Research Interests
Metacognition | Introspection | (Mechanistic) interpretability | NeuroAI
Education
M.S. in Computer Science
Adviser: Daisuke Kawahara
Tokyo, Japan
B.S. in Computer Science
GPA: 3.8/4. Dean's Award.
Tokyo, Japan
Work Experience
Research Assistant
Research on differences between the internal representations of the human brain and language models as part of a JST CREST project.
Tokyo, Japan
Research Contractor
Research on interpretability and defenses of adversarial attacks on robot foundation models.
Tokyo, Japan
Research and Development Center for Large Language Models, National Institute of Informatics (NII)
Technical Assistant
Research on bottom-up interpretability of training dynamics in large language models.
Tokyo, Japan
Publications
Conference
Bottom-Up Interpretability of Pretraining Dynamics via Loss Curve Decomposition
Koshiro Aoki, Masaru Isonuma, Max Müller-Eberstein, Yusuke Oda, Hirokazu Kiyomaru, Takashi Kodama, Chaoran Liu, Yohei Oseki, Yusuke Miyao, Daisuke Kawahara
Third Conference on Language Modeling (COLM 2026)
Workshop
Crosscoders Identify Shared or Specific Features between the Human Brain and Language Models
Koshiro Aoki, Itsuki Hamada, Naho Orita, Daisuke Kawahara, Hiromu Sakai
Mechanistic Interpretability Workshop @ ICML 2026
Koshiro Aoki, Ryota Takatsuki, Gouki Minegishi, Yusuke Haruki, Daisuke Kawahara
Trustworthy AI for Good (AI4GOOD) Workshop @ ICML 2026 Spotlight
Evaluating the Impact of SAE-based Language Steering on LLM Performance
Sebastian Zwirner, Wentao Hu, Koshiro Aoki, Daisuke Kawahara
EACL 2026 Student Research Workshop
Testing Simulation Theory in LLMs' Theory of Mind
Koshiro Aoki, Daisuke Kawahara
IJCNLP-AACL 2025 Student Research Workshop Oral
Domestic Journal
Mechanistic Interpretability: A New Trend in Interpretability Research
Koshiro Aoki, Ryota Takatsuki, Gouki Minegishi, Hakaze Cho
Transactions of the Japanese Society for Artificial Intelligence Best Paper Award
Honors and Awards
Trustworthy AI for Good (AI4GOOD) Workshop @ ICML 2026
The Japanese Society for Artificial Intelligence 40th Anniversary Commemorative Paper Award
Dean's Award for Excellence
School of Fundamental Science and Engineering, Waseda University
Young Researcher Encouragement Award
The 32nd Annual Meeting of the Association for Natural Language Processing
IJCNLP-AACL 2025 Student Research Workshop
The 20th Young Researcher Association for NLP Studies
The 31st Annual Meeting of the Association for Natural Language Processing