We aim to design reinforcement learning algorithms that balance theoretical soundness, safety, and adaptability, enabling trustworthy deployment across domains ranging from finance to home robotics. In this talk, we will introduce Offline RL and Cognitive RL methods that advance this goal.

Yun-Hsuan Lien is an Assistant Research Fellow at the Citi, Academia Sinica, specializing in reinforcement learning. Her work focuses on developing theoretical RL algorithms and applying them to dynamic, real-world environments. She is committed to reducing the “reality gap” by improving the robustness of RL models when leveraging simulators or pre-collected data, thereby overcoming the limitations of direct environmental interaction.