Special Session 2:

Multi-party Collaborative Learning in Computational Intelligence

Aim and Scope:

Data are the oil for the operation of artificial intelligence. Nevertheless, with the exception of a few industries, the data available in most fields are of limited quantity or poor quality, making it hard to realize effective artificial intelligence applications. Due to various data policy restrictions, it is very difficult in many situations to break the barriers between data sources, and thus almost impossible to integrate the data scattered around different institutions, thus data always exist in the form of isolated islands. Therefore, it has been increasingly challenging to build efficient joint models while meeting privacy, security, and regulatory requirements, especially with scattered data or limited computational resources. In this context, there is a rising interest in multi-party learning, where data owners train machine learning models collaboratively without sharing their private data, whereas can benefit from the data of others. This is a privacy-preserving approach that enables collaborative learning by training on individual data and sharing knowledge of their local models, which achieves privacy preservation while leveraging distributed computational resources efficiently. In recent years, multi-party collaborative learning has been shown to be the premier method for solving data silo problems, producing both secure and flexible solutions compared with centralized training. This is particularly true for scattered limited, and private data settings in computational intelligence, which is the case in real-world collaborative learning and optimization problems with many participants.
Multi-party learning can be used to address machine learning challenges under diverse distributed data settings. In addition to the privacy issues, multi-party learning has great advantages in leveraging the computational resources of massive edge devices, and it achieves superior performance in building customized and personalized models according to local individual data. However, existing multi-party learning has its restrictions in the face of increasing multi-source data, and there are new challenges in overcoming the limitations imposed by designing new optimization strategies, privacy and security mechanisms, and concrete implementations in complex data environments.
This special session aims to showcase the importance of multi-party collaborative learning in computational intelligence. The guest editors believe that there is great potential in using multi-party learning to address machine learning challenges such as privacy and security concerns, distributed settings, and collaborative optimizations. The special session promotes developing new models or integrated frameworks in multi-party collaborative learning, improving the efficiency of building joint models, and introducing novel research areas. It also encourages implementing developed frameworks to solve real-world problems in different application areas such as healthcare, finance, and advertising. Special attention will be devoted to addressing challenges in multi-party learning, optimizing communication-accuracy trade-off, developing general multi-party learning, and using collaborative decision-making in prediction and pattern recognition. The goal of this special session is to solicit high-quality, original research contributions on advances in theory, algorithms, systems, and applications of multi-party collaborative learning, thereby capturing the state of the art and stimulating further developments in the related areas.

Topics:

The main topics of this special session include, but are not limited to, the following:

  • Aggregation scheme in multi-party learning
  • Incentives mechanism in multi-party learning
  • Fairness and interpretability in multi-party learning
  • Asynchronous Communication in multi-party learning
  • Privacy, security, and robustness in multi-party learning
  • Fault tolerance and active sampling in multi-party learning
  • Adaptive and personalized techniques in multi-party learning
  • Resource allocation and management in multi-party learning
  • Heterogeneous data, model, and system in multi-party learning
  • Model compression and communication efficiency in multi-party learning
  • Collaborative optimization and convergence analysis in multi-party learning
  • Incomplete, imbalance, and multimodal data in multi-party learning
  • New applications for collaborative multi-party learning
  • Introduction to organizers:

    Yuan Gao received the B. Eng degree and Ph.D. degree from Xidian University in 2018 and 2022. Since 2023, he has been a postdoctoral researcher with School of Electronic Engineering of Xidian University. He was a visiting scholar in School of Computer Science and Engineering, Nanyang Technological University, Singapore, from 2021 to 2022. He is the recipient of the National Program for Support of the Postdoctoral Fellowship Program of CPSF.His current research interests include secure artificial intelligence, multi-party learning, and collaborative optimization. He has so far authored or coauthored numerous refereed academic papers on various high-quality and influential journals in the above-mentioned areas, such as IEEE T-PAMI, T-CYB, T-NNLS, T-KDE, etc. Moreover, he serves as a reviewer for many leading-in-their fields journals and conferences, and as a Program Committee Member for international conferences.

    Submission Process:

    If you wish to participate in this special session--MCLCI, please submit your manuscript through the OpenReview:https://openreview.net/group?id=csaeconf.org/CSAE/2024/Special_Session_2/MCLCI.We will assign your submission to Dr.Yuan Gao for a preliminary review. After passing the preliminary review, your manuscript will undergo a secondary review by experts. Notifications of acceptance will be issued concurrently with the main conference notifications. For any questions, please contact: info@csaeconf.org