UNIST Prof. Han Seung-yeol's Team Develops 'LMAC' Technology to Boost AI Communication Efficiency

Professor Han Seung-yeol's team at Ulsan National Institute of Science and Technology (UNIST) has developed 'LMAC' technology that enhances communication efficiency between multiple agents by utilizing the reasoning capabilities of large language models, and presented this at the International Conference on Machine Learning (ICML), one of the world's top three AI conferences.

The core of LMAC technology lies in automatically designing communication rules between AIs in environments where each device, such as in drone swarms or autonomous vehicles, can only perceive a partial view of the overall situation.

This technology automatically supplements communication codes by analyzing how accurately the AI receiving the message grasps the situation necessary for the mission and the information gap among team members. Each AI adopts a learning structure that compresses and shares only the information necessary for cooperation.