NEWS
  • We have two papers accepted for EMNLP 2026 main (LLM watermarking and steering LMs). Well done co-authors!
  • We have two papers accepted for CIAA 2026; one is about the pattern mining and the other is about the decomposition of regular languages. The second one is joint work with Prof. Kai Salomaa from Queen's University in Canada.
  • Our "ReSyn: A Generalized Recursive Regular Expression Synthesis Framework" paper is accepted for IJCAI 2026. Good job, Su-Hyeon. This is joint work with Prof. Sang-Ki Ko's group from University of Seoul.


    • Research Highlights
    1. Simon's Congruence Pattern Matching & Mining: We design efficient algorithms for pattern matching and mining under Simon's congruence and improve subsequence-based pattern discovery.
  • Pattern Mining under Simon's Congruence [DLT'25]
  • Simon's Congruence Pattern Matching [TCS'24]
  • On Simon's Congruence Closure of a String [TCS'23]

    • 2. LLM Watermarking: We develop robust watermarking techniques for LLMs and AI-generated codes focusing on provenance, quality preservation, linguistic awareness and resistance to attacks.
  • Linguistics-Aware Non-Distortionary LLM Watermarking [EMNLP'26] [요약]
  • A Linguistics-Aware LLM Watermarking via Syntactic Predictability [ACL'26] [요약]
  • WaterMod: Modular Token-Rank Partitioning for Probability-Balanced LLM Watermarking [AAAI'26][요약]
  • DITTO: A Spoofing Attack Framework on Watermarked LLMs [EACL'26] [요약]

    • 3. LLM Control and Safe AI: We study controllable, reliable and safe language models, and develop methods that understand and control model behavior.
  • RV-HATE: Reinforced Multi-Module Voting for Implicit Hate Speech Detection [ACL'26] [요약]
  • Steering Language Models Before They Speak: Logit-Level Interventions [EMNLP'26] [요약]
  • DLM-SWAI: Steering Diffusion Language Models Before They Unmask [arXiv'26] [요약]
  • Adaptive Steering and Remasking for Safe Generation in Diffusion Language Models [arXiv'26] [요약]


    • Recent Preprints
  • arXiv'26 / SLICE: Specification-Level Isolation of Contract Enforcement / GitHub / 요약
  • arXiv'26 / One Adapter Pair per Model: A Universal Activation Interface for Language Models / GitHub / 요약
  • arXiv'26 / EPIC: Efficient and Parallel Inference under CFG Constraints for Diffusion Language Models / GitHub / 요약
  • arXiv'26 / CRaFT: Circuit-Guided Refusal Feature Selection via Cross-Layer Transcoders / GitHub / 요약
  • arXiv'26 / DLM-SWAI: Steering Diffusion Language Models Before They Unmask / GitHub / 요약
  • arXiv'26 / KOTOX: A Korean Toxic Dataset for Deobfuscation and Detoxification / GitHub / 요약
  • arXiv'26 / STAB: Specification-driven Testing for Algorithmic Bottlenecks / GitHub / 요약
  • arXiv'26 / Sequential Behavioral Watermarking for LLM Agents / GitHub / 요약
  • arXiv'26 / Adaptive Steering and Remasking for Safe Generation in Diffusion Language Models / GitHub / 요약
  • arXiv'26 / Cross-Family Universality of Behavioral Axes via Anchor-Projected Representations / GitHub / 요약
  • arXiv'26 / NCO: A Versatile Plug-in for Handling Negative Constraints in Decoding / GitHub / 요약
  • arXiv'26 / How Does the Thinking Step Influence Model Safety? An Entropy-based Safety Reminder for LRMs / GitHub / 요약
  • arXiv'25 / RegexPSPACE: A Benchmark for Evaluating LLM Reasoning on PSPACE-Complete Regex Problems / GitHub / summary / 요약


    Contact Information
    School of Computing, Yonsei University
    50 Yonsei-Ro, Seodaemun-Gu, Seoul 03722, Republic of Korea
    Email: emmous [at] yonsei [dot] ac [dot] kr

    Last updated 2026/8/25