A Giant Step in Evolution of AI | Reinforcement Learning

Learning from action
or
Learning from action rather than learning from data

If you want to do Supervised Learning, you've to create a data-set to train. It's not always an easy task to create a huge data-set. So, here Reinforcement Learning comes into the picture, where an agent learns entirely by itself.


Lets take a small example of a robot that is learning by itself. Suppose the robot is assigned to walk continuously in a hall with dimension 100 m x 50 m:

Step-1: Robot walks straight for 100 meter, it hits the wall and it stops (Error!).

Step-2: Robot walks for 100 meters and then take left run (It took turn because now it knows that there is an obstruction after 100 meters). Robot continues walking but again after 50 meters it hits with another wall and stops (Error!).

Step-3: Robot walks for 100 meters. Take left turn. Walks again for 50 meters and then turn right (It turn right because it knows from its previous collision that after 50 meters there is an obstruction again).

Well, whatever I maintained above, is just a high level view of Reinforcement Learning.

The real implementation is way too complex! Oh, don't get discouraged! If you want to transform this world with AI, it's gonna be so interesting!


Have you ever implemented or thinking to implement Reinforcement Learning for any of your applications? Share your experiences in the comments below and please do share the post with your friends as well!