HUFS Professor Lee Junhyun Has Four Papers Accepted at AI Conferences

Research Recognized at the World's Three Leading AI Conferences

Four papers co-authored by Lee Junhyun, a professor in the Department of Computer Engineering at Hankuk University of Foreign Studies, as corresponding author were accepted in succession to ICLR 2026, ICML 2026, and NeurIPS 2026, three of the world’s leading artificial intelligence (AI) conferences.

Lee Junhyeon, professor in the Division of Computer Engineering at Hankuk University of Foreign Studies. Hankuk University of Foreign Studies

ICLR, ICML, and NeurIPS are internationally renowned conferences in the fields of artificial intelligence and machine learning.

They provide a venue for researchers from universities and global companies around the world to present their latest research, with outstanding papers selected through a rigorous review process.

This year, Professor Lee presented four papers as corresponding author: one each at ICLR and ICML and two at NeurIPS, earning recognition for his research at all three of the world’s leading AI conferences.

In particular, his research demonstrated competitiveness across a broad range of fields, from evaluating AI reasoning capabilities to multi-AI collaboration, information exchange between models, and model compression.

These achievements resulted from collaborative research with academic and industry researchers in Korea and abroad, including LG AI Research, the University of Illinois Chicago (UIC) in the United States, Capital One AI Foundations, and Chung-Ang University.

The paper accepted to ICLR 2026 presented a new methodology for more accurately evaluating the actual reasoning capabilities of AI models.

Moving beyond conventional fixed evaluation problems, the approach has AI agents generate new problems themselves and progressively increase their difficulty to test models’ reasoning capabilities.

The approach drew attention for its potential to enable the continuous and flexible evaluation of rapidly advancing AI technology.

The research presented at ICML 2026 addressed inefficiencies that arise when multiple AIs collaborate.

Unlike conventional approaches in which AIs discuss every problem, the proposed method has them discuss only when they have low confidence in an answer. It maintains performance while reducing computational costs to nearly the level of using a single AI.

Two papers on efficient information exchange between AI models and model compression were accepted to NeurIPS 2026.

The first study developed a technology that enables different large language models (LLMs) to exchange internal information directly, rather than communicating in natural language.

This improved collaboration between AIs while reducing communication time to about one-eleventh of the previous level.

The second study proposed a compression technology that reduces the size of AI models while protecting essential knowledge in key fields such as medicine, law, and finance.

The research is significant in that it proposed a way to minimize the loss of important knowledge and performance during model compression.

Professor Lee is also continuing his active research. In collaboration with researchers from Korea University, he presented one paper each at Findings of EMNLP 2026, a major international conference in natural language processing, and the SCI(E)-indexed international journal Information Sciences.

“Recent AI research increasingly requires different agents to use their respective strengths and collaborate effectively,” Professor Lee said. “This research explored interactions among various AI systems, including evaluating AI reasoning capabilities, discussion among AIs, and efficient communication between different language models.”

He added, “This research direction also aligns with the strengths of Hankuk University of Foreign Studies in linguistic and cultural diversity. We will continue to expand our collaborations with researchers in Korea and abroad and pursue research that enhances the efficiency and collaborative capabilities of AI systems.”

This content was produced with the assistance of AI translation services.