I am a Ph.D. student (starting Fall 2025) at
The University of Hong Kong,
where I focus on the theoretical foundations and algorithmic development of
reinforcement learning (RL) and multi-agent systems for Large Language Models (LLMs).
Research Interests
My research focuses on the algorithmic and structural foundations of intelligent decision-making systems.
I am particularly interested in:
- Multi-agent reinforcement learning frameworks for high-level decision-making
- LLM-based agent architectures for planning, collaboration, and belief modeling
These interests are grounded in prior experience with simulation-based inference and statistical modeling,
and are aimed at advancing the capabilities of autonomous multi-agent systems powered by large language models.
I completed my undergraduate studies in Mathematics and Statistics at
The University of Edinburgh.
My academic background encompassed statistical learning theory, optimization, and applied probability,
with a strong emphasis on machine learning methodologies.
I graduated with a first-class degree.
Before that, I studied Information and Computational Science at
Dalian University of Technology,
where I built a solid foundation in mathematical modeling, computational methods, and probability theory.
My coursework and research performance consistently reflected strong academic engagement and technical proficiency.
- Ph.D. Student
- Research areas: Reinforcement Learning (RL), Multi-Agent Systems, Large Language Models (LLMs)
- Advisor: Prof. Jiayu Chen
- Co-advisor: Prof. Vaneet Aggarwal, Purdue University
|
|
- BSc (Hons) in Mathematics and Statistics
- First-Class Honours (Equivalent to 4.0/4.0 GPA)
|
|
|
- BSc in Information and Computing Science
- Average score: 89.2/100
|
|
|
- Position: Research Collaborator
- Project: Dynamic Self-Rewarding for Medical Large Language Models (Med-LLMs)
- Research Focus:
- Developed a dynamic self-rewarding framework for aligning medical LLMs without human-annotated supervision
- Integrated a two-tier judge system where ChatGPT-4o dynamically refines evaluation prompts to mitigate reward misspecification and scoring bias
- Executed multi-round Direct Preference Optimization (DPO) to align model behavior through self-generated preference pairs and adaptive reward modeling
- Fine-tuned and evaluated LLMs (Mistral-7B) on domain-specific datasets (HealthCareMagic, PubMedQA, MedMCQA), targeting empathy, factuality, and coherence
- Conducted task-specific benchmarking and error analysis to uncover performance bottlenecks due to hallucination and distributional drift across iteration stages
|
|
- Position: Honours Dissertation / Final Year Project
- Supervisor: Dr. Amanda Lenzi
- Thesis Title: A Comparative Study of Simulation-Based Inference Algorithms
- Research Focus:
- Benchmarked three cutting-edge Simulation-Based Inference (SBI) algorithms—BayesFlow, Sequential Neural Likelihood (SNL), and Affine Flow Matching (AFM)—on both synthetic and structured real-world inference tasks
- Demonstrated that AFM outperforms amortized and sequential methods in capturing spatial structure in high-dimensional Poisson–CAR disease mapping models
- Designed a robust evaluation framework using recovery line metrics and ECDF-based posterior calibration to analyze estimation accuracy and uncertainty quantification
- Identified a key trade-off in joint parameter inference: increasing dimensionality improves model expressiveness but amplifies uncertainty due to parameter interaction
- Implemented a full end-to-end SBI workflow and published open-source code: github.com/xudongwu-0/SBI-comparison
|
|
- Position: Summer Research Assistant
- Supervisor: Prof. Chen Li (IEEE Fellow)
- Project: Optimizing Texera, a machine learning-based data analysis workflow platform
- Responsibilities:
- Integrated AI-driven automation for workflow optimization, enabling seamless machine learning pipeline execution
- Developed an automated report generation system that converts data analysis workflows into structured insights
- Enhanced Texera’s data cleaning and visualization capabilities to improve model interpretability
|
|
|
- Position: Research Assistant in Uncertainty Economics
- Research Focus:
- Analyzed the economic impact of uncertainty during the COVID-19 pandemic using stochastic models
- Implemented Monte Carlo simulations and Bayesian inference techniques for probabilistic estimation
- Developed computational tools for visualizing uncertainty in economic forecasting
|
|
|
- Position: Project Leader in Differential Equations
- Research Focus:
- Developed a system of ordinary differential equations (ODEs) to model bacterial infections and antibiotic treatments
- Used Fourier series analysis and Laplace transforms to predict bacterial resistance patterns
- Optimized drug treatment schedules using numerical simulations
|
|
|
- Position: Research Project in Mathematical Modeling
- Research Focus:
- Developed a population dynamics model incorporating fertility rate adjustments
- Improved predictive accuracy by incorporating age-stratified birth rate variations
- Performed parameter sensitivity analysis to optimize demographic forecasting
|
|
|
- BSc in Mathematics and Statistics, First-Class Honours
- Mathematical & Optimization Courses:
- Numerical Ordinary Differential Equations – Stability & convergence analysis, applied to RL optimization
- Honours Differential Equations – Applied in stochastic dynamic programming & Bellman equations
- Honours Complex Variables – Conformal mapping & contour integration, useful for stochastic control
- Honours Analysis – Functional analysis and variational methods, applicable to policy optimization
- Statistics & Probabilistic Modelling:
- Financial Mathematics – Stochastic processes, Ito calculus, and Black-Scholes model
- Stochastic Modelling – Markov chains, Poisson processes, Brownian motion, key concepts for RL
- Statistical Computing – Monte Carlo methods, MCMC, and Bayesian inference
- Statistical Methodology – Hypothesis testing, regression, and statistical decision theory
- Programming & Computational Methods:
- Applied Statistics – Hands-on experience with real-world data, Bayesian updating in ML
- Python, C++, MATLAB – Numerical computing, algorithm development for RL applications
|
|
- BSc in Information and Computing Science, GPA: 89.2/100
- Mathematical & Theoretical Foundations:
- Mathematical Analysis (1-3) – Real & complex analysis, measure theory fundamentals
- Ordinary Differential Equations – Theoretical and numerical solutions to dynamic systems
- Abstract Algebra – Group theory, rings, and fields, applicable in cryptography & optimization
- Complex Function Theory – Applications in function approximation & conformal mapping
- Probability & Statistical Modelling:
- Probability & Mathematical Statistics – Statistical inference, Bayesian learning, decision theory
- Real Analysis – Measure theory and Lebesgue integration, foundation for stochastic processes
- Computational & Algorithmic Skills:
- C++ Programming – Algorithm design, object-oriented programming
- Python Programming – Data structures, numerical computation, scientific computing
- Mathematical Modelling – Applied techniques in RL-based decision-making and optimization
|
|
- First-Class Scholarship - Dalian University of Technology, 2022-2023 (Top 5%)
- Dual Degree Student Scholarship - University of Edinburgh, 2022-2023
- International Study Scholarship - Dalian University of Technology, 2022-2023
|
Last update: 22 May 2025
