Exploration and exploitation
Exploration tries actions to learn about their outcomes; exploitation chooses actions that currently appear best.
Definition
Exploration tries actions to learn about their outcomes; exploitation chooses actions that currently appear best.
Intuition
A restaurant with one good review may look best only because you have not tried alternatives. Uncertainty matters.
Worked example
With epsilon = 0.1 and two actions, epsilon-greedy picks the greedy action with probability 0.9 + 0.1/2 = 0.95.
The math
In epsilon-greedy, take a uniformly random action with probability and an estimated best action otherwise.
In practice
Explore during training where mistakes are controlled. Evaluate separately with a fixed policy so exploration noise does not hide progress.
Sources
More in Reinforcement learning
Assembled from the ReLU.chat curated knowledge base. These explanations are concise on purpose; check the sources for anything important.