Thematic Axis: Distributed AI and Complex Networks
Leader(s)
Description
Here, we explore the social and technological contexts of interacting intelligent entities, covering topics such as multi-agent systems, agent-based simulation, crowd computing, the semantic web, provenance, norms and incentives, knowledge graphs, and mechanisms for trust and reputation. A key focus of this axis is the modeling, analysis, and learning over large-scale complex networks and graphs, which provide the structural backbone for distributed intelligence. This includes studying how information, decisions, and behaviors emerge, propagate, and evolve across interconnected agents and data sources, enabling scalable, robust, and trustworthy AI systems in real-world distributed environments.
Another dimension of this axis concerns confidentiality and data protection constraints. In many practical cases, data cannot be centralized due to regulatory, ethical, or technical limitations. In this context, particular emphasis is placed on federated learning, a privacy-preserving distributed learning framework in which multiple clients jointly train a shared model while keeping their data locally. Our work in this area focuses on the design and analysis of federated algorithms, with particular attention to scalability to a large number of clients, robustness to data heterogeneity, and evaluation of the quality and reliability of client contributions.
The motivation for focusing on distributed AI and complex networks as a research axis relates to the growing recognition that data, models and knowledge are rarely centralized: information is distributed across multiple sources, agents operate in parallel and interact, and data is often subject to constraints that prevent its centralization. In that sense, this axis connects naturally with several others. The knowledge representations and reasoning axis shares with it a common interest in graph-based representations: knowledge graphs and the semantic web are themselves forms of distributed, interconnected knowledge, and reasoning over them raises questions about scalability and distribution. The frugal AI axis is also closely linked, since distributing the training process is itself a response to resource and regulatory constraints, and where the efficiency of distributed algorithms is a direct concern. Finally, the explicability axis is relevant in distributed settings, where the opacity of collectively learned models raises additional challenges for transparency and accountability.