Academic Profile

Research scientist specializing in efficient, knowledge-enhanced, and trustworthy AI systems. My research connects federated and edge intelligence with efficient foundation-model inference and knowledge- and memory-augmented AI agents. I develop algorithms and systems for deployment across financial services, digital government, energy, and healthcare, with a long-term focus on reliable agentic AI under practical data and computing constraints.

Research Interests

Agentic AI and long-term memory; graph-based retrieval-augmented generation and knowledge graphs; efficient foundation-model inference; federated, privacy-preserving, and edge intelligence.

Education

• Ph.D. in Computer Science, The Hong Kong Polytechnic University, 2020 – 2024.
  • Degree conferred Dec. 11, 2024; supervised by Prof. Song Guo.
  • Dissertation: Cooperation & Competition: Mechanism Design for Federated Optimization of Edge Intelligence.
• Visiting Ph.D. Student, École Polytechnique Fédérale de Lausanne, May 2023 – Dec. 2023.
  • School of Computer and Communication Sciences; supervised by Prof. Anne-Marie Kermarrec.
  • Research focus: federated unlearning.
• B.S. in Automation, Beijing Institute of Technology, 2015 – 2019.
  • Xuteli School; research focus: game theory and control theory.

Research Appointments and Experience

• Research Scientist, Dec. 2024 – Present
  • System Group, Theory Lab (Huawei 2012 Labs), Huawei Hong Kong Research Center.
  • Efficient LLM inference: Developed model-quantization algorithms that achieve more than 2× inference acceleration without measurable accuracy degradation, and a speculative-decoding method that delivers more than 50% speedup over the multi-token prediction baseline on DeepSeek models. Both have been incorporated into a joint Huawei–DeepSeek LLM inference-acceleration project.
  • Knowledge-enhanced AI: Collaborated with Prof. Yangqiu Song’s team at HKUST on AutoSchemaKG. Related Graph-RAG and knowledge-graph technologies have been deployed for high-precision question answering over financial transactions, nuclear power plant technical manuals, and digital government services.
  • Agentic AI: Developed graph-based code intelligence and agentic-memory prototypes with measurable gains on SWE-bench Pro and LoCoMo; detailed results are summarized under Research Agenda and Current Pipeline.
  • Multimodal inference acceleration: Developed hardware-aware optimizations for Vision Transformer models on Huawei NPUs; the solution was commercialized for a medical image-recognition project at a leading tertiary hospital.
• Research Intern, Dec. 2023 – Aug. 2024
  • System Group, Theory Lab (Huawei 2012 Labs), Huawei Hong Kong Research Center.
  • Conducted research on AI-empowered time-series forecasting and multimodal LLM-based information retrieval.
• Research Assistant, Sept. 2019 – Sept. 2020
  • Department of Computing, The Hong Kong Polytechnic University; supervised by Prof. Song Guo.
  • Studied deep-reinforcement-learning-based incentive mechanisms and resource optimization for federated learning and mobile edge intelligence.
• Research Assistant, Nov. 2018 – Jul. 2019
  • State Key Laboratory of Intelligent Control and Decision of Complex Systems, Beijing Institute of Technology; supervised by Prof. Kun Liu.
  • Developed cooperative UAV–UGV system models, control laws, and tracking algorithms under dynamic visibility constraints.
• Embedded Systems Software Engineer, Apr. 2018 – Oct. 2018
  • Zhejiang Dahua Technology Co., Ltd.; worked on debugging, automated testing, and code-coverage analysis for embedded AI camera systems.

Research Agenda and Current Pipeline

My current research agenda centers on knowledge- and memory-augmented agentic AI systems, building on my earlier work in federated learning, privacy-aware optimization, and edge intelligence.

