Papers
18
Total Citations
272
H-Index
9
About
Kimin Lee is a prominent researcher specializing in deep reinforcement learning (RL), with a particular focus on generalization, model-based RL, and human-in-the-loop learning systems. His work addresses some of the most pressing challenges in deploying RL agents in real-world settings, where environments are dynamic, data is scarce, and reward engineering is costly. Lee's most recognized contribution, "Network Randomization: A Simple Technique for Generalization in Deep Reinforcement Learning," has accumulated nearly 100 citations across its versions, establishing him as a leading voice in improving RL robustness to unseen environments. He further extended this theme through information bottleneck approaches and context-aware dynamics models, tackling the critical problem of dynamics generalization across varying conditions. His research portfolio also spans model-based RL, one-shot visual imitation learning, and preference-based RL — where his PEBBLE and SURF frameworks demonstrate innovative strategies for reducing human feedback requirements while maintaining training efficiency. His offline-to-online RL work bridges the gap between static datasets and adaptive real-world deployment. With over 230 cumulative citations and contributions spanning robotics, sample efficiency, and human-AI interaction, Lee's research offers foundational tools for building more adaptable and practically deployable intelligent agents.
Research Focus
Key Achievements
Top Papers
- 1
- 2
- 3
- 4Towards More Generalizable One-shot Visual Imitation Learning27 citations · 2022
- 5
- 6
- 7
- 8
- 9Masked World Models for Visual Control10 citations · 2022
- 10