Ruiyi Yan

Ruiyi Yan

Ph.D. Student, Kyoto University

I am a Ph.D. student at the Language Media Processing Lab, Graduate School of Informatics, Kyoto University, advised by Associate Professor Yugo Murawaki.

My earlier work is on the security of text that language models generate — linguistic steganography and watermarking, where I looked at how provable guarantees hold up once a message has to survive a real channel, and at what the tokenizer quietly breaks along the way. I also surveyed that field.

My current focus is memory in large language models and the agents built on them. Long-term memory is still the weak point: a model accumulates history faster than it can organize it, and what it has stored degrades as it keeps being updated. I am interested in managing that from the inside — using signals the model already exposes to decide what is worth keeping, how it should be arranged, and what can be forgotten.

I am supported by a JSPS Research Fellowship for Young Scientists (DC2) and the Kyoto University DoGS SPRING Program. You can find my work on Google Scholar .

I am always happy to talk about steganography, watermarking, or agent memory — feel free to reach out by email.

🔥 News

  • 2026.08:  🎉 Two papers accepted to EMNLP 2026: MemDefrag and our survey on linguistic steganography.
  • 2026.04:  🎉 Two papers accepted to ACL 2026: Efficient Provably Secure Linguistic Steganography via Range Coding and Anchored Sliding Window.
  • 2025.12:  💻 Started a research internship at Tencent (Tokyo), working on memory for LLM agents.
  • 2025.09:  🎖 Selected for the JSPS Research Fellowship for Young Scientists (DC2) (term: 2026.04 – 2028.03).
  • 2025.08:  🎉 One paper accepted to EMNLP 2025 on tokenization inconsistency in steganography and watermarking.
  • 2024.10:  🎓 Started my Ph.D. at Kyoto University as a DoGS SPRING Fellow.

📝 Publications

* All papers below are first-author. See Google Scholar for the complete list.

  • MemDefrag: Latent Memory Defragmentation for Large Language Models, Ruiyi Yan, Zhuoyuan Mao, Yiwen Guo, EMNLP 2026 | arXiv | Code

  • A Comprehensive Survey on Linguistic Steganography: Methods, Countermeasures, Evaluation, and Challenges, Ruiyi Yan, Chenhui Chu, Zhongliang Yang, Yugo Murawaki, EMNLP 2026 | arXiv

  • Efficient Provably Secure Linguistic Steganography via Range Coding, Ruiyi Yan, Yugo Murawaki, ACL 2026, pp. 890–907 | Paper | arXiv | Code

  • Anchored Sliding Window: Toward Robust and Imperceptible Linguistic Steganography, Ruiyi Yan, Shiao Meng, Yugo Murawaki, ACL 2026, pp. 993–1012 | Paper | arXiv | Code

  • Addressing Tokenization Inconsistency in Steganography and Watermarking Based on Large Language Models, Ruiyi Yan, Yugo Murawaki, EMNLP 2025, pp. 7076–7098 | Paper | Code

  • TokenFree: A Tokenization-Free Generative Linguistic Steganographic Approach with Enhanced Imperceptibility, Ruiyi Yan, Tian Song, Yating Yang, IEEE SMC 2024, pp. 449–455 | Paper | Code

  • A Near-Imperceptible Disambiguating Approach via Verification for Generative Linguistic Steganography, Ruiyi Yan, Tian Song, Yating Yang, IEEE SMC 2024, pp. 1638–1643 | Paper | Code

  • A Secure and Disambiguating Approach for Generative Linguistic Steganography, Ruiyi Yan, Yating Yang, Tian Song, IEEE Signal Processing Letters, vol. 30, pp. 1047–1051, 2023 | Paper | Code

Preprints and Manuscripts under Review

  • HiTMS: A High-Throughput Multi-Stream Linguistic Steganography Framework, Ruiyi Yan, Zhongliang Yang, Yugo Murawaki, under review | arXiv | Code

🎖 Honors and Awards

  • 2026.04 – 2028.03, JSPS Research Fellowship for Young Scientists (DC2), Japan Society for the Promotion of Science.
  • 2024.10 – 2027.09, DoGS Fellow, Kyoto University DoGS SPRING Program.

📖 Educations

  • 2024.10 – present, Ph.D. student in Informatics (Intelligence Science and Technology), Kyoto University, Kyoto, Japan.
  • 2021.09 – 2024.06, M.S. in Cyberspace Security, Beijing Institute of Technology, Beijing, China.
  • 2017.09 – 2021.06, B.E. in Network Engineering, University of Electronic Science and Technology of China, Chengdu, China.

💻 Research Experience

  • 2025.12 – present, Research Intern, Tencent, Tokyo, Japan. Memory of LLM agents.
  • 2024.10 – present, Ph.D. Research, Kyoto University, Kyoto, Japan. Steganography and watermarking based on LLMs.

🛠 Skills

  • Programming: Python, PyTorch, C.
  • Languages: Chinese (Mandarin, native), English (CEFR C1), Japanese (studying, around JLPT N3–N2).