• Agentic Memory for Personalized and Domain-Specific AI: Developing a memory harness that enables AI agents to acquire, organize, retrieve, update, and maintain long-term knowledge. A prototype evaluated on LoCoMo outperforms Huawei’s open-source Jiuwen-Memory baseline by 29.6 percentage points in Recall@20 and 12.4 percentage points in question-answering accuracy. Ongoing research investigates memory quality, consistency, privacy, and downstream agent performance.
• Graph-Based Knowledge Construction and Retrieval: Extending autonomous knowledge construction and Graph-RAG toward continuously updated domain knowledge systems. Current enterprise validation focuses on high-precision question answering in finance, nuclear power, and digital government, with follow-up academic research under development.
• Graph-Based Code Intelligence: Constructing graph representations and indexes over large code repositories to support repository-level retrieval, reasoning, and automated repair. On SWE-bench Pro, the current system improves the issue-resolution pass rate by 13.5 percentage points over a Codex baseline.
• Efficient and Trustworthy Agent Deployment: Investigating model compression, speculative decoding, hardware-aware optimization, and privacy-preserving learning to deploy agentic AI under practical latency, computing, and data-governance constraints.

Publications

^* Equal contribution.

Peer-Reviewed Journal Articles

• Leijie Wu, Song Guo, Yaohong Ding, Junxiao Wang, Wenchao Xu, Yufeng Zhan, and Anne-Marie Kermarrec. "Rethinking Personalized Client Collaboration in Federated Learning." IEEE Transactions on Mobile Computing (TMC). 2024. [Paper]

• Yi Liu, Song Guo, Yufeng Zhan, Leijie Wu, Zicong Hong, and Qihua Zhou. "Chiron: A Robustness-Aware Incentive Scheme for Edge Learning Via Hierarchical Reinforcement Learning." IEEE Transactions on Mobile Computing (TMC). 2024. [Paper]

• Leijie Wu, Song Guo, Yi Liu, Zicong Hong, Yufeng Zhan, and Wenchao Xu. "Long-term Adaptive VCG Auction Mechanism for Sustainable Federated Learning with Periodical Client Shifting." IEEE Transactions on Mobile Computing (TMC). 2023. [Paper]

• Leijie Wu, Song Guo, Junxiao Wang, Zicong Hong, Jie Zhang, and Yaohong Ding. "Federated Unlearning: Guarantee the Right of Clients to Forget." IEEE Network. 2022. [Paper]

• Yufeng Zhan, Peng Li, Leijie Wu, and Song Guo. "L4L: Experience-Driven Computational Resource Control in Federated Learning." IEEE Transactions on Computers (TC). 2021. [Paper]

• Yufeng Zhan, Jie Zhang, Zicong Hong, Leijie Wu, Peng Li, and Song Guo. "A Survey of Incentive Mechanism Design for Federated Learning." IEEE Transactions on Emerging Topics in Computing (TETC). 2021. [Paper]

Peer-Reviewed Conference Papers

• Jiaxin Bai, Wei Fan, Qi Hu, Qing Zong, Chunyang Li, Hong Ting Tsang, Hongyu Luo, Yauwai Yim, Haoyu Huang, Xiao Zhou, Feng Qin, Tianshi Zheng, Xi Peng, Xin Yao, Huiwen Yang, Leijie Wu, JI Yi, Gong Zhang, Renhai Chen, and Yangqiu Song. "AutoSchemaKG: Autonomous Knowledge Graph Construction through Dynamic Schema Induction from Web-Scale Corpora." The 64th Annual Meeting of the Association for Computational Linguistics (ACL). 2026: 20557-20584. [Paper]

• Jingze Zhang, Leijie Wu, Xi Peng, Qingqing Yang, Ruilun Liu, and Hong Xu. "PP-OpenNet: Privacy-Preserved Open Set Classification for Network Traffic." IEEE International Conference on Computer Communications (INFOCOM). 2026: 1-6. [Paper]

• Leijie Wu, Yaohong Ding, Akash Dhasade, Martijn De Vos, Anne-Marie Kermarrec, and Song Guo. "QuickDrop: Efficient Federated Unlearning via Synthetic Data Generation." 25th International Middleware Conference (Middleware 2024). 2024. [Paper]

• Leijie Wu, Song Guo, Yi Liu, Zicong Hong, Yufeng Zhan, and Wenchao Xu. "Sustainable Federated Learning with Long-term Online VCG Auction Mechanism." IEEE 42nd International Conference on Distributed Computing Systems (ICDCS). 2022: 895-905. [Paper]

