About
My research mainly focuses on the development, application, and analysis of large vision-language models (LVLMs).
I am a third-year Ph.D. student at the Graduate School of Information Sciences, Tohoku University, advised by Prof. Jun Suzuki.
Affiliation
- Tohoku University, Graduate School of Information Sciences — Ph.D. Student, Fundamental Artificial Intelligence Laboratory
- University of Cambridge, Department of Computer Science and Technology — Visiting Researcher (Departmental Visiting Student)
Research Interests
News
-
2026.10
Gave a talk titled “Large Vision-Language Models: How They See, Pass, and Answer” at the NLIP Seminar Series, University of Cambridge.
-
2026.10
Our paper “It's Not RoPE That Creates Sinks: The Role of Self-Concentration and Value-Non-Mixing in Attention” was accepted to EMNLP 2026.
-
2026.09
Started a research visit at the University of Cambridge (Department of Computer Science and Technology) as a visiting researcher, hosted by Prof. Andreas Vlachos.
-
2026.09
Our book “マルチモーダルモデル入門” (Introduction to Multimodal Models) was published by Gijutsu-Hyoron.
-
2026.09
Our paper “MOMIJI: A Large-Scale Japanese Interleaved Image-Text Dataset” was accepted to the Journal of Natural Language Processing.
-
2026.09
Our paper “Read or Ignore? A Unified Benchmark for Typographic-Attack Robustness and Text Recognition in Vision-Language Models” was accepted to ECCV 2026.
Education
-
Apr 2024 – Mar 2027 (expected)
Tohoku University (Ph.D.)
Ph.D. in Information Science, Fundamental Artificial Intelligence, Department of System Information Sciences, Graduate School of Information Sciences
Advisor: Prof. Jun Suzuki
-
Apr 2022 – Mar 2024
Tohoku University (M.S.)
M.S. in Information Science, Fundamental Artificial Intelligence, Department of System Information Sciences, Graduate School of Information Sciences
Advisor: Prof. Jun Suzuki
-
Apr 2018 – Mar 2022
Tohoku University (B.E.)
B.E. in Information Science, Department of Electrical, Information and Physics Engineering, School of Engineering
Experience
-
Sep 2026 – Present
Visiting Researcher (Departmental Visiting Student) @ University of Cambridge
Department of Computer Science and Technology
Conducting research on Vision & Language and NLP, hosted by Prof. Andreas Vlachos and working closely with Diana Galván Sosa.
-
Apr 2024 – Oct 2026
Machine Learning Research Engineer Intern / Contractor @ Turing Inc.
Fundamental AI Group
Contributed to training large-scale vision-language models (LVLMs), building a large-scale Japanese vision-language pretraining dataset, and constructing a Japanese visual instruction-tuning dataset.
-
Apr 2023 – Mar 2027 (expected)
Research Assistant @ Center for Data-driven Science and Artificial Intelligence, Tohoku University
Supported research activities through experimental setup, documentation, data maintenance, and related project operations.
-
Mar 2023 – Dec 2023
Part-time Machine Learning Engineer @ Future Inc.
Strategic AI Group
Improved OCR models and developed a model for converting foreign personal names into Japanese readings.
Publications
Underline: myself, *: equal contribution
Journal Papers (refereed)3
-
JNLP 2026
MOMIJI: A Large-Scale Japanese Interleaved Image-Text Dataset
Journal of Natural Language Processing, Volume 33, Issue 4 (General Paper; scheduled for publication December 15, 2026). Accepted September 18, 2026.
-
NGC 2026
Instruction-Following Evaluation of Large Vision-Language Models
New Generation Computing, Volume 44, Article 7, January 2026.
-
Advanced Robotics 2023
Aoba_v3 bot: a multimodal chatbot system combining rules and various response generation models
Advanced Robotics, Volume 37, Issue 21, pp.1392-1405, August 2023.
International Conferences (refereed)9
-
-
-
-
AAAI 2026
STRIDE-QA: Visual Question Answering Dataset for Spatiotemporal Reasoning in Urban Driving Scenes
Proceedings of the AAAI Conference on Artificial Intelligence (AAAI 2026), Singapore, January 2026.
-
-
COLING 2025
Evaluating Model Alignment with Human Perception: A Study on Shitsukan in LLMs and LVLMs
Proceedings of the 31st International Conference on Computational Linguistics (COLING 2025), Abu Dhabi, UAE, January 2025.
-
NAACL SRW 2024
Instruction-Following Evaluation for Large Vision-Language Models
Non-archival submission for the 2024 Conference of the North American Chapter of the Association for Computational Linguistics Student Research Workshop (NAACL SRW 2024), Mexico City, Mexico, June 2024.
-
-
EACL SRW 2024
Exploring Large Language Models' Understanding of Shitsukan
Non-archival submission for the 18th Conference of the European Chapter of the Association for Computational Linguistics Student Research Workshop (EACL SRW 2024), Malta, March 2024.
Domestic Conferences / Symposium / Workshop14
-
NLP 2026
MOMIJI: 日本語大規模
インターリーブ 視覚言語 データセット 言語処理学会第32回年次大会 (NLP 2026), Utsunomiya, Tochigi, pp.717-722, March 2026.
-
-
-
-
FT-LLM 2026
チーム dentaku FT-LLM 2026 (数学タスク) システム解説
The Second Workshop on Fine-Tuning and Evaluation of Large Language Models (FT-LLM 2026), Utsunomiya, Tochigi, March 2026.
-
YANS 2025
バッチ単位
収束型 事前学習：子どもの言語発達に着想を得た一歩ずつの学習 NLP若手の会第20回シンポジウム (YANS 2025), Hamamatsu, Shizuoka, August 2025.
Encouragement Award -
-
-
NLP 2025
日本語
バイト 符号化 マスク 言語 モデルの開発と分析 言語処理学会第31回年次大会 (NLP 2025), Nagasaki, pp.3356-3361, March 2025.
-
-
-
YANS 2023
生成応答に含まれる事実に基づかない情報の自動検出の試み
NLP若手の会第18回シンポジウム (YANS 2023), Tokyo, August 2023.
Encouragement Award CyberAgent Award -
-
Book1
No publications match your search.
Projects
Software & Tools
-
Jul 2026
-
Jul 2026
Cosense Zenn Style
A collection of unofficial UserCSS, UserScripts and Tampermonkey scripts that make Cosense (formerly Scrapbox) look like Zenn. It also supports importing and exporting Markdown files, switching between light and dark mode, and embedding GitHub files and X posts as cards.
Local Data Studio
Model & Dataset Development
-
Ongoing
Building a Large-Scale Collection of SFT Datasets Translated into Japanese
Independent project (FineVision_JA).
-
Sep 2024
Building SFT Datasets for Japanese LVLMs
Outcome from my internship at Turing Inc.: The Cauldron JA and Wikipedia Vision JA.
-
Jun 2024
Building a Japanese BERT Model for Long-Sequence Processing
Collaborated on the development and distributed pretraining of a Japanese BERT model supporting long-sequence inputs, using Megatron-LM on the TSUBAME4.0 supercomputer.
Knowledge Sharing
-
Ongoing
Sharing Notes, Articles and Slides
-
Nov 2024
Sharing Slides Summarizing Insights on LVLMs as of 2024
Independent project: “Large Vision Language Model (LVLM) に関する最新知見まとめ (Part 1)”
Awards
-
Mar 2026
1st Place in the Open Division and 2nd Place Overall, FT-LLM2026 Math Task
-
Mar 2024
Committee Special Award, 言語処理学会第30回年次大会 (NLP 2024)
(26/567 ≈ 4.59%)
-
Dec 2022
Excellence Award, 2nd Place, Open Track, 5th Dialogue System Live Competition
Talks
-
Oct 2026
Large Vision-Language Models: How They See, Pass, and Answer
NLIP Seminar Series, Department of Computer Science and Technology, University of Cambridge
-
Aug 2025
Words or Vision: Do Vision-Language Models Have Blind Faith in Text? (CVPR 2025)
Activities
Research Fellowship / Scholarship
-
Sep 2024 – Mar 2027 (expected)
BOOST Program for Next-Generation AI Areas
※ BOOST: Broadening Opportunities for Outstanding young researchers and doctoral students in STrategic areas
Research Fellow (Subsidized Project by the JST (Japan Science and Technology Agency))
-
Apr 2020 – Mar 2025
WISE Program for AI Electronics (AIE)
Research Fellow (Subsidized Project by the Ministry of Education, Culture, Sports, Science and Technology)
-
Apr 2022 – Mar 2024
Japan Student Services Organization (JASSO) Scholarship
Scholarship. Half Repayment Exemption due to outstanding achievements.
Teaching Assistant
-
Oct 2023 – Feb 2024
Natural Language Processing
Graduate-level course, Graduate School of Information Sciences, Tohoku University
-
Oct 2022 – Dec 2023
Introduction to Programming for Artificial Intelligence Technology
General education course, Tohoku University