Lab image

Takagi Laboratory

Department of Computer Science
School of Informatics
Nagoya Institute of Technology
Intelligent Information Systems

The Takagi Laboratory is a newly established research group founded in 2025. Our research aims to uncover the interface between human intellectual activities and rapidly advancing machine learning models.

Humans constantly engage—often unconsciously—in acts of “thinking” and “feeling.” To what extent can computers reproduce and even extend these processes? Addressing this question requires a simultaneous exploration of the internal mechanisms of state-of-the-art foundation models and the information-processing principles generated by human brain activity.

In this sense, our research seeks to elucidate human cognition through a deep understanding of machine learning models, while at the same time leveraging the rich dynamics of the brain to guide machine learning toward a new stage of development. Just as humanity has long pursued the mysteries of the universe and life itself, we aim to gradually unravel the mysteries of intelligence and learning.

Specifically, our research topics include:

  • Analysis of internal representations in large language models and diffusion models
  • Brain information analysis using fMRI, EEG, and physiological sensing data
  • Model interpretation and causal validation based on cognitive theories
  • Dialogue and creative-support systems for human–AI collaboration
  • Robotic systems that flexibly understand human intentions

Through these themes, we develop mathematical models and analytical methods that bridge machine learning and systems neuroscience. We place particular emphasis on the visualization and interpretability of machine learning models, promoting interdisciplinary research that spans linguistics, neuroscience, and information science to decode the “inner thoughts” of both humans and machines.

We are actively seeking master’s and doctoral students, postdoctoral researchers, and industry researchers to join collaborative projects. The laboratory emphasizes a flat and open organizational structure and provides an environment that fosters active discussion with researchers from universities, companies, and research institutes both in Japan and abroad.

Scholarships and research assistant positions are available, and members have access to cutting-edge GPU clusters and biosignal measurement equipment, including brain-imaging and physiological sensing systems, enabling the realization of ambitious and innovative ideas.

If you are interested in research that integrates human intelligence and machine learning to create new forms of intelligence, please feel free to contact us.

News

Aug 2026 Takagi will give an invited lecture at the Neuro2026 citizen participation event and symposium.
Apr 2026 Our paper on interpreting brain activity using large language models (paper) has been accepted to ICLR 2026.
Feb 2026 Our laboratory’s research was featured in Nikkei Science (February 2026 issue), with Takagi contributing through an interview.
Jan 2026 Takagi will give an invited talk at the “Physical AI” event hosted by Accenture and Kyoto University.
Jan 2026 Takagi has been selected as a principal investigator for a JST BOOST project.
Dec 2025 Research from our laboratory was covered by the German daily newspaper Der Tagesspiegel on December 31, 2025. (URL)。
Dec 2025 Our laboratory’s research was featured in the Chunichi Shimbun (morning edition, December 19, 2025, Page 3, Social Affairs section).
Dec 2025 Takagi will give an invited talk at the 39th Optical Communication Systems Symposium.
Dec 2025 A paper exploring the correspondence between deep learning models and human brain activity in response to music has been published in Nature Communications URL. This study was conducted in collaboration with Google DeepMind and was featured in Google Research’s official YouTube series, Field Notes URL.
Nov 2025 A paper exploring the correspondence between deep learning models and human brain activity in response to multimodal stimuli has been published in Nature Communications URL. This paper has also been selected as an Editor’s Highlight by the editorial team of Nature Communications URL.
Nov 2025 Two papers have been accepted at EMNLP 2025, a top-tier conference in natural language processing. A paper on the principles underlying supervised fine-tuning (SFT) arXiv, has been accepted to the Main Conference. Another paper on the learning dynamics of large language models using sparse autoencoders arXiv has been accepted to the Findings track.
Nov 2025 Takagi will give an invited talk at the 15th CiNet Symposium.
Nov 2025 Takagi will give a tutorial (on the first day) and an invited talk at the Neuro-LLM Workshop at the International Conference on Neural Information Processing 2025 (ICONIP 2025).
Oct 2025 Takagi will give an invited talk at the 2nd Online Seminar of the Kyushu Chapter of the Acoustical Society of Japan.
Oct 2025 Takagi has been selected as a principal co-investigator for a JST CREST project (Principal Investigator: Prof. Nishida, Osaka University).