how the SmartBox sensing platform can be used to ease data collection and be coupled with the SmartDash web application to enable active learning-based labeling. Each SmartBox is a hardware prototype that collects data and sends it to a centralized server. Our novelty resides in the SmartDash web application that visualizes data streams from multiple boxes simultaneously and has a section dedicated to letting the user answer queries posed by active learning applications. To the best of our knowledge, this is the first application that enables scalable sensing, data storage, visualization, and time-series active learning. We used SmartBox to collect data at a student’s home and are currently developing active learning based algorithms to label that data for occupancy counting, activities of daily life monitoring, and dynamic emergency evacuation, to be continued in future work.