CV
Education
University of California, Davis
Ph.D. in Computer and Information Science, September 2024 – PresentSouth China University of Technology
M.S. in Computer Technology, September 2021 – June 2024South China University of Technology
B.S. in Computer Science and Technology, September 2017 – July 2021
Research Experience
Research Assistant, University of California, Davis
Advisors: Houman Homayoun and Setareh Rafatirad
September 2024 – Present
Machine Learning for Physiological Signal Modeling
Developed a compact, task-agnostic PPG foundation model via self-supervised knowledge distillation, transferring representations from a pretrained PPG foundation model to a lightweight Transformer using masked temporal reconstruction, global representation alignment, and cosine-similarity distillation for downstream physiological tasks.
Developed a predictor-guided, stage-wise Transformer framework for reconstructing low-quality dual-wavelength PPG signals using masked temporal reconstruction, time-domain and frequency-domain objectives, and an SpO$_2$-prediction constraint to preserve physiologically relevant information.
Developed a transfer-learning framework with a self-attention BiLSTM by pretraining on clinical PPG data and fine-tuning on low-sampling-rate wearable PPG for SpO$_2$ estimation, and co-designed a related BiLSTM-attention architecture for continuous user authentication.
Developed a generalizable blood-pressure estimation framework based on attention-fused CNN features from multi-wavelength PPG signals, integrating curriculum learning from hypertension classification to BP regression and gradient-reversal adversarial training to learn subject-invariant features.
Developed an XGBoost-based SpO$_2$ confidence-assessment model using PPG signal-quality indices and accelerometer data as inputs, with SpO$_2$ prediction error serving as the supervision target.
- Developed a unit-independent wrist-GSR stress-detection pipeline using cvxEDA decomposition and robust z-score normalization to extract phasic nSCR features for stress-task classification with Random Forest classifier.
Wearable Healthcare Firmware Algorithms
Designed a lightweight, real-time wrist SpO$_2$ estimation pipeline for smartwatch firmware using beat-level AC/DC extraction, accelerometer-derived motion weighting, weighted-median ratio-of-ratios aggregation, and perfusion-guided calibration on low-rate PPG under micro-perturbations.
Implemented FFT-based spectral heart-rate estimation with Wiener filtering to suppress motion artifacts, along with PID-controlled optical intensity adjustment in smartwatch firmware.
- Led data management and preprocessing for ECG, PPG, GSR, and respiratory signals collected during stress tasks, and authored IRB protocols for human subject physiological data collection.
Research Assistant, South China University of Technology
Advisor: Jianming Lv
September 2021 – June 2024
Multimodal Pain Assessment System in Children
Developed a multimodal multi-expert framework for postoperative pediatric pain-score regression.
Trained modality-specific expert models on clustered facial and crying expressions and applied confidence-based fusion to address individual variability and label imbalance.
Developed a Django-based web platform and mobile applications for database management and multimodal result analysis in collaboration with a pediatric hospital.
Contributed to a Science and Technology Program of Guangzhou grant proposal.
Authored two Chinese patents on the pain-assessment framework.
Unsupervised Multimodal Domain Adaptation
September 2023 – April 2024
Co-designed an early-stage prototype of MODfinity comprising multimodal and label encoders, label classifiers, and domain discriminators.
Applied joint contrastive learning, domain-adversarial training, and cross-entropy objectives.
Multimodal Psychological Assessment for the Power Grid Industry
March 2021 – June 2021
Proposed a multimodal psychological-assessment framework integrating facial, speech, and gait data through CNN-LSTM, BiLSTM, and ST-GCN models.
Contributed to a paper submission to Software Guide.
Work Experience
Platform and Content Group, Tencent
Backend Development Intern — Self-Study Module in the QQ App
July 2020 – August 2020
Developed Go backend interfaces to map multiple object-ID types from the database.
Collaborated on developing a ranking system with push-pull diffusion logic.
Used MySQL for persistent storage, Redis for caching, and message queues for asynchronous updates.
Publications
Technical Skills
Research Fields: Machine Learning, Deep Learning, Signal Processing, Wearable Healthcare, Multimodal Fusion
Programming: Python, C, C++, JavaScript, Go, HTML, MySQL, Redis, Shell
Tools and Frameworks: PyTorch, Scikit-learn, OpenCV, Django, Segger Embedded Studio
Course Projects
Diversity in LLM-Simulated Survey Responses
January 2025 – March 2025
Implemented Retrieval-Augmented Generation and zero-shot prompting strategies for large language models using GPT and DeepSeek APIs.
Analyzed the distributions of LLM-generated responses against human survey data.