Kyujin Choi

Korea Advanced Institute of Science and Technology

Papers

Total Citations

H-Index

About

Kyujin Choi is a researcher advancing the frontiers of reinforcement learning, with a focus on making AI agents more efficient and cooperative. His work addresses two critical challenges: improving sample efficiency in complex environments and enabling effective coordination among diverse agents. In his highly cited 2020 paper, "Batch Prioritization in Multigoal Reinforcement Learning" (14 citations), Choi introduced a novel method to prioritize training experiences, allowing goal-conditioned policies to generalize more effectively across multiple objectives. This work directly tackles the inefficiency of random sampling in replay buffers, a core bottleneck in multigoal RL. Building on this, his 2021 study, "Two-stage training algorithm for AI robot soccer" (7 citations), extends his expertise into multi-agent systems. Here, Choi proposed a centralized training framework that learns a joint-action set, enabling heterogeneous agents—robots with different roles—to develop sophisticated cooperative behaviors. This two-stage approach is particularly impactful for domains like robotics and autonomous systems, where agents must collaborate seamlessly. With a growing citation record, Choi’s contributions are shaping how we train AI to be both goal-directed and socially intelligent, marking him as a promising voice in modern reinforcement learning.