A new study in IEEE IoT Magazine titled, “From Cloud to Edge: Enabling Offline IoT with Small-Scale Language Models," was authored by faculty members of the SUNY Poly Wireless and Intelligent Next Generation Systems (WINGS) Center. Led by Dr. Amit Sangwan and coauthored by Drs. Priyangshu Sen, Arjun Singh, Yu Zhou, Michael Medley, and Shaila Zaman (now at University of Maryland, Baltimore), introduces a novel architecture that leverages Small Language Models (SLMs) to enable intelligent, conversational Internet of Things (IoT) systems that can operate independently of cloud connectivity.

The team’s work demonstrates how emerging compact AI models—ranging from 2 to 7 billion parameters—can run efficiently on low-power edge devices like Raspberry Pi. Their framework combines natural language processing, physics-based machine learning models, and embedded logic to allow IoT systems to analyze sensor data, detect faults, and respond to user commands through intuitive, human-like interactions, all without requiring continuous internet access. This breakthrough makes IoT systems more resilient, private, and accessible, especially in rural or bandwidth-limited environments.

Through benchmarking and hands-on implementation, the researchers found that models such as Gemma2 and LLAMA3 achieved strong performance and accuracy in offline settings. This proof-of-concept underscores SUNY Poly’s growing leadership in AI-driven edge computing and workforce-aligned innovation, advancing the mission of the WINGS research center to develop intelligent, efficient, and secure technologies for next-generation networks and smart systems.

This work was supported by the National Science Foundation (NSF). 

Read more