Reinforcement Learning is a generalization of Machine Learning framework we seen so far which deals with a sequential decision-making process. ML problems like classification and regression lack of the sequentiality nature. In RL, each decision influences future decisions. The aim is to model process dynamics and choose from different actions in each situation. Two different problems:
- Prediction: given a specific behaviour (policy) in each situation, estimate the expected long-term reward starting from a specific state.
- Control: learn the optimal behaviour to follow in order to maximize the expected long-term reward provided by the underlying process.
RL is used in scenarios where the dynamics are unknown or it is difficult to directly model the problem. It is also used when the model is known but too complex to solve exactly or when there is a need to make sequential decisions.
Resources:
https://www.deepmind.com/learning-resources/introduction-to-reinforcement-learning-with-david-silver