SNU Professor Kyomin Jung’s team has 44 papers accepted to leading AI conferences over two years

Professor Kyomin Jung’s research team in the Department of Electrical and Computer Engineering at Seoul National University College of Engineering has had a total of 44 papers accepted to leading international artificial intelligence (AI) conferences over the past two years, with 21 papers accepted in 2025 and 23 in 2026.

The Machine Intelligence Lab (MILAB), led by Professor Jung, has conducted a broad range of research on the reasoning and training efficiency of large language models (LLMs), consistency evaluation of AI-generated images, and AI agent safety. In 2026, the lab had nine papers accepted to EMNLP and five to ACL, two of the leading international conferences in natural language processing. In machine learning, three papers were accepted to ICML, one to NeurIPS, and one to ICLR. In addition, two papers were accepted to EACL, and one each to AAAI and ECCV, bringing the total number of accepted papers this year to 23. In 2025, the team also had 21 papers accepted—nine to EMNLP, six to NAACL, two to ACL, one each to NeurIPS and ICLR, and one each to CIKM and COLING—marking the second consecutive year in which the lab had more than 20 papers accepted.

One representative study, “Evaluating Visual Narrative Coherence in Story Visualization via Diversified Storylines,” by researcher Minha Jhang and colleagues, was selected for an Oral presentation at ACL 2026. The study proposes a technique for evaluating visual narrative coherence across sequences of AI-generated images. It assesses not only whether each individual image matches its corresponding description, but also whether characters and backgrounds remain consistent across scenes.

Experimental results showed that the agreement between AI-based evaluations and human judgments reached approximately 95% of the level of agreement among human evaluators. The technology is expected to help identify inconsistencies between scenes in AI-generated picture books and narrative content, reducing the burden on creators during the review process.

Another study accepted to ICML 2026, “Beyond Normalization: Rethinking the Partition Function as a Difficulty Scheduler for RLVR,” by researcher Dohyung Kim and colleagues, proposes PACED-RL, a reinforcement learning technique designed to efficiently improve AI performance in mathematical reasoning and code generation. The method uses the partition function, which has conventionally been used primarily for normalization, to estimate the difficulty of individual problems and concentrate computational resources on problems that are more beneficial for training.

In experiments, accuracy on a mathematical reasoning benchmark increased from 23.9% with the existing method to 28.7%. In code generation evaluation, the method reduced the training time required to reach the best performance of the existing method by approximately half. The technology is expected to lower the time and cost required for AI training, helping reduce the burden of AI development for university laboratories and small and medium-sized enterprises.

The team has also achieved significant results in improving the safety of AI agents. “LiSA: Lifelong Safety Adaptation via Conservative Policy Induction,” by researcher Minbeom Kim and colleagues, was accepted to NeurIPS 2026 and introduces a technique that continuously improves an AI agent’s safety judgments based on errors reported by users. The method derives rules for protecting personal information and carrying out tasks from cases involving incorrect judgments, then selects and applies rules supported by sufficient evidence. This enables the system to respond to different security requirements across organizations and to newly emerging risk scenarios without retraining the AI model.

In experiments, the method distinguished safe actions from risky ones more accurately than existing approaches. It also maintained higher performance than the comparison methods even when 20% of user feedback contained incorrect information. The technology is expected to help prevent personal information leakage by enterprise AI assistants and improve the safety of task automation.

The research team has also been working to improve the reliability and practical applicability of AI through studies on technologies that provide personalized advice in long-term conversations, methods for generating language while protecting personal information, and approaches for evaluating AI systems’ ability to follow instructions.

The team has continued to achieve notable results in cultivating research talent as well. Yerin Hwang, who earned her Ph.D. in August 2026, was appointed a professor in the Department of Computer Science and Artificial Intelligence at Dongguk University in September of the same year. Minbeom Kim, who completed his Ph.D. around the same time, received the Best Ph.D. Dissertation Award from the Interdisciplinary Program in Artificial Intelligence at Seoul National University and is currently working at the SK hynix AI Research Center. Hwang’s current faculty affiliation and her previous research at SNU MILAB are also confirmed by her Dongguk University laboratory profile.

Professor Jung, who supervised the research, said, “At a time when AI research based on language processing, including LLMs, is becoming increasingly important, it is encouraging that we have been able to lead research by actively presenting papers over multiple years at the world’s most prestigious conferences.” He added, “We will continue to pursue creative and challenging research and contribute further to strengthening AI research capabilities at SNU Engineering and across Korea’s academic and industrial communities.”

The research introduced here was supported by the National Research Foundation of Korea and the Institute of Information & Communications Technology Planning & Evaluation, with funding from the Ministry of Science and ICT.

□ Introduction to the SNU College of Engineering

Seoul National University (SNU) founded in 1946 is the first national university in South Korea. The College of Engineering at SNU has worked tirelessly to achieve its goal of ‘fostering leaders for global industry and society.’ In 12 departments, 323 internationally recognized full-time professors lead the development of cutting-edge technology in South Korea and serving as a driving force for international development.