To always use the next reward as the estimate for future returns

To interpolate between using a single-step TD update and the full Monte Carlo return

To update the value function using the entire return of an episode

To update the policy after every action