What is Experience replay?

Prerequisite knowledge
　・Deep Q-Network(DQN)

■What is Experience Replay?

Experience Replay is a method to improve learning accuracy in DQN. It memorizes past actions and periodically selects multiple memorized actions at random(batch size) for learning. By performing random learning on a regular basis in this way, you can prevent erroneously learning that there is a high correlation with time.

In addition, rather than randomly learning previous actions, a method of selectively learning actions that are considered to be effective for learning or actions that occur infrequently is called Prioritized Experience Replay.

The method of prioritizing uses the absolute value of the difference between the teaching data and the learning data, and if the absolute value is large, it means that the learning effect is high or the number of learning times is small, so it is selected preferentially. If the absolute value is small, it means that the learning effect is low or that the learning has been repeated enough times, so it is not selected often.

List of related articles TOP MENU Machine learning ・ε-greedy method ・Reinforcement learning ・Temporal Difference learning ・Regression vs Classification vs Clustering ・Parametric vs Nonparametric model ・LSTM ・Error function ・Experience Replay