FCP researchers from the Singapore University of Technology and Design (SUTD) together with researchers from Yonsei University, Inha University and Adelaide University, have developed a breakthrough approach that could make AI assistants faster, more efficient, and better suited for future wireless networks.

The team proposed a new framework called Communication-Efficient and Uncertainty-Aware Hybrid Language Model (CU-HLM), which enables a lightweight AI model on a device to work intelligently with a more powerful cloud-based AI model. Instead of constantly transmitting large volumes of information, the system dynamically decides when communication is truly necessary and how much information needs to be sent.

The results are remarkable:

• Up to 206× higher token throughput under challenging wireless conditions.
• 74.8% fewer transmissions between device and server.
• 97.4% vocabulary compression while maintaining strong performance.
• Approximately 97.4% of the original model accuracy retained.

At a time when AI applications are rapidly expanding from cloud servers to smartphones, wearables, robots, vehicles, and future 6G networks, reducing communication overhead is becoming just as important as improving computing power. This research demonstrates how uncertainty-aware AI can intelligently balance communication, computation, and performance to enable scalable AI services at the network edge.

S. Oh, J. Kim, J. Park, S.-W. Ko, J. Choi, T. Q. S. Quek, and S.-L. Kim, “Communication-Efficient Hybrid Language Model via Uncertainty-Aware Opportunistic and Compressed Transmission,” IEEE Transactions on Communications, 2026. 

#SUTD #FCP #IMDA #NRF #ArtificialIntelligence #EdgeAI #6G #FutureCommunications