Affliation

Ph.D. student at OIST
Email: yuji.kanagawa [at] oist.jp, yuji.kngw.80s.revive [at] gmail.com

Education

2019 - 2021: Department of General Systems Studies, The University of Tokyo (M.S. degree)
- Recieved Ichiko Memorial Award
2017 - 2019: College of Arts and Science, The University of Tokyo (B.S. degree)
2015 - 2017: Human Sciences III, The University of Tokyo

Job Experiences

2023: Interned at stealth neurotech startup
2019: Interned at Preferred Networks, Tokyo

Research interests

I’m interested in the evolutionary simulation of virtual agents to get insights of real brain evolution. Particularly, reward function is of my main interest. From the age of early vertebrates (you can imagine fishes with eel-like mouth), reward function in our brains helped us survive in the world by providing learning signal to helpful events like food intake. I’m doing simulation where many (100~500 so far) agents learn from their innate reward functions and observing how they evolve.

I did some deep reinforcement learning stuff in my masters and undergrad.

Publications

Evolution of Rewards for Food and Motor Action by Simulating Birth and Death
- Paper
- Code
- Demo site
Learning Diverse Options via InfoMax Termination Critic
- Presented at NeurIPS DeepRL Workshop 2020
- Paper
- Code
Rogue-Gym: A New Challenge for Generalization in Reinforcement Learning
- Presented at IEEE Conference on Games 2019
- Paper
- Environment code
- Experiment code

Sub project?

For evolutionary simulation with many agents, I made 2D physics simulator that works on GPU. Check phyjax2d repo.

OSS Projects

I am a fan of open sources and open communities since I met the Rust community during my undergraduate. I have continuously contributed to some OSS projects, although I am not very active now. I received Google Open Source Peer Bonus for my contributions to PyO3.

See my Github account for more.

Kanji

金川裕司 (かながわゆうじ)

Yuji Kanagawa