About me
I am an applied math PhD candidate at the University of Tokyo, supervised by Professor Tomonari Sei and Professor Takeru Matsuda. My research interests include statistics, copulas, LLMs, neuroscience, and AI for healthcare.
My work explores how dependence structure in data and worlds should be modeled and incorporated into AI systems. In my doctoral research, I have developed dependence modeling methods based on the minimum information principle, extending from bivariate copulas to Markov models and graphical models.
π₯ News
- Aug. 2026 π Our paper was accepted in Journal of Multivariate Analysis!
- Feb. 2026 π Our paper was accepted in Neural Computation!
- Dec. 2025 π Our paper was accepted in ACM Multimedia Asia 2025 Workshop!
- Oct. 2025 π Two papers were accepted in IJCNLP-AACL 2025 Main conference!
- Aug. 2025 π Our paper was accepted in SIGGRAPH 2025 Posters!
- Oct. 2024 π Our paper was accepted in Statistical Papers!
- Oct. 2024 π Our paper was accepted in Statistics & Probability Letters!
- Oct. 2024 π My paper was accepted in NeurIPS Workshop 2024 AIM-FM!
- Sep. 2024 π Our paper was accepted in ACM SenSys 2024!
- Apr. 2024 π Our paper was accepted in Artificial Intelligence for Health.
- Dec. 2023 π My research was accepted as KAKUSEI Project by AIST.
- Oct. 2023 π Our paper was accepted in NeurIPS Workshop 2023 DGM4H ! See you in New Orleans !
- Jan. 2023 π Our paper was accepted in European Journal of Operational Research !
- Nov. 2022 π Our paper was accepted in ACM SenSys 2022 !
- Feb. 2022 πͺ Founded EQUES Inc. as CTO !
Research Interests
- Dependence modeling: copula theory; information-theoretic statistical models
- Graph network: statistical graphical models; network analysis; circular data
- AI for healthcare: biomedical data analysis; clinical AI applications
π Medical/Pharma LLM Projects
- JMedLoRA is the first Japanese medical LLM finetuned from Llama2. A work at UTokyo Hospital.
- MedSwallow is the second Japanese medical LLM finetuned from Swallow-70b-instruct-hf. (work at UTokyo Hospital).
- Japanese Medical Language Model Evaluation Harness is the evaluation codes and leaderboard for medical LLMs. (work at AIST KAKUSEI)
- JPharmatron-7B is the first Japanese pharmaceutical domain specific LLM, developed along with new benchamrks JPharmaBench. Accepted at IJCNLP-AACL 2025 Main. (work at EQUES Inc.)
π Statistics Projects
- Minimum Information Copula is the library that numerically calculates density function of minimum information copulas, defined as the most natural copula under predetermined constraints.
- Minimum information Markov model is a brand new Markov model that models the marginal (stationary) distribution and the dependence function seperately. Accepted in JMA.
- Sparse Torus Graph introduces a high-dimensional circular model with regularization to estimate dependence structure behind EEG phase, i.e., phase-based connectivity. Accepted in Neural Computation.
π° Fundings & Grants
- GENIACοΌas CTO of EQUES Inc., 2024οΌ
- RIKEN CBS Junior Research Associate [JPY 7.8M + 1.5M]
- JSPS DC2 (selected and declined due to CTO position)
- IST-RA, funded by UTokyo IST.
- AIST KAKUSEI, funded by AIST [JPY 3M]
- JEES Softbank AI scholarship [JPY 1M]
π Awards
2025
- π© ICSDS 2025 Student Award
- π KAKUSEI Award, AIST.
2024
- π₯ 1st Place, CBS Retreat 2024 - Poster Theory/Tech Award, RIKEN CBS
- π Best Presentation Award, Japan Society of Statistics
- π₯ 1st place, Sports Data Science Competition (MLB division), Japan Society of Statistics
2023
- π ISI Tokyo Memorial Award, Japan Society of Statistics
- π₯ Student Best Presentation Award, Japan Society of Statistics
γ2021
- π Finalist, Fin SUM Business Contest 2021
- π Yahoo Prize, Civictech Challenge Cup 2021 U-22
- π₯ 1st place, AI/SUM Business Contest 2020
- π₯ 2nd place, UTokyo MDS Data Analysis Contest 2019
- π₯ Bronze Award, Chemistry Grand Prix 2016
- π₯ 3rd place, All Japan Diabolo Competition 2015