Within machine learning, we have Supervised learning, Unsupervised learning and Reinforcement learning. We are here to learn more about reinforcement learning.
As an activity to learn, you decide to learn Swimming. So you look out for some guidance. Luckily you found an instructor who teaches swimming to beginners. What do you do next? You go to his class and learn how to swim. Some might have a little fear of water but that too is taken care of by the instructor. You learn as he teaches and he is able to teach because of his experiences. After a couple of weeks, you finish the course and voila!!! Now you know how to swim.
What is Reinforcement Learning?
Taking the analogy with a machine, in reinforcement learning, Machines are just exposed to their world. Machines have no instructor like it has in supervised learning. It learns from the actual experience. Machines are provided with an initial state as an input and their task is to reach the final point without any external help. So how will a machine learn if it has no instructor? The answer to this question is itself. The machine learns from itself.
When I said machines are exposed to their world, what I meant was our virtual world. This virtual world is designed by the programmers itself which is a replica of the real world.A simple task of taking a napkin from the table is an example here. This is played as a game with a machine. Just like you get points while collecting coins in Mario, the same is with a machine. It is provided with reward points if it took the correct path.
If it took the wrong path, some points are deducted. With this game of points, the machine decides what is wrong and right. In Reinforcement learning, the output of the previous stage is the input of the next stage. Hence they are dependent on their experience. It removes all the junk information and stores the vital information for future purpose.
Reinforcement Learning in daily Lives
When the machine finally takes the napkin, it is rewarded with points. This experience is now stored in a machine. It will be used when any similar situation occurs. Hence, the experience of the machine matters the most. Just like a baby learns to walk across a room. He will fall several times and rise up again.
The reward is counted as his mother standing across the hall and calling out his name. Same is with the machine. But the only difference is that it is its own teacher. Once the machine is ready to deal right in the virtual world, it is ready to test in the real world. For this, several algorithms are applied. These algorithms help machines to learn more and more accurately.
Another concept which should be spotlighted is Deep Reinforcement Learning or Deep RL. It is reinforcement learning but with the use of Neural networks. Neural network is a field of study which implements the algorithms on a machine to make it more like the human brain. This adds more to reinforcement learning to learn more accurately. Machines are termed as agents who work in an environment with their effectors attached.
Their actions on the environment are perceived via sensors such as cameras, GPS, etc. and the decision is taken on the basis of sensor data. Based on the sensor data, feature extraction is done which helps to identify whether the decision is correct or not.
conclusion
The useful knowledge obtained is used for reasoning. Reasoning defends the knowledge obtained. It could be done by image recognition or audio recognition or video recognition. Based on that, planning to take further action is done. Once it is realised that it is correct, action is taken by the effector.
Till now, we have seen robots being made, people drawing their attention towards AI and now we have children learning coding. Machines are made to help humans but the only promise we need is supremacy of human and humanity over machines. Kudos to the progress!!!
Written by: simran ahuja
reviewed by: shivani yadav
If you are Interested In Machine Learning You Can Check Machine Learning Internship Program
Also Check Other Technical And Non Technical Internship Programs
