Papers
7
Total Citations
165
H-Index
6
About
Jinwoo Shin is a researcher specializing in deep reinforcement learning (RL), with a particular focus on generalization, model-based learning, and reward shaping. His most influential 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 addressing one of RL's most persistent challenges — the inability of trained agents to generalize to unseen environments when learning from high-dimensional image-based inputs. By introducing a straightforward yet effective randomization technique, Shin demonstrated that simple architectural interventions can yield significant performance gains. Beyond generalization from a model-free perspective, Shin extended this inquiry into model-based RL with his context-aware dynamics model, tackling the difficult problem of learning transferable environment models. His work on offline-to-online RL and preference-based learning with semi-supervised reward estimation further reflects a broad and systematic approach to making RL more practical and data-efficient. His more recent exploration of multi-view masked world models signals a growing interest in visually grounded robotic manipulation. Across these contributions, Shin has consistently pushed toward making reinforcement learning more robust, adaptable, and deployable in real-world settings.
Research Focus
Key Achievements
Top Papers
- 1
- 2
- 3
- 4
- 5
- 6Multi-View Masked World Models for Visual Robotic Manipulation6 citations · 2023
- 7