Exploration Strategies in RL

Methods for Exploration

Balance exploration/exploitation.

Epsilon-Greedy

Random action with epsilon.

UCB

Upper confidence bound. Optimism in face of uncertainty.

Boltzmann

Softmax over Q-values.

Key Takeaways

1. Epsilon-greedy baseline
2. UCB for balanced exploration
3. Boltzmann exploration