Monte Carlo sampling
Also: Monte Carlo, Monte Carlo estimation, Monte Carlo methods
Draw samples from a distribution and average a function of them; the estimate converges to the true expectation, with error shrinking like 1/√n. Used for RL returns, dropout uncertainty and intractable integrals.
E[f(X)] ≈ (1/n) Σᵢ f(xᵢ), xᵢ ~ p
Example
REINFORCE estimates the value of acting by playing complete episodes and averaging what they actually returned: episode returns of 10, 4 and 7 give an estimate of 7.