credit assignment
The problem of determining which actions or decisions in a sequence were responsible for the outcomes, particularly in reinforcement learning. Effectively solving credit assignment helps improve learning efficiency by linking rewards back to prior actions.
- DISCOVER: Automated Curricula for Sparse-Reward Reinforcement Learning
- Error Broadcast and Decorrelation as a Potential Artificial and Natural Learning Mechanism
- Group-in-Group Policy Optimization for LLM Agent Training
- Meta-learning how to Share Credit among Macro-Actions
- NAVIX: Scaling MiniGrid Environments with JAX
- Optimizing Anytime Reasoning via Budget Relative Policy Optimization
- PRIMT: Preference-based Reinforcement Learning with Multimodal Feedback and Trajectory Synthesis from Foundation Models
- PRIMT: Preference-based Reinforcement Learning with Multimodal Feedback and Trajectory Synthesis from Foundation Models
- Reward-Aware Proto-Representations in Reinforcement Learning
- Segment Policy Optimization: Effective Segment-Level Credit Assignment in RL for Large Language Models
- Stop Summation: Min-Form Credit Assignment Is All Process Reward Model Needs for Reasoning