Towards Ubiquitous Internet of Things Applications with Deep Domain Adaptation and Generalization

PhD Thesis Proposal Defence


Title: "Towards Ubiquitous Internet of Things Applications with Deep 
Domain Adaptation and Generalization"

by

Miss Hua KANG


Abstract:

The rapid development of Internet-of-Things (IoT) has equipped the 
physical world with sensing, computation, and communication capabilities, 
which have changed human-centered interaction and emerged promising 
applications. Recent advances in deep learning have revolutionized many 
fields and are being adopted in IoT sensing applications. However, there 
are still some unaddressed issues regarding ubiquitous IoT applications. 
We specify four main problems faced across IoT systems. First is the data 
heterogeneity problem owing to different data collection contexts. Second 
is the time-consuming labeling burden. The third is the privacy issues of 
distributed collected data. The last is the limited communication 
bandwidth resources for IoT devices. This dissertation elaborates on these 
four problems and their emerging solutions to step towards ubiquitous IoT 
applications.

For the first two problems, we consider how to adapt the model quickly to 
unlabeled data with different distributions via unsupervised domain 
adaptation and how to make the model general for different data 
distributions via domain generalization. We present novel approaches for 
two specific applications, i.e., wireless-based gesture recognition and 
human activity recognition with partial sensor sets. For the privacy issue 
of distributed collected data and limited communication bandwidth, one of 
my works adopted a federated learning scheme to mitigate computation and 
communication overhead while protecting privacy. The other work considers 
the network layer for efficient communication. We design deep learning 
models for wireless channel prediction leveraging the stripe features 
existing in the CSI matrix to reduce bandwidth overheads due to the 
transmission of large downlink CSI matrix.


Date:			Tuesday, 10 January 2023

Time:                  	2:00pm - 4:00pm

Venue: 			Room 3494
 			Lifts 25/26

Committee Members:	Prof. Qian Zhang (Supervisor)
 			Prof. Gary Chan (Chairperson)
 			Prof. Kai Chen
 			Dr. Yangqiu Song


**** ALL are Welcome ****