Special Section Overview
With the exponential growth of data generated by cloud-based services and edge devices, distributed machine learning has become a foundational pillar of modern AI applications. In particular, decentralized learning schemes—such as federated learning—have emerged as promising solutions that facilitate collaborative model training across multiple clients without requiring the exchange of raw data. This paradigm is particularly well-suited to cloud computing environments, where stringent privacy regulations, heterogeneous network conditions, and limited communication bandwidth pose significant challenges. Nonetheless, achieving robust security and maintaining high system efficiency in such distributed settings remain critical and unresolved issues.
Cloud infrastructures, despite offering substantial computational capabilities, are inherently characterized by resource asymmetry, dynamic connectivity, and diverse privacy constraints. Conventional distributed training frameworks, originally designed for centralized data centers with homogeneous hardware and stable networking, often prove inadequate in these heterogeneous and privacy-sensitive scenarios. Moreover, the distributed nature of decentralized learning introduces unique challenges, including Non-IID data distributions, client unreliability, asynchronous updates, and susceptibility to adversarial behaviors.
This special session aims to explore the convergence of secure and efficient distributed machine learning with cloud computing. We welcome original research contributions, forward-looking position papers, and preliminary investigations that address the theoretical foundations, algorithmic innovations, and practical deployments of secure, scalable, and efficient learning frameworks in distributed cloud environments.
Topics of interest include, but are not limited to:
Topics of Interest
- Optimization Techniques for Distributed Learning in the Cloud (e.g., beyond first-order methods, adaptive and asynchronous optimization)
- Communication- and Resource-Efficient Distributed Learning (e.g., communication compression, update sparsification, hardware co-design)
- Handling System and Statistical Heterogeneity (e.g., data, model, and device heterogeneity; personalization strategies; cross-device and cross-silo settings)
- Privacy-Preserving and Secure Learning Protocols (e.g., differential privacy, secure aggregation, robustness to adversarial attacks)
- Incentive Mechanisms and Economic Models (e.g., game-theoretic designs, reputation systems, auctions, pricing)
- Foundation Models in Federated/Distributed Contexts (e.g., parameter-efficient fine-tuning, knowledge distillation)
- Responsible and Trustworthy Learning (e.g., fairness, bias, interpretability, accountability, ethical considerations)
- Domain-Specific Applications (e.g., secure medical collaboration, financial forecasting with sensitive data, smart city and IoT analytics)
Important Dates
- Paper Submission: September 30, 2025
- Notification of Acceptance: October 15, 2025
- Camera-Ready: October 30, 2025
- Workshop Date: November 16, 2025 (TBC)
Submission Guidelines
Authors are invited to submit original, unpublished research papers. Submissions should not exceed 8 pages, including tables, figures, and references in IEEE CS format. The template files for LATEX or WORD can be downloaded from the IEEE site https://www.ieee.org/conferences/publishing/templates.html. Submission must be made in PDF format only with savable text and embedded fonts.
The review process will be doubly blind, and so a submission should not include any information that may identify the authors of the manuscripts. Technical content of the camera-ready manuscript must be identical to the submitted version except for changes made to address TPC review comments.
For each accepted submission, at least one of the co-authors must have a full conference registration and present the work in person.
Special Section Chairs
- Xuehe Wang, Sun Yat-sen University, wangxuehe@mail.sysu.edu.cn
- Zhe Wang, Nanjing University of Science and Technology, zwang@njust.edu.cn
- Yiyang, Pei, Singapore Institute of Technology, yiyang.pei@singaporetech.edu.sg
- Xiao Zhang, South-Central Minzu University, xiao.zhang@my.cityu.edu.hk
Program Committee (Tentative)
- Zhiguang Cao, School of Computing and Information Systems, Singapore Management University, Singapore
- Wenya Wang, College of Computing and Data Science, Nanyang Technological University, Singapore
- Chuan Chen, School of Computer Science and Engineering, Sun Yat-sen University, China
- Pan Lai, Department of Computer Science, South-Central Minzu University, China
- Wenhao Yuan, School of Artificial Intelligence, Sun Yat-sen University, China
Contact Information
For any inquiries about the section, please contact:
Email: wangxuehe@mail.sysu.edu.cn
Keynote Speaker
Prof. Lingjie Duan
Professor
Hong Kong University of Science and Technology (Guangzhou), China
Human-in-the-Loop Learning: Taming Agency and Incentives for Smart Cities
Abstract
IoT and Machine Learning (ML) are transforming urban systems, but face a fundamental challenge: they are Human-in-the-Loop. When humans act as rational sensors and users, their personal incentives often introduce systemic inefficiencies. This talk presents Human-in-the-Loop Learning solutions that merge ML with human-aware mechanism design to proactively account for human agency and improve collective outcomes.
We address three critical, human-created problems:
1. The Traffic Paradox: Crowd-sourced navigation platforms prioritize exploitation (fastest route), causing users to converge and creating a tragedy of the commons (congestion). We introduce strategic information disclosure mechanisms that guide user routing decisions and provide strong, provable performance guarantees.
2. The Perils of Real-time Information: In IoT-enabled services, providing real-time queue updates can encourage humans to game the system, paradoxically reducing efficiency. We develop optimal algorithms to determine the ideal, non-real-time frequency of information updates, significantly stabilizing system performance.
3. Truthfulness in Personalized Systems: In personalized systems (e.g., clinical trials), agents may strategically misreport private contexts (e.g., symptoms) to gain an advantage. We show that traditional ML algorithms fail to ensure truthfulness. We propose a novel mechanism using linear programming that guarantees agents report truthfully while maintaining near-optimal learning performance.
By explicitly designing for human behavior, our approach transforms these systemic bottlenecks into opportunities for efficient, user-centric smart city management.
Biography
Lingjie Duan is a Professor in the Internet of Things Thrust of Hong Kong University of Science and Technology (Guangzhou). He earned his PhD from The Chinese University of Hong Kong in 2012. Following a Visiting Scholarship at the University of California, Berkeley, he joined the Singapore University of Technology and Design (SUTD), a collaboration with MIT.
His research is distinctively interdisciplinary, covering human-centric AI, distributed machine learning, and computer networking. His editorial leadership is evident through his roles as an Editor for IEEE/ACM Transactions on Networking and an Associate Editor for IEEE Transactions on Mobile Computing. He previously contributed as an Editor for IEEE Transactions on Wireless Communications and a Guest Editor for IEEE Journal on Selected Areas in Communications, and notably served as General Chair for the 21st IEEE/IFIP WiOpt International Conference (2023) in Singapore.
His contributions to interdisciplinary research "Network Economics" with publications in top engineering, AI, and business venues have garnered significant acclaim. Since 2021, Stanford University has recognized him among the World's Top 2% Scientists, and ScholarGPS named him a 2022 Highly Ranked Scholar. His awards include the First Prize Paper Award from the Computer Academy of Guangdong (2024), a Finalist for the Best Paper Award at IEEE/IFIP WiOpt 2024, and the SUTD Excellence in Research Award. Earlier in his career, he was the Runner-up for the IEEE ComSoc Best Young Professionals Award (2016) and received the 10th Asia-Pacific Outstanding Young Researcher Award from the IEEE Communications Society (2015). His research has been robustly supported by over $10 million USD in funding, spanning grants from the National Research Foundation of Singapore, Ministry of Education, Ministry of Defence, Singapore DSO National Laboratories, and partnerships with global tech giants. His mentorship has also been highly successful, with numerous Ph.D. and postdoc graduates now holding senior faculty positions in prestigious academic institutions and leading industry roles globally.