What is Reinforcement learning?

Reinforcement learning is training of machine learning agent to make sequence of decisions. The machine learning agent learn to achieve its goal in uncertain and potentially complex environments. To make the machine do what the programmer wants, AI either get rewards or penalties based on the action it performs.

To understand reinforcement learning simply we can differentiate between preprogrammed agent and reinforcement learning agent.

So let's take the example of robot dog.

A robot dog which is preprogrammed already has the hard coded algorithm in it which includes various predefined methods like sit, stand, walk, etc.

So the robot can only perform the actions which are defined in it.

On the other hand if we take the example of robot dog which has reinforcement learning algorithm, that dog do not have any predefined methods which leads him to perform action.

The robot dog will learn  what the programmer wants based on getting rewards or penalties digitally.

The main goal of the machine is to maximize total number of records.

So the dog will perform the action like sit, stand, walk, etc to get rewards.

By avoiding the actions like falling down, etc which push towards penalties.

Currently reinforcement learning is very efficient in machines creativity.