CARE-VI: Conservative Adaptive Reliability Estimation for Value Improvement in Off-Policy Actor-Critic Learning
Conservative Adaptive Ranking and Screening, Selector-Evaluator Value Assessment, and Dynamic Adaptive Risk-aware Enhancement form CARE-VI, an evidence-regulated target construction framework that preserves the backbone interfaces for critic regression and actor updates.
Xian-Mao Zou, Sheng-Zhu Shi, Jun Gao et al.
· 0 citations