Xiyuan Zhang is a Ph.D. student at Computer Science and Engineering, University of California, San Diego. She is advised by Prof. Rajesh Gupta and Prof. Jingbo Shang. Prior to UCSD, she obtained her B.S. degree in Computer Science with honors from Zhejiang University in 2020. Her research interests are in robust and efficient machine learning with applications in time-series analysis and context-aware sensing. She received the 2022 Qualcomm Innovation Fellowship. She has also held internships in AWS AI.
Machine learning (ML) recently gains increasing popularity in Internet-of-Things (IoT) applications, including healthcare, smart city, smart building, robotics, automation, and beyond. However, unlike classical vision and language applications, real-world IoT devices directly interact with humans and the physical world and are therefore in close proximity to the heterogeneity and instability of the real-world operating conditions. This often causes current ML models trained under controlled experimental settings to collapse. Our goal is to design robust and efficient ML methods for IoT devices that learn to function with real-world noisy, heterogeneous and missing data. Towards achieving this goal, we designed methods leveraging sensor data properties and incorporating contextual knowledge, and successfully improved the robustness and efficiency of current IoT devices across different domains.