Zoom Meeting: https://umass-amherst.zoom.us/j/95518149029?pwd=ZWY1enpOZWVPOXc4dGtFWU1OVlIvZz09
Abstract:
Understanding the mobility of users and their devices has become ever more important in the era of mobile Internet---mobile behavior has broad implications on the design of mobile services, wireless networks, edge computing, and urban infrastructure. With the ever so increasing availability of digital device traces left by user devices, understanding user and device mobility through passively sensed data has become feasible. Over the past decade, there has been extensive work on understanding mobility at urban scales, modeling such mobility, activity identification, next location prediction, and point of interest areas. This body of work has focused on characterizing and modeling outdoor mobility at large spatial scales, such as cities and campuses, as well as different temporal scales, by using a variety of data sources such as cellular, WiFi, social media check-ins, and vehicular data. Studies have shown that humans spend over 80% of their lives indoors and inside buildings. Consequently, understanding indoor mobility is equally important for modeling and system design, but this area has seen much less work than outdoor mobility of users. Recent research has recognized that indoor mobility of users inside buildings, where many users spend a significant portion of the day, is very different from outdoor mobility exhibited when walking in a city or traveling in vehicles. While a few recent efforts have specifically focused on indoor mobility it has been from the network perspective and many research questions remain unanswered.
This thesis proposal explores the challenges and opportunities in leveraging WiFi logs generated by mobile devices to analyze and characterize mobility, infer actionable insights for designing accurate and effective mobility-aware applications, and model device mobility. My thesis has 4 main parts - First, characterization, and analysis of human and device mobility indoors and outdoors. Second, leverage WiFi logs to extract past mobility of users to design a network-centric contact tracing application- WiFiTrace. Third, design a machine learning based system, iSchedule, that leverages WiFi logs to learn the occupancy patterns of buildings and predict customized HVAC schedules of buildings for higher energy savings and user comfort. Fourth, modeling multiple devices of a user as a group using deep learning models.
Committee member details:
Prashant Shenoy
Deepak Ganesan
Jeremy Gummeson
Tauhidur Rahman