What is entropy regularization in cost functions?

Entropy regularization encourages exploration by penalizing overly confident predictions.

Entropy regularization adds a term that increases uncertainty in model outputs, preventing premature convergence to deterministic policies. In reinforcement learning, it is critical for exploration. It modifies the cost landscape to favor distributions with higher entropy, improving robustness and preventing collapse into suboptimal policies.