Batch Constrained Q-learning (BCQ)

Value Function Estimation

Process of estimating Q-values from offline data while accounting for potential bias due to lack of exploration. Modern methods use conservative underestimation techniques to avoid over-optimization.

← Tillbaka