Research Topics
Our research focuses on designing intelligent mobile and pervasive computing systems that utilize mobile/embedded platforms and innovative algorithms.
Key topics include
Advancing sensing technologies by developing algorithms to calibrate and enhance sensor data accuracy.
Building applications and systems to address the needs of diverse populations, with a focus on empowering vulnerable communities.
Designing user-centric solutions that integrate accessibility and adaptability into real-world mobile/embedded environments.
We investigate how machine learning and AI can be optimized for mobile and IoT devices, while addressing the low-power, resource-constrained limitations of mobile and embedded platforms to enable innovative sensing capabilities.
Our focus includes
Developing federated learning frameworks for privacy-preserving AI on edge devices.
Designing energy-efficient AI model training and inference methods for resource-limited platforms.
Creating adaptive learning systems that respond effectively to dynamic mobile environments and sensing needs.
Our healthcare research aims to assist clinicians and improve patient care through intelligent systems.
Key areas include
Designing wearable devices and hospital systems for real-time health monitoring.
Analyzing medical data to provide actionable insights for diagnosis and treatment.
Creating autonomous systems for managing and predicting patient symptoms.
Projects
AttFL is a federated learning framework designed to enhance personalized deep learning models for analyzing time-series data from mobile and embedded sensing applications. By adding attention modules to baseline RNN models, it captures features efficiently and exchanges feature map information across devices via the server. The server groups devices based on cosine similarity and redistributes updated parameters to improve local inference. AttFL outperforms five existing frameworks in accuracy and efficiency, as demonstrated through evaluations on three applications and tests on a CPU emulator and a 12-node embedded testbed.
HeartQuake is an innovative geophone-based sensing system that non-intrusively extracts accurate ECG waveforms by detecting heartbeat vibrations through a bed mattress. Designed for scalability, HeartQuake's low-cost and non-intrusive approach makes it ideal for large-scale deployment compared to traditional ECG systems. Additionally, qualitative studies with physicians confirm its potential as an effective screening tool for detecting abnormal cardiovascular conditions. HeartQuake represents a significant step forward in affordable, non-intrusive cardiovascular monitoring solutions.