ISKE 2025 Special Session on Trustworthy Federated Learning for Knowledge Discovery

The 2025 International Conference on Intelligent Systems and Knowledge Engineering (ISKE 2025) is the 20th in a series of ISKE conferences. The conference will be held in Shunde, a beautiful place in southeast of China.

The conference proceedings will be published by IEEE Press (sent to be EI indexed). Special issues of SCI indexed journals will be devoted to a strict refereed selection of extended papers presented at ISKE 2025.

Scope and Motivation

The rapid growth of decentralized data ecosystems, driven by applications in healthcare, finance, smart cities, and IoT, demands collaborative knowledge discovery while preserving privacy and ethical accountability. Federated learning (FL) offers a promising framework for training models across distributed data silos without raw data sharing. However, the practical adoption of FL hinges on its trustworthiness—ensuring robustness against adversarial attacks, fairness in heterogeneous and non-IID data environments, explainability of distributed decision-making, and compliance with evolving regulations (e.g., GDPR, CCPA).

Trustworthy federated learning (FL) has emerged as a transformative paradigm for decentralized knowledge discovery, yet its integration with advanced frameworks—such as federated graph neural networks (GNNs), foundation models, and spatio-temporal analysis or continual learning—introduces novel challenges in balancing privacy, fairness, and robustness. FL systems must address the special challenge from heterogeneous graph-structured data, non-IID distributions in spatio-temporal contexts, and catastrophic forgetting in continual learning scenarios. This session explores methodologies to embed trustworthiness into these specialized FL domains, focusing on secure aggregation for federated GNNs, alignment of foundation models in decentralized settings, and adaptive mechanisms for fairness-aware, federated continual learning. Contributions are encouraged to tackle interdisciplinary challenges, fostering FL systems that are not only scalable and accurate but also inherently trustworthy.

This invited session aims at providing a promising direction, fostering open collaboration among FL co-creators while upholding transparency, fairness and robustness, without compromising sensitive local data.