Yuan Bian

Postdoctoral Research Scientist

Columbia University

About Me

I am a Postdoctoral Research Scientist in the Department of Biostatistics at Columbia University, working with Professor Yuanjia Wang. Currently, I also serve on the Statistical Society of Canada ’s Student and Recent Graduates Committee.

I was very fortunate to be supervised by Professors Grace Y. Yi, Wenqing He, Shelley B. Bull, and Rahim Moineddin during my PhD and Master’s studies. I also have the privilege to work as research trainees at the Fields Institute for Research in Mathematical Sciences, Lunenfeld-Tanenbaum Research Institute, University of Toronto, and University of Wetsern Ontario.

During my graduate studies, my research focused on developing statistical methodologies and machine learning techniques to address challenges posed by noisy data, including missing data, censored data, and measurement errors. My current research explores machine learning and reinforcement learning approaches, with applications to interventional and observational biomedical data, such as electronic health records, multi-modal neuroimaging/omics, behavioral tasks, clinical assessments, and environmental exposures.

Interests

Statistical learning
Incomplete data
Precision medicine
Uncertainty quantification
Causal inference

Education

PhD in Statistics, 2024
University of Western Ontario
MSc in Biostatistics, Thesis Option, 2020
University of Toronto
HBSc in Mathematics and Statistics, 2017
University of Toronto

Research

Last updated on Jan 7, 2025

Characterizing Decision-Making Dynamics in Reinforcement Learning Experiments

This work is ongoing.

Last updated on Jan 7, 2025

Boosting Learning with Incomplete Data

This work advances boosting methods by developing strategies to handle missing or interval-censored response variables, implementing them with functional gradient descent, establishing theoretical guarantees, and demonstrating strong finite-sample performance through numerical studies.

Last updated on Jan 7, 2025

Causal Inference with Artificial Intelligence

This work examines the causal impact of stroke and myocardial infarction on self-rated health, revealing a negative effect on individuals’ perceptions of having very good or excellent health and highlighting the importance of interventions to support subjective health in aging populations.

Last updated on Jan 7, 2025

Assessing Physician Performance in Critical Care

This work evaluates ICU physician performance by applying tree ensemble methods with TreeSHAP, and addresses confounding, achieves covariate balance, and assesses physician effects using propensity weighting with parametric models and super learning methods.

Last updated on Jan 7, 2025

Identifying the Popularity of TED Talks

This work uses network analysis, principal component analysis, and LASSO to predict TED Talk popularity, revealing that Number of Languages significantly influences popularity and that publishing on weekends, particularly Saturdays, enhances idea dissemination.

Last updated on Jan 7, 2025

Examining COVID-19 Incubation Times

This work explores different models to analyze incubation times of COVID19 from different angles, finding that the current recommended 14 day quarantine time is not long enough to control the probability of an early release of infected individuals to be small.

Last updated on Jan 7, 2025

Hypothesis Testing with Genetic Variants in Longitudinal Measurements and Time-to-Event Data

This work investigates joint models for genetic association with longitudinal biomarkers and time-to-event outcomes, developing a closed-form sample size formula, using spline functions for robustness against non-linearity, and evaluating validity, sensitivity, and power through simulations and application to genetic data from the Diabetes Control and Complications Trial.

Last updated on Jan 7, 2025

Predicting the Survival of Titanic Passengers

This work employs several machine learning techniques to predict Titanic passenger survival, confirming that women and wealthy individuals were more likely to survive, with the Conditional Forest method achieving the highest accuracy of 81.34%.

Last updated on Jan 7, 2025

Analysing Randomized Controlled Trials with Missing Response

This work evaluates the efficacy of Diclectin for nausea and vomiting during pregnancy (NVP) using a double-blind randomized controlled trial, finding that statistical inferences about its effectiveness depend on the choice of missing data methods and models, with results suggesting its benefit is not clinically significant under the pre-specified minimal clinically important difference.

Last updated on Jan 7, 2025

Inference and Variable Selection with Missing Data

This work addresses challenges in missing data analysis by proposing a unified modeling framework that uses generalized additive models to handle various missing data mechanisms without requiring specific assumptions, while incorporating regularized likelihood for concurrent estimation and variable selection, with rigorous theoretical guarantees and demonstrated empirical performance.