• Yi Liu*, Leijie Wu*, Yufeng Zhan, Song Guo, and Zicong Hong (* indicates co-first authors with equal contribution). "Incentive-Driven Long-term Optimization for Edge Learning by Hierarchical Reinforcement Mechanism." IEEE 41st International Conference on Distributed Computing Systems (ICDCS). 2021: 35-45. [Paper]

Preprints

• Leijie Wu, Song Guo, Junxiao Wang, Zicong Hong, Jie Zhang, and Jingren Zhou. "On Knowledge Editing in Federated Learning: Perspectives, Challenges, and Future Directions." arXiv. 2023. [Paper]

• Leijie Wu, Song Guo, Yaohong Ding, Junxiao Wang, Wenchao Xu, Jie Zhang, and Richard Yida Xu. "Demystify Self-Attention in Vision Transformers from a Semantic Perspective: Analysis and Application." arXiv. 2022. [Paper]

Teaching Experience

• Teaching Assistant, Department of Computing, The Hong Kong Polytechnic University, 2020 – 2023.
  • Led English-medium tutorials across all listed courses for classes of approximately 50–100 students, preparing tutorial slides and supporting teaching materials.
  • Managed student-facing teaching responsibilities as a TA, including consultation and learning support, assignments, course projects, examinations, and grading.
  • Courses:
    • COMP4431: Artificial Intelligence, Fall 2020 — Level 4 undergraduate course covering problem solving and search, knowledge representation and reasoning, and machine learning.
    • COMP4434: Big Data Analytics, Fall 2021 — Level 4 undergraduate course covering NoSQL data management, Hadoop and MapReduce, machine learning, graph analytics, recommender systems, and data visualization.
    • COMP5511: Artificial Intelligence Concepts, Spring and Fall 2022 — Level 5 postgraduate course covering search and game playing, knowledge representation and reasoning, uncertainty management, expert systems, machine learning, and neural networks.
    • COMP4121: E-Commerce Technology and Applications, Spring 2023 — Level 4 undergraduate course covering web technologies, cryptography and internet security, electronic payment systems, and e-commerce system design and implementation.

Research Supervision and Mentoring

• Master’s research mentoring: Mentored at least five master’s students on dissertation research. In Fall 2022, provided day-to-day research guidance to Zhaoyang Zhang on his dissertation, Personalized Federated Prompt Learning, which received an outstanding dissertation award at PolyU.
• Undergraduate research mentoring: Mentored at least 10 undergraduate students. Among them, Yaohong Ding coauthored QuickDrop, published at Middleware 2024, and is now pursuing a Ph.D. at The Education University of Hong Kong under the supervision of Prof. Yu Yang.
• Industry research mentoring: Mentored at least five Huawei research interns on research development, system implementation, and translation into industrial applications.

Honors and Awards

• Huawei Hong Kong Research Center Star Award (港研之星) – Team Award, Aug. 2026.
  • Awarded as a member of the Knowledge Engineering and Knowledge Graph Team.

Selected Conference Presentations

• QuickDrop: Efficient Federated Unlearning via Synthetic Data Generation. 25th International Middleware Conference (Middleware 2024), Kowloon, Hong Kong SAR, December 2024. Presented by Leijie Wu. [Event] [Paper]

• Long-term Adaptive VCG Auction Mechanism for Sustainable Federated Learning with Periodical Client Shifting. COMP 50th Anniversary Research Student Conference, Hong Kong SAR, June 2024. Presented by Leijie Wu. [Event] [Paper]

• Sustainable Federated Learning with Long-term Online VCG Auction Mechanism. IEEE 42nd International Conference on Distributed Computing Systems (ICDCS), Bologna, Italy, July 2022. Presented by Leijie Wu. [Event] [Paper]

Academic Service

• Conference reviewer: IJCAI, ICML, AAAI, NeurIPS, and IEEE INFOCOM.
• Journal reviewer: IEEE Transactions on Mobile Computing, IEEE Transactions on Cloud Computing, IEEE Internet of Things Journal, and IEEE Transactions on Sustainable Computing.

Technical Skills

• Programming: Python, C, C++, MATLAB.
• Machine learning and distributed systems: PyTorch, TensorFlow, Spark, Ray, Docker, and Linux.
• Languages: Mandarin (native), Cantonese (fluent), and English (proficient).