TwitterGoogle ScholarLinkedIn

dematsunaga [at] ai [dot] kaist [dot] ac [dot] kr

Daiki E. Matsunaga

I am a PhD student at the Artificial Intelligence and Probabilistic Reasoning Lab (AIPR Lab) at KAIST Graduate School of AI.

I am very fortunate to be advised by Professor Kee-Eung Kim. 

My research goal is to better understand intelligence from the perspective of agent incentives and multi-agent interactions which drive complex social behaviors. To do this, I primarily conduct fundamental research in Multi-Agent Reinforcement Learning (MARL), which is a standard mathematical formulation for sequential multi-agent decision-making under uncertainty.

In the long-term, my goal is to help build agents with meta-cognitive and moral capabilities. I believe that rather than trying to engineer these traits explicitly, they can emerge from two ingredients in a multi-agent system: (1)  a sufficiently flexible and adaptive policy for each agent and (2) the right evolutionary and incentive pressures such that these traits are necessary and/or beneficial for survival.  

Education

2020. 09 - Current: MS-PhD, Graduate School of AI, KAIST, Seoul, South Korea (Advisor: Kee-Eung Kim)

2012. 09 - 2016. 09: BA, School of Political Science and Economics, Waseda University, Tokyo, Japan

Experience

2019. 04 - 2020. 08: Data Scientist at IBM Japan, Tokyo, Japan

2017. 04 - 2019. 03: IT Specialist/Consultant at IBM Japan, Tokyo, Japan

Publications (Conference)

* indicates equal contribution

[C2]  Stitching Sub-Trajectories with Conditional Diffusion Model for Goal-Conditioned Offline RL
 [Arxiv] [Video] [Code]


[C1]  AlberDICE: Addressing Out-Of-Distribution Joint Actions in Offline Multi-Agent RL via Alternating Stationary Distribution Correction Estimation

[OpenReview] [Arxiv] [Code]


Publications (Workshops)

* indicates equal contribution


[W1] Exploring Graph Neural Networks for Stock Market Predictions with Rolling Window Analysis
[Arxiv]

Reviewer

Teaching