Implicit Q-Learning (IQL)

Batch-Constrained Optimization

Strategy in IQL that constrains learned actions to remain close to those observed in the dataset to avoid unreliable extrapolations.

← Tillbaka