Overcoming Data Distribution Shift in AIoT Mobile Sensing

PhD Thesis Proposal Defence


Title: "Overcoming Data Distribution Shift in AIoT Mobile Sensing"

by

Mr. Tianlang HE


Abstract:

The Artificial Intelligence of Things (AIoT) is reshaping mobile sensing,
where devices like smartphones, wearables, and drones collect data from
diverse users and environments to enable smarter services spanning mobile
health, location AI, and the low-altitude economy. Increasingly, mobile
sensing employs AI models to process sensor data, such as IMU measurements,
magnetic fields, and radio signals. However, the distribution of sensing
data is often non-stationary owing to factors like user heterogeneity and
environmental changes, giving rise to data distribution shifts and hence
unexpected failures of AI models. This thesis develops AIoT models to
overcome data distribution shifts in mobile sensing, advancing robust
capabilities through adaptation and generalization.

First, we establish adaptation capabilities with TASFAR, a source-free
domain adaptation model that enables robust regression in target-agnostic
AIoT tasks, such as pedestrian localization, people counting, and travel
time estimation.

Then, to enhance generalization, we introduce two classifiers based on
causal representation learning: PRID, which achieves robust IMU-assisted BLE
proximity detection amidst multipath fading and device carriage states, and
ELESON, which classifies conveyor states using phone IMU and magnetometer
data under arbitrary user behaviors.

Finally, by extending generalization to sequential decision-making, we
introduce LoRaCompass, a deep reinforcement learning model that guides a
LoRa sensor to efficiently and robustly search for an IoT tag across
diverse, unknown environments.


Date:                   Friday, 28 November 2025

Time:                   9:00am - 11:00am

Venue:                  Room 2408
                        Lift 17/18

Committee Members:      Prof. Gary Chan (Supervisor)
                        Prof. Raymond Wong (Co-supervisor)
                        Prof. Song Guo (Chairperson)
                        Prof. Qian Zhang