Reinforcement learning

RL in ML

Reinforcement Learning is a subfield of machine learning that teaches an agent how to choose an action from its action space, within a particular environment, in order to maximize rewards over time.


Reinforcement Learning has four essential elements:

Supervised, Unsupervised, and Reinforcement Learning: What are the Differences?

RL vs ML

Difference #1: Static Vs.Dynamic

The goal of supervised and unsupervised learning is to search for and learn about patterns in training data, which is quite static. RL, on the other hand, is about developing a policy that tells an agent which action to choose at each step — making it more dynamic.

Difference #2: No Explicit Right Answer

In supervised learning, the right answer is given by the training data. In Reinforcement Learning, the right answer is not explicitly given: instead, the agent needs to learn by trial and error. The only reference is the reward it gets after taking an action, which tells the agent when it is making progress or when it has failed.

Difference #3: RL Requires Exploration

A Reinforcement Learning agent needs to find the right balance between exploring the environment, looking for new ways to get rewards, and exploiting the reward sources it has already discovered. In contrast, supervised and unsupervised learning systems take the answer directly from training data without having to explore other answers.

Difference #4: RL is a Multiple-Decision Process

Reinforcement Learning is a multiple-decision process: it forms a decision-making chain through the time required to finish a specific job. Conversely, supervised learning is a single-decision process: one instance, one prediction.


There are currently two principal methods often used in RL: