Research Scientist working on Vision & Language with Deep Learning

Research Scientist @ Image and Video AI Dept., LY Corp.

Kioi Tower 1-3 Kioicho, Chiyoda-ku, Tokyo, 102-8282, Japan

Project Researcher @ Hosei University

S603, 3-7-2, Kajino-cho, Koganei-shi, Tokyo 184-8584, Japan

Intelligent information processing laboratory

Researcher @ The University of Electro-Communications

Nishi 5 Building, 1-5-1, Chofugaoka, Chofu-shi, Tokyo 182-8585, Japan

Hara Laboratory

Biography

Shunsuke Kitada （北田俊輔 in Japanese) is a Research Scientist at Image and Video AI Dept., LY Corp. His research interest is now on computational advertising with a focus on automatic generation/evaluation and assistive technology for multi-modal ad creatives.

Previously, he got his Ph.D. in 2023 at major in applied informatics, graduate school of science and engineering, Hosei University under the supervision of Prof. Hitoshi Iyatomi. His project was to improve prediction performance and model interpretability through attention mechanisms from basic and applied research perspectives.

His personality traits can be summarized as follows:

❤️ Love research and development. He is enjoying research life, and he is currently focusing on three research topics simultaneously, e.g., natural language processing, medical image-based computer vision, computational advertising.
📝 Every day read and implement the SoTA models. He has released many re-implementations of models using mainly Chainer and PyTorch. Therefore, based on state-of-the-art cases, he can advise on deep learning-based product design.
😄 High technical communicativity. Summarize what he made and what he studied, and spread such information. This shows that he can input and output regularly.

The resume is available in PDF .

Background image borrowed from OpenAI ChatGPT illustration by Ruby Chen.

Interests

🤖 Natural Language Processing
💻 Computer Vision
🏥 Medical Image Processing
📃 Computational Advertising

Education

PhD in Engineering, 2023

Graduate School of Science and Engineering, Hosei University
MSc in Engineering, 2020

Graduate School of Science and Engineering, Hosei University
BSc in Engineering, 2018

Hosei University

📢 Recent News

All news »

[2026.08] Press Releases about Our FOUND Workshop at ECCV 2026

[2026.08] Our Presentations at YANS 2026

[2026.08] Our Presentations at MIRU 2026

[2026.07] Accepted our Workshop Proposal on Foundation Data to ECCV 2026

[2026.05] Accepted our papers to ACL2026 SRW

See all

📝 Journal Article

All Journal Articles »

Majority or Minority: Data Imbalance Learning Method for Named Entity Recognition

IEEE Access (Impact Factor: 3.4 in 2024; 1st place in Engineering & Computer Science (general) at Google Scholar Metrics)

Sota Nemoto, Shunsuke Kitada, Ph.D., Hitoshi Iyatomi

Majority or Minority: Data Imbalance Learning Method for Named Entity Recognition

人工知能学会誌博士論文集 2022.10～2023.9授与

人工知能学会誌の博士論文特集号に寄稿しました。

Shunsuke Kitada, Ph.D., Others

DM$^2$S$^2$: Deep Multi-Modal Sequence Sets with Hierarchical Modality Attention

IEEE Access (Impact Factor: 3.476 in 2021; 1st place in Engineering & Computer Science (general) at Google Scholar Metrics)

Shunsuke Kitada, Ph.D., Yuki Iwazaki, Riku Togashi, Hitoshi Iyatomi

DM$^2$S$^2$: Deep Multi-Modal Sequence Sets with Hierarchical Modality Attention

Making Attention Mechanisms More Robust and Interpretable with Virtual Adversarial Training

Springer Applied Intelligence (Impact Factor: 5.019 in 2021)

Shunsuke Kitada, Ph.D., Hitoshi Iyatomi

Making Attention Mechanisms More Robust and Interpretable with Virtual Adversarial Training

Ad Creative Discontinuation Prediction with Multi-Modal Multi-Task Neural Survival Networks

MDPI Appl. Sci. (Impact Factor: 2.838 in 2021; Top 10 ranking in Engineering & Computer Science (general) at Google Scholar Metrics)

Shunsuke Kitada, Ph.D., Hitoshi Iyatomi, Yoshifumi Seki

Ad Creative Discontinuation Prediction with Multi-Modal Multi-Task Neural Survival Networks

See all

📝 Conference Paper (Refereed)

All Conference Papers »

A11y-Compressor: A Framework for Enhancing the Efficiency of GUI Agent Observations through Visual Context Reconstruction and Redundancy Reduction

Proc. ACL SRW 2026 (Acceptance rate = 32.41%)

Michito Takeshita, Takuro Kawada, Takumi Ohashi, Shunsuke Kitada, Ph.D., Hitoshi Iyatomi

A11y-Compressor: A Framework for Enhancing the Efficiency of GUI Agent Observations through Visual Context Reconstruction and Redundancy Reduction

EnsemHalDet: Robust VLM Hallucination Detection via Ensemble of Internal State Detectors

Proc. ACL SRW 2026 (Acceptance rate = 32.41%; non-archival publication)

Ryuhei Miyazato, Shunsuke Kitada, Ph.D., Kei Harada

EnsemHalDet: Robust VLM Hallucination Detection via Ensemble of Internal State Detectors

TAUE: Training-free Noise Transplant and Cultivation Diffusion Model

Findings of CVPR2026 (Acceptance rate = 36.09%)

Daichi Nagai, Ryugo Morita, Shunsuke Kitada, Ph.D., Hitoshi Iyatomi

TAUE: Training-free Noise Transplant and Cultivation Diffusion Model

SciGA: A Comprehensive Dataset for Designing Graphical Abstracts in Academic Papers

Findings of CVPR2026 (Acceptance rate = 36.09%)

Takuro Kawada, Shunsuke Kitada, Ph.D., Sota Nemoto, Hitoshi Iyatomi

SciGA: A Comprehensive Dataset for Designing Graphical Abstracts in Academic Papers

Layout-Corrector: Alleviating Layout Sticking Phenomenon in Discrete Diffusion Model

Proc. of ECCV2024 (Acceptance rate = 27.9%)

Shoma Iwai, Atsuki Osanai, Shunsuke Kitada, Ph.D., Shinichiro Omachi

Layout-Corrector: Alleviating Layout Sticking Phenomenon in Discrete Diffusion Model

See all

📝 Preprint

All Preprints »

VASCAR: Content-Aware Layout Generation via Visual-Aware Self-Correction

Large language models (LLMs) have proven effective for layout generation due to their ability to produce structure-description …

Jiahao Zhang, Ryota Yoshihashi, Shunsuke Kitada, Ph.D., Atsuki Osanai, Yuta Nakashima

VASCAR: Content-Aware Layout Generation via Visual-Aware Self-Correction

Making Attention Mechanisms More Robust and Interpretable with Virtual Adversarial Training for Semi-Supervised Text Classification

Preprint version of the Springer Applied Intelligence paper “Making Attention Mechanisms More Robust and Interpretable with Virtual Adversarial Training”

Shunsuke Kitada, Ph.D., Hitoshi Iyatomi

Making Attention Mechanisms More Robust and Interpretable with Virtual Adversarial Training for Semi-Supervised Text Classification

Skin lesion classification with ensemble of squeeze-and-excitation networks and semi-supervised learning

Manuscript for the International Skin Imaging Collaboration 2018.

Shunsuke Kitada, Ph.D., Hitoshi Iyatomi

Skin lesion classification with ensemble of squeeze-and-excitation networks and semi-supervised learning

🎓 Dissertation

Improving Prediction Performance and Model Interpretability through Attention Mechanisms from Basic and Applied Research Perspectives

Ph.D. dissertation

Shunsuke Kitada, Ph.D.

Improving Prediction Performance and Model Interpretability through Attention Mechanisms from Basic and Applied Research Perspectives

🇯🇵 Domestic Conference / Presentation in Japan

All domestic conference »

西山天, 川田拓朗, 北田俊輔, 永井大地, 彌冨仁 (2026). Activation Steeringにおける文章崩壊の抑制に向けた初期検討. YANS 2026.

Poster

陳冠帆, 川田拓朗, 守田竜梧, 北田俊輔, 彌冨仁 (2026). Where Do Vision-Language Models Preserve and Use Depth Information?. YANS 2026.

Poster

Shunsuke Kitada, Ph.D., Shoma Iwai, Ryota Yoshihashi, Atsuki Osanai (2026). DesignCorrection-R1: Learning to Reason for Graphic Design Correction. MIRU 2026.

PDF Poster MIRU

Takuro Kawada, Shunsuke Kitada, Ph.D., Hitoshi Iyatomi (2026). Generation and Evaluation of Editable Graphical Abstracts for Academic Papers. MIRU 2026.

PDF Poster MIRU

竹下理斗, 川田拓朗, 大橋巧, 北田俊輔, 彌冨仁 (2026). Compressed-a11y: 視覚的文脈の再構成と冗長性削減による GUI エージェント観測の効率化. NLP 2026.

PDF Poster

菊谷幹, 北田俊輔, 原聡 (2026). MLP の重みを反映した Sparse Autoencoder の初期化手法の提案. NLP 2026.

PDF Poster

川田拓朗, 北田俊輔, 彌冨仁 (2026). SciGA-Vec: 学術論文におけるベクタ画像形式の Graphical Abstract データセット. NLP 2026.

PDF Poster

北田俊輔, 原聡 (2026). ステアリングベクトルは日本語LLMを堅牢に制御できるか？. NLP 2026.

PDF Poster

永井大地, 守田竜梧, 北田俊輔, 彌冨仁 (2025). 潜在拡散モデルでのレイアウト指定生成に向けたアテンションに基づく初期ノイズ最適化. IBIS 2025.

Poster

竹下理斗, 川田拓朗, 大橋巧, 北田俊輔, 彌冨仁 (2025). ローカル LLM を用いた AI エージェントの現状と課題. YANS 2025.

Poster

See all

🎙️ Recent & Upcoming Talks

All recent and upcoming talks »

[Journal Club] PSDesigner: Automated Graphic Design with a Human-Like Creative Workflow (CVPR 2026)

CVPR 2026 論文読み会で、生成後も要素を編集できるデザイン生成手法を紹介

Jul 18, 2026 1:45 PM — 2:10 PM DeNA / IRIAM

Shunsuke Kitada, Ph.D.

[Journal Club] PSDesigner: Automated Graphic Design with a Human-Like Creative Workflow (CVPR 2026)

[Journal Club] Natural Language Autoencoders Produce Unsupervised Explanations of LLM Activations

大規模言語モデルが入力に反応する理由を自然言語で説明する手法を紹介

Jun 5, 2026 12:00 PM — 12:00 PM Online

Shunsuke Kitada, Ph.D.

[Journal Club] Natural Language Autoencoders Produce Unsupervised Explanations of LLM Activations

その LLM 制御、本当に信頼できますか？ / Can We Reliably Control LLMs?

NLP2026 参加報告会で、大規模言語モデルの応答を調整する技術の信頼性と限界を発表

Apr 21, 2026 7:45 PM — 7:55 PM Sansan株式会社本社

Shunsuke Kitada, Ph.D.

その LLM 制御、本当に信頼できますか？ / Can We Reliably Control LLMs?

[Journal Club] The Landscape of Agentic Reinforcement Learning for LLMs: A Survey

LLM が試行錯誤から行動を学ぶ強化学習手法と課題を紹介

Apr 1, 2026 12:00 PM — 12:00 PM Online

Shunsuke Kitada, Ph.D.

[Journal Club] The Landscape of Agentic Reinforcement Learning for LLMs: A Survey

[Journal Club] LLM の内部を操るベクトル？

大規模言語モデル内部の情報を調整して、応答の傾向を制御する手法を紹介

Feb 17, 2026 12:00 PM — 12:00 PM Online

Shunsuke Kitada, Ph.D.

See all events

🏆 Awards & Grants

[2023.01] Got Hosei University Research Grant 2022

[2022.07] Got JASSO Scholarship for Top 30% Excellent Graduate School Students

[2021.12] Got Hosei University Research Grant 2021

[2021.05] Got Grant-in-Aid for Japan Society for the Promotion of Science (JSPS) Fellows

[2021.04] Adopted as a JSPS DC2

See all

📌 Featured Post

All posts »

Shunsuke Kitada, Ph.D.

Last updated on Aug 29, 2026 4 min read DC

日本学術振興会特別研究員 (DC2) 申請書・学振

補欠繰り上げで 2021 年度学振特別研究員 DC2 に採用していただいた際の申請書です。

📝 Posts & Articles

About Awards and Interviews (Almost In 🇯🇵)

FOUND: Foundation Data for Industrial Tech Transfer @ ECCV 2026

The 2nd FOUND Workshop at ECCV 2026 focuses on Foundation Data for Industrial Tech Transfer, with emphasis on domain-grounded data, robustness under domain shift, Physical AI, and World Models.

Yoshihiro Fukuhara, Hirokatsu Kataoka, Shunsuke Kitada, Ph.D., Püren Güler, Risa Shinoda, Shashanka Venkataramanan, Jan-Nico Zaech, Keisuke Tateno, Koichiro Yamaguchi, Masaki Inaba, Hina Otake

Last updated on Aug 29, 2026 2 min read Workshop, ECCV, ECCV2026

A Coding-Agent-Friendly Environment Is Friendly to Humans Too: ghq × gwq × fzf

Coding agents thrive on parallelism, and by unifying repository clones and git worktrees under a single ghq root and navigating them with fzf, you can create a workflow that’s faster, cleaner, and friendlier for both humans and AI.

Last updated on Aug 29, 2026 5 min read Tech

A Coding-Agent-Friendly Environment Is Friendly to Humans Too: ghq × gwq × fzf

ICCV2025 Conference Participation Report

At ICCV 2025, our team from LY Corp. successfully organised and participated in a workshop and paper presentation, gaining deep insights into the shift of computer‐vision research toward structured, controllable design generation and fast, reliable foundation-model technologies.

Shunsuke Kitada, Ph.D.

Last updated on Aug 29, 2026 5 min read Tech Blog, Conference Report

ICCV2025 Conference Participation Report

On Python Type Hints that Save Everything!

This article introduces the background, concepts, practical usage, and benefits of Python type hints, explaining how they enhance readability, maintainability, and development efficiency while preserving the flexibility of dynamic typing.

Shunsuke Kitada, Ph.D.

Last updated on Aug 29, 2026 7 min read

Testable Dotfiles Management With Chezmoi

This article explains an approach to dotfiles management that emphasizes testability, using the author’s dotfiles repository shunk031/dotfiles as a case study.

Shunsuke Kitada, Ph.D.

Last updated on Aug 29, 2026 9 min read Tech

Testable Dotfiles Management With Chezmoi

See all posts

👨‍💻 Selected Experiences

All experiences can be found on my LinkedIn.

Research Scientist

LY Corporation

October 2023 – Present Kioi Tower 1-3 Kioicho, Chiyoda-ku, Tokyo, 102-8282, Japan

Transferred due to merge between LINE Corp. and Yahoo! Japan Corp.

Research Scientist

LINE Corporation

April 2023 – September 2023 Yotsuya Tower 23rd FL., 1-6-1 Yotsuya, Shinjuku-ku, Tokyo, 160-0004, Japan

Lecturer

DAY1 COMPANY Inc.

June 2023 – June 2026 Teheran-ro, Gangnam-gu, Seoul, Korea

Taught lectures on image-generating AI at Coloso, an online education service
The lectures can be found at the following URL: https://bit.ly/KitadaXColosoJP1

Teaching Assistant / Mentor

NVIDIA Japan

January 2016 – April 2021

Assisted in several workshops in Tokyo relating to deep learning and CUDA:

Served as a mentor, providing guidance on large-scale GPU distributed training using HPC:

Fundamental Information Technology Engineer

Information-technology Promotion Agency, Japan

June 2016 – Present

Fundamental Information Technology Engineer Examination is a yardstick for measuring IT knowledge and skills as a team member by asking a range of questions about algorithm, network, database, information security, practical programming, etc.