Image
About
Decentralized collaborative systems encompass a broad spectrum of information processing tasks, extending from federated model training across devices and enterprises to autonomous multi-agent workflows. Across these varied modalities, a central barrier to practical deployment remains establishing trust, defined here as the rigorous assurance of safety, privacy, and security among entities that may not fully trust one another. While decentralized architectures provide a useful structure for decomposing complex workloads, they do not inherently guarantee protection. Indeed, exchanging model updates, intermediate representations, or agent instructions can introduce unique vulnerabilities and expand the attack surface. Furthermore, adapting theoretical privacy and security primitives to these distributed environments often creates significant friction regarding efficiency, accuracy, and verification. This workshop explores principled approaches for building trustworthy decentralized systems, addressing the distinct technical and theoretical requirements of both traditional federated learning and nascent agentic ecosystems.
If you require special accommodation, please contact our access coordinator at simonsevents@berkeley.edu with as much advance notice as possible.
Please note: the Simons Institute regularly captures photos and video of activity around the Institute for use in videos, publications, and promotional materials.
Chairs/Organizers
