Title: Machine Intelligence of Ubiquitous Computing in the Internet of Things
Student: Fei Dou
Major Advisor: Dr. Jinbo Bi
Associate Advisors: Dr. Song Han, Dr. Sheida Nabavi, Dr. Chun-Hsi Huang (external)
Review Committee Members: Dr. Derek Aguiar, Dr. Suining He
Date/Time: Thursday, Dec. 15, 2022, 9:30am – 11:00am
Location: WebEx Online
Meeting link: https://uconn-cmr.webex.com/uconn-cmr/j.php?MTID=m089ddba974f254a020bcfef34b493926
Meeting number: 2620 785 2478
Meeting password: 7hbJummWU87
Join by phone +1-415-655-0002 US Toll
Access code: 26207852478
Abstract:
Penetration of technologies such as machine learning (ML), artificial intelligence (AI), wireless broadband and Internet of Things (IoT) is propelling a rapid adoption of ubiquitous devices across the household, industrial, and commercial sectors. These devices can capture voluminous amounts of data in various formats that can be turned into valuable information for public safety, mobile healthcare, smart manufacturing, autonomous driving, service robots, assisted living, urban planning, humanitarian assistance and disaster response, etc. To reach the potential, new methods are needed for efficiently and effectively extracting, transferring, and sharing and learning useful information from ubiquitous devices while preserving user privacy.
There are challenges associated with enhancing Machine Intelligence of Ubiquitous Computing in the Internet of Things, including: 1) inefficiency and low scalability of trained models, especially when the solution space is large; 2) security and privacy preserving of user data; 3) data heterogeneity among devices and high imbalance in data distribution on individual devices; and 4) communication bottlenecks and high computational costs. Three interconnected research thrusts are investigated in this dissertation research using techniques and perspectives from reinforcement learning, federated learning, computer vision, and edge computing. 1) Novel architecture for location-based services in IoT is designed to ensure efficient and effective model training when the solution space is very large based on deep reinforcement learning. 2) Innovative privacy preserving techniques is developed to rigorously protect the privacy of personal data on IoT devices over wireless edge networks while maintaining high model accuracy and reducing communication cost under system heterogeneity. 3) Contrastive learning is employed and optimized to analyze remote sensing imagery data taken from drones/satellite to learn an accurate model under high imbalance data distribution and resource constraints.