Beginner's note -Reinforcement Learning

     

     The potential benefits that Reinforcement Learning has brought up is enormous.This is something beyond learning the concepts of Machine learning and Deep Learning.We step up ourselves to a higher level,a bit.

All of it connects to the idea of making improvements in the performance of an automated system by making it learn by itself by exposing our agent to the Environment.And the result is evident as the RL technology was able to beat the best players in gaming.For eg, RL can be used to train the bot to play ping-pong games too.

     

    

     In this article we are entering sinto the world of reinforcement learning,an important part of machine learning and artificial intelligence where an agent learn how to behave in the environment based on the actions done . 

     The point to note is that RL focuses primarily on the decision making process and learning from it  when it encounters with an environment and also with other agents and receiving rewards for actions they take.That is it must learn the consequences by his own actions through trial and error methods.Rewards can be positive value or negative value.Ultimately our aim is to maximize our reward move to better condition.

      For example in our daily life if you on the switch bulb will glow , it will be a positive reward. Instead if you  put your hand in the plug socket you will definitely get electric shock, it is a negative reward 

The key Terms in reinforcement learning :-

    1. Agent -The agent is the action taker in the system.In the above example the agent is you who takes the actions

    2. Environment - Environment is the surrounding part of agent where he interacts

    3. State - State represents the current situation in which the agent finds itself

    4. Action - Action is the way which the agent interact and changes the environment

    5. Reward - Reward is the immediate feedback from the environment based on the agents action in a particular state

    6. Policy - Policy is the strategy that the agent takes to perform the action  in a particular in the environment

      Let us assume a bot that is in a specified location of an office,suppose it wishes to assemble some components to build some kind of gadget by itself and that the components are at different spaces of the office,then we can say that on reaching the exact place where the components kept you get a positive reward.Then the bot initially moves and tries to find out which path is most suitable so as to reach the destination.Assuming each step is an action,you get the maximum rewards by taking certain steps and finally the one that gives you the maximum reward becomes its destination.The kind of place or location where the bot reaches by performing an action that is the state,in this case let us assume this as may be the next room.

   As we discussed earlier , the agent in an environment will take the decisions and perform the action by observing the environmental state and achieve the reward which results in the formation of new state . The goal of the agent is to select the best action and hence to achieve the maximum reward.

   Here we have discussed the key terms that one must primarily get familiar with.As the topic covers a large area of the Technology ,to learn and understand more we recommend you take courses available these days. Online teaching platform like Coursera have been providing students with free access to courses.You can also search out courses on edX and also for free on Youtube.

Get a good read everyday and learn more.

Stay tuned till our next post!!!