Home » Machine Learning/Artificial Intelligence

Main Points of Reinforcement Learning in Artificial Intelligence

This article is in continuation of the previous article on Reinforcement Learning. In this article, we will further study the components of the Reinforcement Learning agent and will also discuss some main points related to it.
Submitted by Monika Sharma, on June 18, 2019

As discussed earlier, in Reinforcement Learning, the agent takes decisions in order to attain maximum rewards. These rewards are the reinforcements through which the agent learns in this type of agent.

The reinforcements are of two types:

1. Positive Reinforcement:
   When the agent completes any task, if the feedback or the points for the task are in a positive response, then it is termed as the positive reinforcement. This type of reinforcement increases the performance of the agent as the agent now gets a hint that it has to make decisions and perform tasks in this particular manner to earn more rewards in the future also.
2. Negative Reinforcement:
   Whenever the agent fails to perform any task as required, in that case, the agent is provided with negative reinforcement. This can be thought as of giving punishment to a child for doing mischiefs. The negative reinforcements tell the agent that such type of performance or such type of decisions must be avoided in the future while solving similar types of problems.

Factors on which the performance of the agent which learns through Reinforcements depend:

1. Input:
   The Agent seeks the initial stage as the input from which it has to start. This is an important phase because all the observations and inferences will be drawn starting from this state, and the past state of the agent will not be considered.
2. Output:
   The output state that the system will reach after solving a certain problem is not fixed as there are multiple ways of solving a problem and the agent can choose different solution whenever it tries to solve the same type of problem.
3. Training/Learning:
   The training phase or the Learning Phase is when the agent builds its Knowledge Base from the reward or punishment that it gets based on the output it produces. It is a very important phase in Reinforcement Learning because it helps the agent to understand and learn in the same way as humans. This implements the human behavior in agents which is the main target in Artificial Intelligence.