Woojun Kim
[CV] [Google scholar] [Github]
Postdoctoral Fellow (Advised by Prof. Katia Sycara)
Advanced Agent-Robotics Technology Lab
Robotics Institute, School of Computer Science, Carnegie Mellon University
Contact -> woojunk [at] andrew [dot] cmu [dot] edu & kwj123123 [at] gmail [dot] com
Research Interests
My research currently focuses on multi-agent reinforcement learning (MARL). Specifically, I'm interested in developing MARL algorithms that achieve high coordination among multiple agents and consider practical issues such as scalability, robustness, and generalization, and their applications to real-world problems such as autonomous driving and robotics. My research on MARL can be organized into three intertwined thrusts:
(1) Correlated exploration method in MARL;
(2) Learning to communicate among multiple agents in MARL; and
(3) Development of scalable and robust MARL algorithms.
Besides the research on MARL, I'm also interested in {single-agent, multi-task, offline} reinforcement learning and other broad topics in machine learning such as imitation learning, transfer learning, and active learning.
Recent News
(Jan. 2024) One paper is accepted to ICLR 2024 as a spotlight presentation
(Oct. 2023) I joined CMU as a postdoc!
(Sep. 2023) Two papers and one workshop paper are accepted to NeurIPS 2023.
(Apr. 2023) Two papers are accepted to ICML 2023.
(Jan. 2023) Two papers and one workshop paper are accepted to AAMAS 2023.
(Dec. 2022) I successfully defended my Ph.D. thesis!
Employment
(2023.10 ~ Present) Postdoctoral Fellow in the Robotics Institute, Carnegie Mellon University, Pittsburgh, PA, USA (Advised by Prof. Katia Sycara)
(2023.03 ~ 2023.09) Postdoctoral Researcher in Electrical Engineering, KAIST, Daejeon, Korea (Advised by Prof. Youngchul Sung)
Education
(2018.03~2023.02) Ph.D. in Electrical Engineering, KAIST, Daejeon, Korea (Advised by Prof. Youngchul Sung)
(2016.03~2018.02) M.S. in Electrical Engineering, KAIST, Daejeon, Korea (Advised by Prof. Youngchul Sung)
(2010.03~2016.02) B.S. in Electrical Engineering, Korea University, Seoul, Korea (Including 2 years of military service)
Publications
(C: conference / P: preprint / W: workshop / U: under-review)
* denotes equal contribution or co-advised
Journal and Conference Publications
[C10] Decision ConvFormer: Local Filtering in MetaFormer is Sufficient for Decision Making [pdf] [project]
Jeonghye Kim, Suyoung Lee, Woojun Kim*(co-advised), Youngchul Sung*
International Conference on Learning Representation (ICLR), 2024
Spotlight Presentation (Top 5%)
[C9] Sample-Efficient and Safe Reinforcement Learning via Reset Deep Ensemble Agents [pdf]
Woojun Kim*, YongJae Shin*, Jongeui Park, Youngchul Sung
Conference on Neural Information Processsing Systems (NeurIPS) 2023 (acceptance rate: 26.1%)
[C8] Domain Adaptive Imitation Learning with Visual Observation [pdf]
Sungho Choi, Seungyul Han, Woojun Kim, Jongseong Chae, Whiyoung Jung, Youngchul Sung
Conference on Neural Information Processsing Systems (NeurIPS) 2023 (acceptance rate: 26.1%)
[C7] LESSON: Learning to Integrate Exploration Strategies for Reinforcement Learning via an Option Framework [pdf] [code]
Woojun Kim*, Jeonghye Kim*, Youngchul Sung
International Conference on Machine Learning (ICML) 2023 (acceptance rate: 1827/6538=27.9%)
[C6] An Adaptive Entropy-Regularization Framework for Multi-Agent Reinforcement Learning [pdf]
Woojun Kim, Youngchul Sung
