- Proposal for an AI-based synthesis framework to address the challenge of automatically generating complex regular expressions

  The paper titled “RESYN: A generalized recursive regular expression synthesis framework," by Professor Sang-Ki Ko's research team in the Department of Artificial Intelligence at the University of Seoul, has been accepted for presentation at IJCAI 2026 (International Joint Conferences on Artificial Intelligence), one of the most prestigious international conferences in the field of artificial intelligence.

  This study addresses the programming by example (PBE) problem of automatically generating regular expressions based solely on examples and proposes RESYN, a new recursive AI framework that can effectively synthesize complex regular expressions used in real-world environments. Previous studies rely on benchmarks with simple structures, which fail to adequately reflect the complex patterns of real-world environments. However, the research team quantified the structural complexity of real-world regular expression data and proposed a novel approach to address this issue.

△ set2regexv3

  In particular, the research team developed a lightweight neural network model called SET2REGEX, which incorporates a divide-and-conquer-based composite structure that recursively decomposes the problem and reflects the permutation invariance of the input example set. The proposed approach achieved high performance using far fewer parameters than conventional large-scale models. Moreover, it outperformed conventional methodologies and large-scale general-purpose language models on complex real-world datasets.

△ running_example 

  The paper was coauthored by Professor Sang-Ki Ko and Seongmin Kim, a master’s student, of the University of Seoul, along with a research team led by Professor Yo-Sub Han of Yonsei University. This study is expected to serve as a general methodology applicable not only to the synthesis of regular expressions but also to the synthesis of various programs with complex structures and artificial intelligence problems.

△ (From left) Professor Sang-Ki Ko of the Department of Artificial Intelligence and Seongmin Kim, a master’s student