Xuange (Alex) Liang
梁轩阁
Hong Kong · New York
Ph.D. @ HKUST · CSE
MPH @ Columbia · Biostatistics
About
Incoming Ph.D. student in Computer Science and Engineering at HKUST (Advisor: Prof. Qian Zhang), and MPH graduate in Biostatistics (Outstanding Student) from Columbia University. My research focuses on applying foundation models and causal inference to clinical EHR data, with publications at AAAI workshops. I also bring engineering depth from internships in AI, back-end systems, and large-scale health databases.
Publications
Foundation Models for Heterogeneous Treatment Effect Estimation: Adapting Pretrained EHR Embeddings in Small Clinical Cohorts
Liang, X., Yang, B., & Wang, Y. — AAAI-26 Workshop on Health Intelligence (W3PHIAI-26), 2026
Experience
AI Intern
Shanda Group · Shanghai
Jun 2025 – Present
- Health analysis system using LLM agents on wearable & physical examination data.
- Reorganized a PostgreSQL database with 3,000+ health indicators and 200,000+ time-series rows.
- Synthesized virtual health data with C-GAN, CPAR, and Gaussian Copula models.
Back-End Engineer Intern
JD Tech (京东科技) · Beijing
Aug 2023 – Sep 2023
- Developed 10+ APIs in Golang for database management and migration scripts.
- Deployed applications on Docker and Kubernetes; optimized MySQL query performance.
Data Scientist Intern
CloudWalk Technology (云从科技) · Guangzhou
Jul 2022 – Sep 2022
- Cleaned 140+ table health database using SQL; reduced query runtime by 70% via optimized joins.
- Predicted hotel health risk with Random Forest and XGBoost (cross-validated).
Education
Ph.D. · Computer Science and Engineering
HKUST · Hong Kong SAR · Advisor: Prof. Qian Zhang
Sep 2026 – Present
M.P.H. · Biostatistics Outstanding Student · CEOR
Columbia University · New York
Sep 2024 – May 2026
B.S. · Intelligence Science and Technology
Peking University · Beijing
Sep 2020 – Jul 2024
Selected Projects
Heterogeneous Treatment Effect via EHR
Adapted CLMBR EHR embeddings with PCA + R-learner on Stanford Medicine data. Ranked #1 across 4 clinical outcomes, +7.9% over best baseline.
3D Object Detection · Multimodality
Lift-Splat-Shoot + ZoeDepth framework for continuous depth prediction. Built detection head and data pipeline on MMDetection 3D.
SDV-theta
Synthetic tabular data generation with C-GAN, CPAR, Gaussian Copula — extended fork of SDV.
IEEE-CIS Fraud Detection
XGBoost + Random Forest on 590K transactions. PR_AUC 0.95 — top 5% on Kaggle leaderboard.
Skills
PythonC++Golang
JavaSQLR
SASGit
PyTorchXGBoostDockerKubernetes
Links