Yuanyi Zhong
Papers
3
Total Citations
22
H-Index
2
About
Yuanyi Zhong is a researcher specializing in deep reinforcement learning, with a particular focus on credit assignment, visual learning, and video prediction. His work addresses some of the most fundamental challenges in training intelligent agents, including how to effectively attribute rewards across complex decision sequences and how to leverage visual information to improve agent performance in real-world environments. Zhong's most notable contribution, "Sequence Modeling of Temporal Credit Assignment for Episodic Reinforcement Learning" (2019), tackles the long-standing temporal credit assignment problem — determining which actions in a sequence are responsible for eventual outcomes. By framing this as a sequence modeling challenge, his approach reduces dependence on dense, hand-crafted reward functions, a significant practical advancement that has garnered 17 citations. His subsequent work on disentangling controllable objects through action-conditioned video prediction demonstrates creative thinking at the intersection of computer vision and reinforcement learning, enabling agents to better understand and exploit the consequences of their own actions in visually rich environments such as video games and robotic manipulation tasks. Zhong's research reflects a consistent drive to make reinforcement learning more sample-efficient, generalizable, and applicable to complex, vision-based real-world problems — making his work valuable reading for students and practitioners in AI and robotics.
Research Focus
Key Achievements
Top Papers
- 1
- 2
- 3