Yu Zhang is a PhD student in the Department of Computer Science at Aalborg University (AAU), working with the DATA Section under the supervision of Prof. Arijit Khan. His research focuses on graph neural network explainability and interpretable deep learning, aiming to enhance transparency, accountability, and trust in real-world AI systems.
Yu amassed extensive experience across the internet, finance, and energy sectors. As a senior AI algorithm engineer and technology manager at the State Power Investment Corporation, he led development of award-winning AI platforms deployed in over 1,000 power plants. Holder of 10+ patents and publications, he is committed to bridging cutting-edge research and large-scale applications, fostering innovation at the intersection of AI+X and intelligent systems.
PhD Computer Science (AI Focus)
Aalborg University (AAU), Denmark
ME Mechatronic Engineering
Chongqing University of Posts and Telecommunications (CQUPT), China
BE Mechanical Design, Manufacturing and Automation
CQUPT, China
My research lies at the intersection of graph neural networks (GNNs), explainable AI, and secure machine learning.
Recently, I developed a unified framework that bridges adversarial attacks and counterfactual explanations, enabling more faithful, concise, and plausible interpretations of GNN predictions.
Looking ahead, my PhD research will explore graph data watermarking and LLM-GraphRAG, aiming to design radio-active datasets that embed detectable watermarks. This work links to data poisoning, robustness, and trustworthy AI, while offering practical tools for intellectual property protection in machine learning.
More broadly, I am excited about bridging explainability, security, and foundation models, and I actively welcome collaborations on GNN explainability, watermarking across modalities, and LLM-based graph reasoning.
This paper proposes ATEX-CF, a novel framework that integrates adversarial insights into counterfactual explanation generation for graph neural networks (GNNs). It demonstrates …
This work proposes a hybrid Luenberger–sliding mode state observer to estimate engine combustion torque, enabling effective ANN-based misfire fault diagnosis under both steady and …
After starting my PhD at Aalborg University in April 2025, my first research paper on explainable Graph Neural Networks (GNNs) was accepted at ICLR 2026, presenting novel methods …
Training in Aalborg University's Problem-Based Learning (PBL) framework as a mentee, while contributing to student project evaluation and supervising projects on GNN watermarking …
GPA: 3.55 (Top 1%).
Awarded Excellent Master’s Thesis Award.
Led two municipal-level research projects and published multiple papers in international journals (e.g., Acta Automatica Sinica, Mechanical Systems and Signal Processing).
First Prize, National Graduate Mathematical Modeling Competition (2015) [PDF]
Thesis: Research on Misfire Fault Diagnosis Algorithm of Automobile Engine [PDF]
GPA: 3.22 (Top 5%).
Received multiple scholarships including National Encouragement Scholarship (five times).
Built a solid foundation in engineering design and automation, which later supported my transition into computer science and AI research.
Thesis: Fault Diagnosis of Mechanical Bearings Based on Hidden Markov Models
Experienced in algorithm development and system implementation
ML (XGBoost, SVM) and DL (CNN, RNN, Transformer, BERT)
Entity recognition, text classification, KG construction & query
Text mining, clustering, LDA, PCA, prediction modeling
AI service platform, MySQL, Neo4j, SPARQL, Cypher
Enjoy long-distance trekking and outdoor exploration
Building custom PCs and tinkering with hardware/software
Passionate about sci-fi, history, and social service
Project Background: Designed and led the development of a fiscal intelligence system for Zhong Jin Suo, focusing on policy document retrieval and intelligent recommendation. …
Project Background: Developed a service transaction knowledge graph for ZBJ.com, China’s leading crowdsourcing and service marketplace platform. Objective: Enhance search accuracy, …
Feel free to reach out via email or social media:
