Zhiyuan Wu (Jerry)

alt text 

  • Zhiyuan Wu                                                                                                            

  • Address: No.6, Kexueyuan South Road, Zhongguancun, Haidian District, Beijing, China

  • Email: wuzhiyuan22s [at] ict.ac.cn; wuzhiyuan2000 [at] ieee.org

          [Google Scholar]        [Github]

"The greatest truths are the simplest."

About Me

I am a master’s candidate and a research assistant at the Institute of Computing Technology, Chinese Academy of Sciences (ICT/CAS). My research lies at the intersection of federated learning, edge intelligence, and distributed computing. Through the lens of distributed systems, I aim to develop scalable frameworks that bridge the gap between federated learning and edge computing, with a focus on communication efficiency, resource adaptability, and privacy preservation. My long-term goal is to build intelligent network systems that enable seamless collaboration across highly distributed and resource-heterogeneous environments.

I am best known for my work on Agglomerative Federated Learning (FedAgg), a pioneering resource-adaptive federated learning framework for large-scale model training through end-edge-cloud collaboration. FedAgg has set a new paradigm in hierarchical federated learning, and has been recognized as IEEE Xplore Top 1 Popular Article in INFOCOM 2024. I have also contributed to system-heterogeneous federated learning (e.g., FedCache and FedICT). I am a recipient of several awards, including the President Special Prize (the highest award at ICT/CAS) and multiple National Scholarships. My publications have been recognized as Top Cited and Popular Articles in federated learning and edge computing communities.

Publications

Featured Publications

FedICT: Federated Multi-task Distillation for Multi-access Edge Computing.

Zhiyuan Wu, et al. ESI Highly Cited Paper in TPDS 2024

We propose FedICT, a novel framework that resolves the trade-off between global model aggregation and personalized services in federated edge learning. FedICT enables end devices to perform multitask training in a communication-efficient and model-heterogeneous manner, and is suitable for multi-access edge computing.

Agglomerative Federated Learning: Empowering Larger Model Training via End-Edge-Cloud Collaboration.

Zhiyuan Wu, et al. IEEE Xplore Top 1 Popular Article in INFOCOM 2024

We propose FedAgg, a novel hierarchical federated learning framework that leverages end-edge-cloud collaboration to perform recursive agglomeration with model-agnostic interaction protocols. FedAgg is the first framework to enable training larger models with ever-increasing capability tier by tier up to the cloud, and satisfy the privacy constraints of federated learning as well as the flexibility requirements of end-edge-cloud computing.

FedCache: A Knowledge Cache-driven Federated Learning Architecture for Personalized Edge Intelligence.

Zhiyuan Wu, et al. IEEE Xplore Popular Article in TMC 2024.

We propose FedCache, the first cache-driven federated learning architecture that guarantees satisfactory performance while conforming to personalized devices-side limitations in edge computing. FedCache improves communication efficiency by up to 278.9 times over previous architectures and can accommodate heterogeneous devices and asynchronous interactions.

Selected Publications

Note: Bold & underlined names indicate that I or my directly supervised students are the first authors. An asterisk (*) denotes the corresponding author. A well number (#) indicates the citation ranking based on Google Scholar (as of Jan. 1, 2025), and a dollar sign ($) indicates the popularity ranking based on IEEE Xplore (as of Sep. 2024). Both rankings are determined specifically by comparing papers published in the same year within the corresponding journal or conference.

Selected Awards

Academic Service

Journal Reviewer

Conference Technical Program Committee/ Conference Reviewer

Selected Activities

Invited Talks/ Technical Report

Conference/ Symposium Suborganizer

Conference/ Symposium Participation

Professional Standards

Projects

Featured Projects

MindSpore Network Model and Federated Innovation Collaboration.

Huawei Technologies Co. Ltd.

MindSpore is an open AI framework supported by Huawei Technologies Co. Ltd. that matches with Ascend processors and supports multi-processor architectures. Before Feb. 2023, MindSpore has gained more than 3.3k stars on github, and is awarded the World's 3rd Most Popular Federated Learning Framework, only inferior to PySyft and FATE. I was involved in integrating the server communication module with the asynchronous federated learning and client selection functional modules in the framework.

China's National Standard Acquisition.

Standardization Administration of the People's Republic of China.

China's national standard acquisition is to convert the content of international standards into China national standards after investigation, analysis, research and experimental verification. The converted international standards are approved and released according to Chinese regulations. I have participated in the sinicization of 3 IEEE professional standards related to computer networks and communications, including IEEE Std 802.1Qch-2017, IEEE Std 802.1Qci-2017 and IEEE Std 802.1Qbz-2016.

Federated Learning-based Intelligent Scheduling and Collaborative Optimization at the Edge of Cloud Networks.

Institute of Computing Technology, Chinese Academy of Sciences.

This project conducts prospective research on intelligent scheduling and collaborative optimization of federated learning based on end-edge-cloud collaboration, aiming to improve the resource utilization of computing networks under hierarchical architecture. I contributed to the application form, the development of an end-edge-cloud collaborative federated learning framework, and the interim report.

Selected Projects

Mentorship

Master Students

Undergraduate Students

​