Exploring model adaptation methods that enable real-time inference on extremely resource-constrained devices
Adapts pre-trained representation from large-scale images to diverse IMU sensing tasks
Exploring mobile system that uses a flickering flashlight and the rolling shutter effect to quickly detect hidden cameras amidst rising illegal filming crimess
Addressed the extreme noniid scenario, where the classification label and features are simultaneously non-iid via Domain-aware Contrastive Federated Learning with Major Domain Group Selection
Developing MANET distributed resource allocation algorithms, routing protocols, and routing metrics suitable for cluster robot operation
Developed prototype chair with a tablet app, featuring AI-based heart height estimation, BLE-controlled laser and motor, and cloud server measurement management.
