Data-Enabled Predictive Control of Greenhouse Climate

MSc Thesis, Leiden University (2025) · Supervisor: Xiaodong Cheng

- Developed data-driven predictive controller achieving 94% performance of model-based MPC with 5× computational speedup
- Identified and analyzed safety-critical constraint violations under out-of-distribution conditions & Investigated robustness challenges in data-driven control
- Manuscript under review at ECC 2026

Predicting Depression Risk Using Machine Learning (CHARLS Dataset)

Statistical Learning Course Project, Leiden University (2024)

- Developed predictive models on large-scale health survey data (CHARLS, 15k+ individuals)
- Implemented Random Forest, Gradient Boosting, MARS, and Logistic GAM with cross-validation
- Achieved best classification accuracy of 72.9% and identified key nonlinear risk factors using interpretable ML techniques

Causal Effect of Smoking on Stroke Risk

Causal Inference Course, Leiden University (2024)

- Estimated 2.8% absolute risk increase using propensity score weighting on Framingham data (4,434 subjects, 24-year follow-up)
- Applied DAG-based confounder selection with backdoor criterion; verified positivity and balance assumptions via Love plots
- Conducted sensitivity analyses for missing data mechanisms and weight truncation strategies

Brain Iron Accumulation and Pathology Analysis

Statistical Consulting, Leiden University (2024)

- Applied non-parametric tests with multiple comparison corrections to assess regional brain iron differences across pathology groups
- Developed stepwise regression models using AIC-based variable selection, identifying significant predictors

Generative Models and Sequence-to-Sequence Learning

Introduction to Deep Learning, Leiden University (2024)

- Trained VAE and GAN models for anime face synthesis
- Built encoder-decoder architectures(RNN/LSTM) for multimodal arithmetic tasks using MNIST-based representations

Scalable Similarity Search with Locality Sensitive Hashing

Advances in Data Mining, Leiden University (2024)

- Designed a MinHash + LSH pipeline to detect similar user pairs in the Netflix Prize dataset
- Achieving efficient large-scale Jaccard similarity search without exhaustive comparisons

Product Manager

H3C Technologies, Beijing, China (Jul 2020 – Aug 2022)

Translated customer needs into data-driven product requirements. Analyzed market trends and developed bundling strategies for enterprise networking solutions.

Laboratory Assistant

National Institute of Metrology, China (Aug 2018 – Dec 2019)

Assisted in building solar cell calibration systems. Designed and tested pulse-to-continuous light conversion devices for precision measurement.