Hirohane Takagi

Hirohane Takagi

高木 洋羽

About

I’m a PhD student at the University of Tokyo working on machine learning and artificial intelligence. My research focuses on trustworthy inference-time control of foundation models and the statistical theory behind it. Alongside my research, I’ve been working professionally for several years, building ML applications on large-scale data infrastructure in close collaboration with clients.

Interests
  • Generative models
  • Statistical Machine Learning
  • Large Language Models
  • Explainable AI, Interpretability
Education

  • PhD Engineering

    The University of Tokyo

  • MSc Computer Science

    The University of Tokyo

  • BEng Applied Mathematics

    The University of Tokyo

Refereed Conference Papers

  1. Hirohane Takagi, Masashi Sugiyama. (2026, December). Asymptotically Exact Negative Guidance of Diffusion Models via Positive-Unlabeled Learning. In Proceedings of the 40th Conference on Neural Information Processing Systems (NeurIPS 2026).
    Poster, arXiv, Code

  2. Hirohane Takagi, Atsushi Nitanda. (2026, April). Alternating Diffusion for Proximal Sampling with Zeroth Order Queries. The 14th International Conference on Learning Representations.
    Paper, Poster, arXiv, Code

  3. Hirohane Takagi*, Gouki Minegishi*, Shota Kizawa, Issey Sukeda, Hitomi Yanaka. (2025, December). Interpreting Multi-Attribute Confounding through Numerical Attributes in Large Language Models. In Proceedings of International Joint Conference on Natural Language Processing & Asia-Pacific Chapter of the Association for Computational Linguistics (pp. 1098-1115).
    Paper, Poster, arXiv, Code
    Also featured in the AIP Symposium poster

  4. Hirohane Takagi, Shoji Moriya, Takuma Sato, Manabu Nagao, Keita Higuchi. (2025, March). A Framework for Efficient Development and Debugging of Role-Playing Agents with Large Language Models. In Proceedings of the 30th International Conference on Intelligent User Interfaces (pp. 70-88).
    Paper, PFN Tech Blog (Japanese)

Domestic Conference Presentations (国内学会発表)

  1. 高木洋羽, 杉山将, 正・ラベルなし学習に基づく漸近的厳密性を有する拡散モデルの負方向誘導, 第29回情報論的学習理論ワークショップ (IBIS2026), (2026.10).
    プログラム (1-180)

  2. 高木洋羽, 佐藤大地, 長谷川貴巨, 大福泰樹, 武田峻悟, 山内有倫, 高橋孝樹, 国藤靖彦, 機械学習と波形補正を用いたスマートメーター計測値からPV出力と実需要への分離推定, 令和7年電気学会全国大会, (2025.3).
    要旨(6-102)，松尾研究所と東京電力パワーグリッドの共同研究⚡️
    Blog記事 @ 松尾研究所, Zenn

  3. 峰岸剛基*, 高木洋羽*, 木澤翔太*, 助田一晟, 谷中瞳, 大規模言語モデルにおいて数値属性間で共有されるスケーリングベクトルの解析とその応用, 言語処理学会第31回年次大会, (2025.3).
    予稿，プログラム (A6-2) (口頭発表), 若手奨励賞 (20件/487件)🎉

  4. 高木洋羽, 守屋彰二, 佐藤拓真, 永尾学, 樋口啓太, 大規模言語モデルを用いたロールプレイエージェントの効率的な開発と動作検証のためのフレームワーク, インタラクション2025, (2025.3).
    予稿 (査読あり), 登壇発表 (Youtube), Webカタログ (17)

Other Activities

Academic Services

  • Reviewer: NeurIPS 2026

Book Reviewing

  • Joined the volunteer review of Deep Learning from Scratch Series, which first introduced me to Python and deep learning, as a small way of giving something back.
    • Vol. 5: Generative Models — 『ゼロから作るDeep Learning ❺―生成モデル編』
    • Vol. 6: LLMs — 『ゼロから作るDeep Learning❻—LLM編』)

GDG Tokyo / GDGoC