PerAgents focuses on the critical systems-level challenges of transitioning from narrow, single-task edge AI to general-purpose, autonomous AI agents operating in physical environments. Currently, foundation models are trapped in the cloud due to massive computational requirements. This workshop explores how large-scale models such as Large Language Models (LLMs), Vision-Language Models (VLMs), and Multimodal Language Models (MLMs) can be compressed, decentralized, and securely integrated with pervasive sensor networks.
A defining feature of pervasive agentic systems is their domain-agnostic architectures. The exact same edge-computing architecture, split learning pipeline, and VLM integration used to fuse IoT telemetry and satellite data for a real-time decision-support system in precision agriculture can be deployed in a clinical setting to fuse wearable sensor streams for continuous patient monitoring. PerAgents investigates the underlying system architecture that remains constant even as the environmental context changes.
Systems-Level Technical Topics
The workshop is structured as a half-day to full-day event prioritizing extended debate over lengthy presentations.
Presented by a distinguished speaker from academia or industry, highlighting emerging bottlenecks in pervasive agent deployment.
Accepted papers organized into thematic sessions, prioritizing floor time for extended debate over lengthy presentations.
If submissions warrant it, interactive demonstrations and posters will showcase ongoing research, prototypes, and industrial solutions.
