Xinrong (Claire) Zhou
Hi there👋! I'm a master's student in Machine Learning at Carnegie Mellon University, graduating in December 2027. Before CMU I studied Statistics and Computer Science at The Chinese University of Hong Kong. My research focuses on turning generative and multimodal models into reliable, evaluable systems for complex decision making. Glad you're here.
Education
- Master of Science in Machine Learning, School of Computer Science, Carnegie Mellon University
- Bachelor of Science, Major in Statistics, Double Major in Computer Science (First-Class Honours), CUHK
Honors and awards
- Best Original Research Award, HKSSH Free Paper Session, 37th HKSSH Annual Congress (Combined APOA Hand & Upper Limb Society 2nd Meeting), Hong Kong, 2025
- HKSAR Government Scholarship (HK$80,000)
- Asia-Pacific Economic Cooperation Scholarship (APEC Honorary Award)
- Cheung Chuk Shan Scholarship, CUHK (HK$42,100)
- Tsang Shiu Tim Scholarship, CUHK (HK$42,100)
- T. C. Cheng Postgraduate Scholarship, 2025-2026 (HK$25,000)
- Fan Fang Qi Ying Memorial Scholarship, CUHK (HK$5,000)
- Tung Wah Group of Hospitals Prize (HK$3,000)
- Departmental Scholarship, Statistics and Data Science, CUHK (HK$2,000)
- Dean's List, CUHK (2024-2025, 2023-2024, 2022–2023)
- United College Head's List (Rank #1 of College in major Statistics), CUHK (2024-2025, 2023-2024, 2022-2023)
Research
LLM-driven clinical simulation for medical education
I developed the core LLM-driven Clinical Story Tree for medical training simulators, using Pydantic-based schemas to transform expert-reviewed case reports into structured narratives that support hypothesis testing, evidence synthesis and treatment selection in adaptive doctor-patient interactions. I also engineered the dependency resolution layer for multimodal generation, coordinating visual assets, conversational speech synthesis, and audio driven video while preserving execution order and character consistency across scenes. Preprint: arXiv:2607.21570.
Upper limb necrotizing fasciitis: a 20‑year review
This research provided an epidemiological study. Statistical association between the response variables and the predictors were studied, we also produced a practical scoring system to predict mortality with a simple calculation step, good predictive power, and intuitive interpretation.
Directional dependence statistic based on Chatterjee’s correlation
This is an ongoing theoretical research project aiming to establish the foundational properties of a novel directional statistic, Δq = q(X,Y) − q(Y,X), derived from Chatterjee's rank correlation. We aim to rigorously analyze its consistency, key invariance properties, and asymptotic behavior (LLN/CLT) under conditions of mild dependence. Stay tuned to see what we discover!
Publications
MedGame: An LLM-Driven Framework for Scalable, Decision-Based Medical Simulation
A framework that compiles case-report-derived patient summaries into executable, decision-based clinical simulations, with a layered evaluation protocol that separates structural executability from clinical and educational quality.
Projects
Robust open-set animal re-identification
This project tackled open‑set wildlife individual re‑identification. I led data‑split design, built an data augmentation pipeline, handled class imbalance, and fine‑tuned pre-trained MegaDescriptor with cross‑entropy and triplet losses. A broad grid search plus species-specific thresholding delivered 110%-140% gains in GEO_MEAN. Selected as the best project in SEEM2460 and placed in the top 15% of the Kaggle AnimalCLEF25 leaderboard.
Statistics capstone: generating realistic faces
This project built realistic face generators with VAE, GAN, and a hybrid approach. I implemented the pipelines and built a VAE–GAN hybrid optimized with a composite objective using coordinated encoder/decoder/ discriminator updates and carefully tuned hyperparameters; evaluated with Inception Score and Laplacian variance, then investigated latent-space structure for facial-feature interpolation. The hybrid reached 4.42× the Laplacian sharpness of the VAE and converged in half the epochs of the GAN.
AI image generator platform
This project delivered an end‑to‑end image generation web app. I co‑designed the architecture, built Flask REST APIs and a React UI, added authentication and an SQLite data layer, and wrote integration tests. The system supports SDXL generation and Real-ESRGAN upscaling, and the Pytest, Jest and Cypress suites cover authentication, database operations and critical user journeys at 90% codebase coverage.
Internships
Xingyin Fund Management Co., Ltd
• Designed 18 alpha factors across momentum, liquidity, order-flow imbalance, volatility and behavioural signals, using daily bars, Level-1 quotes and Level-2 order book data across the full A-share universe.
• Reached Sharpe 2.1 and 37% annual excess return after style and industry neutralization, by stratifying signals on market capitalization, correcting behavioural biases and flipping signals under high-volatility and high-turnover regimes.
• Refactored the Level-2 pipeline with vectorized NumPy, rolling-window aggregation and multiprocessing, cutting end-to-end computation from about 10 minutes to under 1 minute across roughly 5,000 stocks.
Deloitte Digital Consulting
• Designed the deduplication strategy for migrating 350k ordinary and 35k VIP member records alongside transaction and invoice data, authoring 50+ merge rules agreed across client and Deloitte teams.
• Built SQL validation and key-metric reporting that reduced data integrity issues by 98%, and engineered Azure Data Factory pipelines for MongoDB and Cosmos DB migration.
• Received a full-time return offer.
Skills
- XGBoost
- Gradient Boosting
- Random Forest
- VAEs
- GANs
- LSTM
- SVM
- Clustering
- CNNs
- Transformers
- LLMs
- Fine-tuning
- Metric learning
- Diffusion models
- Web scraping
- OOP
- SPSS
- SAS
- Power BI
- Tableau
- Bloomberg Terminal
- Excel (VBA)