Join our group at UTokyo!
blog · GitHub · benchmarks · email
I welcome motivated students and researchers interested in model evaluation, alignment, and safety, to join our group. We are still a small group, currently just me, 1 postdoc, 2 PhD students, 1 master's student, and 1 lab assistant. Some specific research topics include: irreducible error estimation, LLM contamination, benchmark design, model monitoring, auditing, scalable oversight, alignment, and reward hacking.
Please check our recent publications to get a sense of our research directions. If you have any questions about joining our group, feel free to contact me. I will attend ICLR 2026 (Rio de Janeiro) and will likely attend ICML 2026 (Seoul), so let me know if you'd like to meet and chat.
Ways to Join the Lab
I can supervise or mentor students or work with researchers through the following opportunities:Postdocs
I do not currently have funded postdoc openings at UTokyo. However, I'm happy to hear from self-funded candidates (e.g., JSPS or other fellowships) who share similar research interests. If you are planning to apply to such programs, please feel free to email me. I may have funded postdoc openings in the future, so let me know if you are interested so I can keep you informed if such opportunities arise.
Graduate students (PhD)
I supervise PhD students in Department of Complexity Science and Engineering (Kashiwa Campus) and Department of Computer Science (Hongo Campus) at the University of Tokyo.
If you are interested in applying, I would appreciate it if you could first review our recent papers at UTokyo to see whether our papers resonate with you. If you feel the vision and research directions of the lab are strongly aligned with your research interests, skills, and experience, please review the official guidelines at the links above and feel free to contact me with your CV, academic transcripts (for bachelor and master degrees), any previous papers you have published, and a few-page research proposal. Don't include a timeline in the proposal as it is usually less informative for pre-screening and is not needed. If you have questions before preparing a proposal, feel free to reach out.
If you are considering applying via the Embassy track of MEXT scholarship, please contact me only after you have passed the first screening stage. Please note that I do not accept students through the University track of MEXT.
I receive many emails and may not be able to reply to all of them. Apologies in advance.
Graduate students (Master's)
For Master's course, I supervise students in Department of Complexity Science and Engineering, and I co-supervise students with Prof. Masashi Sugiyama in Department of Computer Science.
Prospective master students should have a strong foundation in mathematics and coding:
- Mathematics. Students should already be very comfortable with statistics, linear algebra, and calculus. A concise refresher.
- Coding. Strong Python programming skills are essential. Depending on the research topic, experience with training/fine-tuning deep networks or LLMs and experience with LLM inference is required. Fluency with coding agents (such as Claude Code and Codex) is increasingly important, especially given our work on model cheating and reward hacking.
4th year undergraduates
I co-supervise undergraduate thesis students with Prof. Masashi Sugiyama from the Department of Information Science at the University of Tokyo.
Interns / Research Students
I usually don't accept interns or research students at the University of Tokyo, with a few exceptions (such as Embassy track of MEXT scholarship, opportunities for already admitted students who is waiting to enroll, or students with concrete ideas to extend or build upon our previous work).
However for 1st–3rd year undergraduates, I may have the chance to mentor students in Japanese universities through the RIKEN-AIP Undergraduate Research Program (PI: Prof. Masashi Sugiyama, Imperfect Information Learning Team, Nihonbashi Office). See Japanese official guide.
Some information about the Lab
- We are a sub-group within Machine Learning and Statistical Data Analysis Lab (mslab), and we work closely with Prof. Masashi Sugiyama's lab and Prof. Naoto Yokoya's lab.
- All lab meetings and seminars are conducted in English, so students from abroad can join us without any Japanese proficiency.
- For computing resources, we have access to Miyabi Supercomputer System, Wisteria Supercomputer System, mslab GPU servers, and occassionally cloud GPUs (e.g., Lambda Labs).
- We have lab seminars on Mondays and Wednesdays (each is around 2 hours). There are no mandatory lab hours outside of seminars.
- Day-to-day communication and 1-on-1 meetings are conducted in English or Japanese, depending on the preference of the student.
- I am the main supervisor for only a few students, so I typically have 1-on-1 meetings frequently, e.g., once a week, but this can vary depending on the student's needs and progress.
- Our labs are located in Kashiwa and Hongo. The Kashiwa campus offers a peaceful research environment with beautiful Japanese gardens and traditional tea houses nearby. The Hongo campus is in central Tokyo, providing easy access to city life.
- Currently all PhD students in our lab receive good financial support! However, I cannot make any promises since it may depend on my future grants/funding.
Lab photos and campus life
