Machine Learning  Categories
There are several broad categories of machine learning, including the following:
As shown in the above diagram, machine learning evolved from left to right.

In the case of housing price prediction, researchers initially used Supervised Learning.

Following that came unsupervised learning, in which the machine learned on its own without any supervision.

Researchers discovered that rewarding the machine for doing the expected job may be a good idea, which led to Reinforcement Learning.

It is becoming increasingly obvious that the amount of data available today is so enormous that conventional techniques cannot be used to analyze the big data and make predictions based on it.

In our binary computers, Artificial Neural Networks (ANN) simulate the human brain, resulting in deep learning.

Using today's high computing power and large memory resources, the machine learns on its own.

Deep Learning has now been observed to be able to solve many previously unsolvable problems.

Now Deep Reinforcement Learning is being developed by giving incentives to Deep Learning networks.
Now let's examine each of these categories in more detail.
Supervised Learning
When you supervise a child's learning, you hold his hand, demonstrate how to take his foot forward, walk yourself as a demonstration, and so on, until the child learns to walk independently.
Regression
In supervised learning, you provide concrete examples to the computer. You say that for given feature value x1, the output is y1, for given feature value x2 it is y2, for given feature value x3 it is y3, etc. As a result of these data, you allow the computer to determine a relationship between x and y empirically.
Assuming that you know the real value of Y for a given X, you can deduce whether the machine's prediction is valid once it has been trained in this way with enough data points.
Once you are satisfied that the machine can perform the predictions with a desired level of accuracy (say 80 to 90%), you can stop further training the machine.
Now you can use the machine for predicting unknown data points, or ask the machine to predict Y for a given X for which you don't know the real value of Y. This is regression, which we discussed earlier.
Classification
You may also use machine learning techniques for classification problems. In classification problems, you classify objects of similar nature into a single group. In the case of 100 students, you may want to divide them into three groups based on their heights  short, medium, and long. By measuring the height of each student, you can place them in the right group.
As soon as a new student arrives, you will place him in the appropriate group by measuring his height. In regression training, you will train the machine to classify students based on their height. Once the machine understands how groups are formed, it will be able to classify any unknown new student correctly. Before putting the developed model into production, you would use the test data to verify that the machine has learned your classification technique.
This technique has been applied successfully in a number of cases. The handwriting recognition on your machine was done using this model. A number of algorithms have been developed for supervised learning. In the following chapters, you will learn about them.
Unsupervised Learning
In unsupervised learning, we do not specify a target variable to the machine, rather we ask the machine “What can you tell me about X?”. Specifically, we might ask questions like: “What are the five best groups we can form from X?” or “What features occur together most frequently in X?”. It is clear that the machine would need a large number of data points to arrive at the answers to such questions. With supervised learning, even a few thousand data points can be used to train the machine. In unsupervised learning, however, the number of data points accepted for learning starts in the hundreds of millions. These days, the data is generally abundantly available. The data should be curated ideally. However, because of the volume of data that is continuously flowing in a social area network, data curation is usually impossible.
This image shows the boundary between yellow and red dots as determined by unsupervised machine learning. You can see that the machine would be able to determine the class of each black dot fairly accurately.
Source:
https://chrisjmccormick.files.wordpress.com/2013/08/approx_decision_boun dary.png
Many modern AI applications, such as face detection, object detection, etc., have shown great success with unsupervised learning.
Reinforcement Learning
When we train a pet dog to bring a ball to us, we throw it a certain distance and ask him to fetch it back to us. We reward the dog every time he does this correctly. In the future, as the dog learns that doing a task correctly will earn him a reward, he will start doing the task correctly every time. As a matter of fact, this concept is utilized in “Reinforcement” learning techniques that were initially developed for playing games by machines. The machine is given an algorithm to analyze all possible moves at each stage of the game. It may select one of the moves at random. If the move is correct, the machine is rewarded, otherwise penalized. Over time, the machine will be able to differentiate right from wrong moves and eventually solve the game puzzle with a higher degree of accuracy. As the machine plays more games, its chances of winning will improve.
The following diagram illustrates the entire process
The key difference between this machine learning technique and supervised learning is that you do not need to provide labelled input/output pairs. The focus is on balancing exploration and exploitation of the solutions that have already been learned.
Deep Learning
Artificial Neural Networks (ANN), specifically Convolutional Neural Networks (CNN), are the basis of deep learning. As part of deep learning, several architectures can be used, including deep neural networks, deep belief networks, recurrent neural networks, and convolutional neural networks.
Computer vision, speech recognition, natural language processing, bioinformatics, drug design, medical image analysis, and games have all been successfully solved with these networks. Several other fields are actively using deep learning. Deep learning requires a huge amount of processing power and a large amount of data, both of which are generally readily available these days.
In the following chapters, we will discuss deep learning in more detail.
Deep Reinforcement Learning
In Deep Reinforcement Learning (DRL), deep learning algorithms like Qlearning are combined with deep learning algorithms in order to create a powerful reinforcement learning model. This technique has been extremely successful in the fields of robotics, video games, finance, and healthcare. Many previously unsolvable problems can now be solved with DRL. A lot of research is being conducted in this area, and the industry is very active in this area.
Let us take a closer look at some of the algorithms that are available under various machine learning models now that you have had a brief introduction to them.