Reason-guided human reward modeling for preference-based reinforcement learning
In brief
- New preference-based RL framework integrating textual justifications to guide human alignment.
- Model combines preferences and reasons to enhance interpretability of optimized policies.
- Human reward is merged with environment reward to balance performance and alignment.
This study introduces a preference-based reinforcement learning (RL) framework that incorporates human reasoning into reward modeling. The model learns a reward function from trajectory feedback, combining preference labels with textual justifications to align behaviors with human expectations. Results demonstrate improved interpretability and alignment of optimized policies.
This summary gives you the essentials. The full article, its figures and quotes are on cs.AI updates on arXiv.org.
Read the original articleAI-generated summary of the source, checked by rules. Report an error