New framework improves credit assignment in reinforcement learning

Researchers have introduced Counterfactual Shapley Credit Assignment (CSCA), a new framework for reinforcement learning agents. CSCA uses the Counterfactual Shapley Value ($\phi$-value) to better attribute credit and blame for outcomes, distinguishing between an agent's skill and environmental randomness. This approach enhances temporal credit assignment, particularly in scenarios with sparse causality, high stochasticity, and delayed rewards. An efficient estimator for $\phi$-values has been developed, leading to a new policy gradient method called $\phi$-PPO, which demonstrates superior sample efficiency in challenging environments. AI

IMPACT Enhances sample efficiency and explainability in reinforcement learning agents, potentially improving performance in complex environments.

RANK_REASON The cluster contains a research paper detailing a new framework and method for reinforcement learning. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Counterfactual Shapley Credit Assignment
• Counterfactual Shapley Value
• $\phi$-PPO
• $\phi$-values
• Prioritized Trajectory Replay
• reinforcement learning

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →