Federated Foundation Models at the Edge

As large language and multimodal models increasingly move from centralized clouds toward the edge, the assumptions of abundant resources, stable connectivity, and centralized control no longer hold. While federated learning provides a privacy-preserving framework for collaborative model adaptation, existing approaches have not been designed to support foundation models operating across heterogeneous, resource-constrained, and intermittently connected edge systems.

This workshop focuses on theoretical and algorithmic design challenges at the intersection of federated learning and on-device foundation models, including but not limited to scalability, communication efficiency, heterogeneity, personalization, security and privacy risks, and operational issues in large-scale deployment. Emphasizing cross-layer perspectives, the workshop seeks contributions that examine fundamental trade-offs between model capability, system efficiency, robustness, and privacy. This workshop is particularly relevant as regulatory pressure, privacy concerns, and time-sensitive applications are pushing foundation models toward the edge, whereas theory, principled frameworks, and empirical understanding remain largely limited.

To meet privacy, latency, and bandwidth requirements, foundation models are increasingly being deployed toward edge environments. However, existing federated learning frameworks are not designed to support foundation models under edge constraints. The scale of foundation models, combined with limited computation, communication, and energy resources at the edge, introduces tremendous new challenges. This workshop aims to provide a focused forum on federated foundation models at the network edge. We encourage submissions on novel algorithmic design, fundamental trade-offs, and real-world deployments of federated learning and on-device foundation models, including but not limited to

Topics (non-exhaustive)

• Federated/split learning for foundation models
• Communication-efficient federated/split learning
• Communication protocols for federated/split learning
• Optimization of on-device foundation model deployment
• Federated and on-device RAG systems
• Privacy and security in on-device foundation models
• Multi-agent and collaborative learning at the edge
• Cloud-edge small-large language model collaboration
• Economics in federated/split learning
• Applications of on-device foundation models
• Real-world testbeds, demos, and empirical evaluations for on-device foundation models

Submissions to the Workshop must be original and unpublished and must not be submitted concurrently for publication elsewhere. All submissions should follow the IEEE 8.5" x 11" two-column format using 10pt fonts and the IEEE Conference template (downloadable by selecting “Conferences” in the IEEE-Template Selector https://template-selector.ieee.org/).

Each submission can have up to 6 pages (including figures, tables, appendices, and references). Submissions exceeding this page limit or with smaller fonts will be desk-rejected without review. The review process is single-blind review.

For each accepted paper, at least one author is required to register and attend the workshop in-person to present their poster/paper on-site. All accepted and presented papers will be included in the IEEE ICDCSW companion conference proceedings and IEEE digital library.

Note that the authors should adhere to ethical and professional standards of IEEE. Please refer to IEEE Code of Ethics and IEEE Policy of AI-Generated Text.

Please submit via EasyChair: EasyChair submission page