Hands-On Artificial Intelligence for Beginners

Experience replay buffer

While we touched upon it briefly in Chapter 8, Reinforcement Learning, let's dive into the technical details of experience replay buffers. Experience replay is a biologically inspired tool that stores an agent's experience at each time step process. Specifically, it stores the [state, action, reward, next_state] pairs at each time step:

Instead of running Q-learning on state-action pairs as they occur, experience replay stores these pairs as they are discovered. Experience replay buffers help with two things: 

• Remember past experiences by storing and randomly sampling them 
• Reduce the chance that experiences will ...