Published in 2022 IEEE International Conference on Communications, Control, and Computing Technologies for Smart Grids (SmartGridComm), 2022
PowerBERT, a self-supervised learning model for anomaly detection trained with minimal labeled data using the BERT model. Achieved 93% accuracy, surpassing other SOTA models by over 15%.
Published in , 2023
Further exploration of PowerBERT. New designs on PowerBERT to address the common data imbalance issue with unlabeled data in industry. Achieve 9.3% increase in accuracy on imbalanced unlabeled datasets.
Published in , 2024
Designed an insulin management system providing injection recommendations based on diet descriptions and historical glucose/injection data. Very first research work on insulin management system with supervised learning models, which dramatically reduced the dependence on the professional assistance while model training, and give patients a chance to do customized adjustments for themselves. DIETS achieved superior performance compared with SOTA models.
Published in ACM International Conference on Mobile Computing and Networking (MobiCom 2026), 2026
A lightweight framework that distills human-structure priors from a vision foundation model (Sapiens) into a compact IMU-based pose estimator — achieving 10–30% lower error and near-zero anatomically implausible poses under sparse 1–3 IMU input.
Recommended citation: Hanyu Zeng, et al. (2026). "ViPoser: Sparse-IMU Based Human Pose Estimation with Distilled Vision Foundation Priors." ACM MobiCom 2026.
Download Paper
Postgraduate course, University of Pittsburgh, Information Science Department, 2024
Postgraduate course, University of Pittsburgh, Information Science Department, 2024
Postgraduate course, University of Pittsburgh, Information Science Department, 2024