Learning Stateful Predictive Knowledge From Experience
Published in ICML 2026 AIWILD Workshop, 2026
Predictive knowledge is central to reinforcement learning agents that must reason about temporally extended experience rather than isolated observations. This work proposes a general framework for learning stateful predictive knowledge from experience, enabling agents to build and update internal predictive state representations that support downstream decision making in sequential environments.
Recommended citation: Song, Y., Zhang, G., Ma, W., Zhang, C., Liu, Y., Liu, Y., Kang, H., Guo, Z., Yang, S., Schaul, T., Wang, J., & Zhao, J. (2026). Learning Stateful Predictive Knowledge From Experience. ICML 2026 Workshop on Agents, Interaction, and Intelligence in the Wild (AIWILD).
Download Paper