Dieqiao Feng
Papers
1
Total Citations
6
H-Index
1
About
Dieqiao Feng’s research lies at the intersection of deep reinforcement learning (RL) and combinatorial AI planning, with a focus on developing methods that enable agents to solve complex, long-horizon problems. Their most notable contribution is a novel automated curriculum strategy that successfully trains RL agents to tackle hard instances of Sokoban—a notoriously challenging planning domain where traditional RL often fails. This work, published in 2021 and garnering 6 citations, demonstrates how carefully structured training sequences can bridge the gap between RL’s success in domains like Go and video games and its struggles in structured planning tasks. Feng’s approach not only advances the theoretical understanding of curriculum learning but also provides a practical framework for applying RL to other combinatorial optimization problems. By systematically increasing problem difficulty during training, their method enables agents to discover solutions that were previously out of reach. This research is particularly valuable for students and researchers exploring how RL can be extended beyond reactive control to deliberate, multi-step reasoning—a critical step toward more general artificial intelligence.
Research Focus
Key Achievements
Top Papers
- 1