International Conference on Machine Learning (ICML) 2023 (acceptance rate: 1827/6538=27.9%)
The preliminary version was presented at ICML 2022 workshop (DARL)
[C5] Parameter Sharing with Network Pruning for Multi-Agent Deep Reinforcement Learning [pdf]
Woojun Kim, Youngchul Sung
International Conference on Autonomous Agents and Multiagent Systems (AAMAS) 2023 (acceptance rate: 237/1015=23.3%)
[C4] A Variational Approach to Mutual Information-Based Coordination for Multi-Agent Reinforcement Learning [pdf]
Woojun Kim, Whiyoung Jung, Myungsik Cho, Youngchul Sung
International Conference on Autonomous Agents and Multiagent Systems (AAMAS) 2023 (acceptance rate: 237/1015=23.3%)
The preliminary version was presented at ICML 2022 workshop (AI4ABM)
[J1] Enabling Technologies for AI Empowered 6G Massive Radio Access Networks [pdf]
Md. Shahjalal, Woojun Kim, Waqas Khalid, Seokjae Moon, Murad Khan, ShuZhi Liu, Suhyeon Lim, Eunjin Kim, Deok-Won Yun, Joohyun Lee, Won-Cheol Lee , Seung-Hoon Hwang, Dongkyun Kim, Jang-Won Lee, Heejung Yu, Youngchul Sung, Yeong Min Jang
ICT Express, July 2022
[C3] MASER: Multi-Agent Reinforcement Learning with Subgoals Generated from Experience Replay Buffer [pdf]
Jeewon Jeon, Woojun Kim*(co-advised), Whiyoung Jung, Youngchul Sung*
International Conference on Machine Learning (ICML) 2022 (acceptance rate: 1235/5630=21.9%)
[C2] Communication in Multi-Agent Reinforcement Learning: Intention Sharing [pdf]
Woojun Kim, Jongeui Park, Youngchul Sung
International Conference on Learning Representation (ICLR), 2021 (acceptance rate: 860/2997=28.7%)
[C1] Message-dropout: An Efficient Training Method for Multi-Agent Deep Reinforcement Learning [pdf]
Woojun Kim, Myungsik Cho, Youngchul Sung
AAAI Conference on Artificial Intelligence (AAAI), 2019 (acceptance rate: 1150/7095=16.2%)
Workshop Papers
[W4] Decision ConvFormer: Local Filtering in MetaFormer is Sufficient for Decision Making
Jeonghye Kim, Suyoung Lee, Woojun Kim*(co-corresponding), Youngchul Sung*
Foundation Model for Decision Making (FMDM) Workshop at NeurIPS 2023
[W3] Off-Policy Multi-Agent Policy Optimization with Multi-Step Counterfactual Advantage Estimation
Seongmin Kim, Seungyul Han, Woojun Kim, Jeewon Jeon, Youngchul Sung
Adaptive and Learning Agents (ALA) Workshop at AAMAS 2023
[W2] An Adaptive Entropy-Regularization Framework for Multi-Agent Reinforcement Learning [pdf]
Woojun Kim, Youngchul Sung
Decision Awareness in Reinforcement Learning (DARL) Workshop at ICML 2022
[W1] A Variational Approach to Mutual Information-Based Coordination for Multi-Agent Reinforcement Learning
Woojun Kim, Whiyoung Jung, Myungsik Cho, Youngchul Sung
International AI for Agent-Based Modelling (AI4ABM) Workshop at ICML 2022
Preprint Papers
[P1] A Maximum Mutual Information Framework for Multi-Agent Reinforcement Learning [pdf]
Woojun Kim, Whiyoung Jung, Myungsik Cho, Youngchul Sung
Awards & Honors
Honorable Mention, 25th Samsung HumanTech Paper Award, Feb. 2019 ($2,000)
Championship, The 1st AI Soccer Worldcup with Team AFC_WISRL, Aug. 2018 (Team Leader) ($10,000)
Second Prize, The 1st AI Soccer Korean Cup with Team AFC_WISRL, Dec. 2017 (Team Leader) ($2,000)
Invited Talks
"Message-dropout: An Efficient Training Method for Multi-Agent Reinforcement Learning" at NAVER, Feb. 2019
"AI Worldcup and Multi-Agent Reinforcement Learning" at Electronics and Telecommunications Research Institute (ETRI), Nov. 2018