New AI framework cuts compute by two-thirds, triples success on virtual tasks

A team led by professor Minhae Kwon at the School of Electronic and Electrical Engineering developed two AI technologies for complex, long-horizon tasks that reduce computational costs and improve safety. According to the authors, compute use drops by two-thirds while success on virtual tasks more than triples. The findings are published on the arXiv preprint server.

• Compute use cut by two-thirds on complex tasks
• Success rate on virtual tasks more than tripled
• Technologies target robotics and autonomous driving
• Findings published on the arXiv preprint server