Build generalized linear mixed model to analyze outcomes of surgeries for coronary artery diseases using patient level pulled data from different studies.

Payment authorization optimization- Proposed both short-term (A/B testing) and long-term (multi-armed bandit) solutions for payment authorization optimization. - Implemented the ML-based (random forest, LGBM, multi-layer perceptron) contextual bandit in Python. Offline simulation indicated the model can increase net authorization rate by 3.6%.- Developed tools in Python for offline experimentation analysis automation and real-time performance monitoring of 800+ hardcoded segments.

Courses: AMS 310 Probability and Statistics (Spring 2022)AMS 316 Time Series (Fall 2021)

- Consulted on statistical considerations for medical device and in-vitro diagnostic clinicaltrials. - Developed protocols and statistical analysis plan to guide data analysis of clinical trials. - Conduct simulations to analyze power and determine sample sizes. - Collaborated with data management team on data collection, data cleaning and exploratory data analysis. Built data pipelines with SQL for data extraction, transformation and load.- Analyzed A/B test results and interpreted results into insights for VP and sponsors. - Implemented quantitative analysis methods including power analysis, sample size calculation, hypothesis testing and segmentation analysis with R and SAS. - Lead 10+ clinical and analytical studies, including but not limited to:A. Utilized machine learning algorithms (density-based clustering algorithms and SVM) to analyze large-scale medical data (millions of rows) in flow cytometer.B. Investigated and prototyped a Bayesian adaptive trial design to allow adaptive sample size selection based on accumulating data.C. Using Gage Repeatability and Reproducibility methodology used to quantify variation in repeated measurements of device of interest.

- Developed a quasi-Poisson linear mixed-effect model to predict the COVID-19 cases in Wuhan using the reported data in cities outside of Wuhan when COVID-19 was in the initial period of outbreak. Proved that the number of cases in Wuhan were substantially underreported and results were published in peer-reviewed journals. (link for publicaton: https://ajph.aphapublications.org/doi/pdfplus/10.2105/AJPH.2020.305893)- Assessed the programmed cell death-ligand 1 (PD-L1) expression in lung cancer patients using logistic regression and zero-inflated Poisson regression models. Results demonstrated that PD-L1 expression is significantly associated with cancer type and sites.(link for publication: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8039214/)- Evaluated the time to develop adverse cardiovascular outcomes in cancer patients treated by tyrosine kinase inhibitor (TKI) therapies using proportional hazards regression model with competing risk. Developed R package to implement random lasso (lasso with bootstrap) from scratch. Selected ~200 features using regularized regression. Trained Cox regression model to predict time-to-develop serious outcomes of cancer patients. Implemented causal inference model (propensity score matching) to compare the probability to develop cardiovascular outcomes of TKI therapies. Results have been used to guide treatment selection. (https://www.jacc.org/doi/full/10.1016/S0735-1097%2822%2902915-1)

Courses: Finite Element Method, Mechanics of Materials, Instrumental Engineering, Heat Transfer, Engineering Mechanics

We are interested in long-time-scale simulation of nano materials and the elementary mechanism of deformation in bio-materials and polymers. We developed algorithms to substantially increase the simulation time scale such that long time events that can't be analyzed by traditional simulation algorithms such as creep, unfolding of protein under experimental force levels and slow strain rate deformations can be simulated. - Developed a Monte Carlo model with gradient descent to sample local minima of atomistic potential energy surfaces, and revealed the deformation mechanisms of nanomaterials with varying external forces.- Prototyped the algorithm in C++. Contributed 8,000 lines of codes to an open-sourced parallel computing (with MPI) package. Enhanced the computational efficiency by ~10 times.- Applied the algorithm to study deformation mechanisms of nano materials and biomaterials. Results were published in 6 papers in high impact journals. (Google scholar: https://scholar.google.com/citations?user=E5pXqH4AAAAJ&hl=en)- Served as a graduate teacher in five different courses. Awarded the outstanding graduate student teacher in 2017.