South Korean researchers develop AI optimization and thermoelectric energy technologies

Researchers at Kookmin University, in collaboration with Hyundai Motor Company, have developed ‘SharedKD’, a technology designed to lighten 3D object detection models for autonomous driving. Unlike traditional knowledge distillation methods that use separate teacher and student models, SharedKD utilizes a single model where the entire network acts as the teacher and a pruned sub-network acts as the student. This approach allows for the simultaneous exploration and training of lightweight models suitable for real-world deployment, reducing computational load and execution time for real-time AI systems.

Separately, a research team led by Professor Kim Woo-chul at Yonsei University has developed a high-efficiency thermoelectric material capable of converting waste heat into electricity. By adding 3% manganese to tin selenide (SnSe), the team achieved a thermoelectric figure of merit (zT) of 2.3, a world-class level for polycrystalline SnSe. The technology uses magnetic nanoparticles to enhance electrical conductivity while reducing thermal conductivity by approximately 31% through lattice softening. This advancement holds potential for energy harvesting in automobiles, power plants, and electronic devices.