In AI & ML
The difference between the reward an agent actually obtained and the reward it could have obtained if it had known the optimal policy from the start.
In Economics
Regret measures the expected loss of a classifier relative to the ideal decision rule under a given cost trade-off and threshold.
Know first
Named as a prerequisite by the lectures that define this. The number is how many say so. A grey one has no page yet.
Reinforcement learning3Action-value2Expected value2Optimal policy2Expected reward1Exploration algorithms1Multi-armed bandit1Optimal value1
Opens up
Concepts that name this one as a prerequisite.
In AI & ML
4Each in their own terms, from the easiest level.
Pieter Abbeel
David Silver
The opportunity loss for a single step, measured as the difference between the optimal value and the action-value of the chosen action.Sergey Levine
The difference between the expected reward obtained by an exploration algorithm and the optimal expected reward over time.Andrea Zanette
The sum of the suboptimality gaps of the policy played by the agent over episodes.
In Economics
1Lectures that assume it
1Connections
16The lectures that define it or assume it, and the concepts next to it.
Concepts8
- Reinforcement learning
- Action-value
- Expected value
- Optimal policy
- Decaying $\epsilon$-greedy
- Optimistic exploration