I am a Ph.D. student in Computer and Information Engineering at The Chinese University of Hong Kong, Shenzhen, advised by Prof. Guangxu Zhu, and a visiting student at the Shenzhen Research Institute of Big Data.
My research focuses on intelligent networks and AI-RAN, exploring how machine learning,large language models and agents can advance wireless network intelligence. My earlier work studied uncertainty quantification, sensitivity analysis, and surrogate-based optimization.
Outside research, I enjoy playing tennis and photography. I am happy to connect, exchange ideas, and collaborate on intelligent networks and AI-driven systems. Get in touch!
Submitted to IEEE Journal on Selected Areas in Communications (JSAC); arXiv preprint
QoEReasoner is an end-to-end agentic system for explainable RAN diagnosis. Deterministic tools convert KPIs into evidence, a domain knowledge base constrains causal propagation, and a stateful planner coordinates anomaly detection and root localization.
IEEE Transactions on Networking
DK-Root combines weak operational rules with scarce expert labels for mobile-network QoE diagnosis. Contrastive pretraining mitigates rule noise, conditional diffusion supplies task-faithful augmentation, and expert-guided fine-tuning sharpens the final classifier.
IEEE ICC Workshops
SemiRoot addresses label scarcity through a two-stage semi-supervised framework. Rule-guided contrastive learning structures the representation space before a small set of expert labels calibrates the decision boundaries.
IEEE ACAI
S-PPR augments Personalized PageRank with semantic information for academic collaborator discovery, combining co-authorship frequency with title-embedding similarity in graph transitions and personalization.
Engineering Optimization
This work revises expected improvement to better balance exploration and exploitation in Kriging-based optimization. REGO uses sample-distribution information to avoid premature convergence while retaining efficient global search.
International Journal for Uncertainty Quantification
A sequential Morris framework reduces the cost of sample-based sensitivity analysis. Progressive Latin hypercube sampling preserves space-filling properties while an adaptive stopping rule avoids unnecessary model evaluations.
International Conference on Algorithms, Computing and Artificial Intelligence
This paper develops a data-driven polynomial-chaos Kriging method for systems with correlated inputs. Adaptive Lasso selects a compact polynomial basis before it is embedded in a Kriging model.
IEEE International Conference on System Reliability and Safety
A global sensitivity-analysis workflow is built on arbitrary polynomial chaos expansion for cases where input distributions are only partially known, extending Sobol analysis to more realistic data-limited settings.
Visiting Student
Ph.D. · Computer & Information Engineering
Financial Technology Manager
M.Sc. · Mathematics
B.Sc. · Information & Computing Science
Which problems are merely technically feasible—and which ones truly matter?
Which incremental change actually alters the nature of the problem?
Which question could allow a reviewer to puncture the whole argument?
So what? · Why now? · Why is the existing solution insufficient? · What changes after your method succeeds?