More about HKUST
Optimizing Multimodal Sensing and Inference for Resource-Constrained Devices: A Literature Review
PhD Qualifying Examination
Title: "Optimizing Multimodal Sensing and Inference for Resource-Constrained
Devices: A Literature Review"
by
Mr. Runxi HUANG
Abstract:
Real-time multimodal sensing and inference on resource-constrained devices is
essential for enabling applications such as autonomous driving, human-computer
interaction, smart health care and mobile agents. Such systems continuously
collect, process and fuse data from heterogeneous sensor modalities (e.g.,
videos, audios, IMUs) to enhance performance on complex tasks. However,
deploying end-to-end multimodal perception systems on devices presents
several major challenges, including the tight coupling between sensing
dynamics and model execution, the complex inter-modality dependencies, as
well as limited and dynamic resource availabilities. To systematically
investigate advancements in this field, this survey categorizes existing
optimization approaches according to the key stages of on-device multimodal
sensing and inference, including: 1) data input optimization which focuses on
adaptively selecting informative sensory data, 2) model optimization which
aims to model size, 3) inference strategy optimization which designs adaptive
inference and scheduling mechanisms, and 4) end-to-end optimization which
jointly optimizes data acquisition and model execution under resource
constraints. For each category, we outline the fundamental challenges, review
representative methodologies, and analyze their strengths and limitations.
Finally, we discuss the current progress and highlight promising research
directions for advancing end-to-end on-device multimodal inference systems.
Date: Monday, 26 January 2026
Time: 4:00pm - 6:00pm
Venue: Room 5501
Lifts 25/26
Committee Members: Dr. Xiaomin Ouyang (Supervisor)
Prof. Gary Chan (Chairperson)
Dr. Chaojian Li