Research Interests
Throughout my Ph.D., I conduct research on the
Sensing and Computing for Edge AI Systems. My research involves the following three directions:
-
Sensor-Specific Data Processing
I have been working on the data processing algorithms on a variety of sensors, including RGB-D cameras, IMUs, and mmWave radars. Applications include object detection, object tracking, activity recognition and vital sign sensing.
-
Machine Learning Algorithms for Real-World Data
Due to the imbalanced data distribution, data insufficiency and domain shift, machine learning using the real-world data is highly challenging. I have been working on corresponding solutions on federated learning, long-tailed learning, few-shot learning and domain adaptation, etc.
-
High-Performance Inference on the Edge
The deployment of the deep learning models on edge devices is the last mile of building an AIoT application. My studies aim to accelerate the on-device inference of neural networks via hardware-aware tensor program generation using both intra-operator and inter-operator optimizations.