Towards Ubiquitous IoT Sensing: From Low-Cost Devices to Highly-Generalizable Models

PhD Thesis Proposal Defence


Title: "Towards Ubiquitous IoT Sensing: From Low-Cost Devices to
Highly-Generalizable Models"

by

Mr. Yinan ZHU


Abstract:

Internet of Things (IoT) sensing has enabled numerous smart applications, yet
still fails to reach full ubiquity, due to two core challenges: (i) Cost:
precise sensing always relies on expensive professional devices, impeding
large-scale deployment, while commercial low-cost devices have limited sensing
granularity; (ii) Generalizability: cross-domain adaptability is severely
hindered by complex domain shifts, such as device deployment and environmental
gaps. This proposal addresses these two challenges on four typical
applications, with innovative research attempts. The first two attempts
develop algorithms to enhance the resolutions of low-cost sensors: layout
features utilization for radio-frequency identification tag groups and
task-oriented spectral reconstruction for multispectral cameras, thus
succeeding high-cost sensors. The last two attempts explore feature-wise
optimization for dataset-level domain generalization: feature distribution
alignment with source-resemble target-domain data selection and generalizable
semantic-related feature extraction across domains with large multimodal
models. Overall, our attempts advance IoT sensing ubiquity with notable cost
reduction and generalizability gain.


Date:                   Friday, 26 September 2025

Time:                   2:00pm - 4:00pm

Venue:                  Room 5501
                        Lifts 25/26

Committee Members:      Prof. Qian Zhang (Supervisor)
                        Prof. Bo Li (Chairperson)
                        Dr. Wei Wang