When it comes to Machine Learning, there are various types of learning, and we can classify it generally into three categories, viz., Supervised Learning, Unsupervised Learning, and Reinforcement Learning. Essentially, machine learning is an implementation of artificial intelligence that allows systems to learn and progress based on their experiences rather than being clearly programmed.
In several situations, it is seen people tend to get confused about the various types of machine learning. Thus, here I will be explaining and discussing the types of machine learning while taking some interesting real-life examples.
Firstly, when we talk about machine learning, our main aim is “How can we make our machines intelligent”, isn’t it? And the most important factor concerned with making machines intelligent is that they should have the ability to learn and implement the actual concept (logic) by themselves!
So, even though we have various types of learning, we will be confined to the three profound ones. Without any further ado, let’s dive right in.
Supervised Learning
Let’s start with Supervised Learning. As the name suggests, we have a supervisor who is providing the instructions, and it is nothing but the training data itself. In other words, we have input data used for training and testing, as well as the output data. In supervised learning, the input data used for training have labels that will help the machine to understand the data better.
A model is developed using that labeled data, and a test input is given to check whether the output obtained is correct or not. Let’s see some simple examples to understand the concepts clearly.
An Example Of Supervised Learning
For instance, consider the case of a kid from a lower primary class and his teacher wants to teach him the fundamentals of math like addition and subtraction. Subsequently, what the teacher will do is acquaint him with numbers and teach him how to add with small numbers like 2+2 = 4(training).
She will take unit tests (try on test cases) with entirely new questions like 8+5 = ? to check on the knowledge of kids, for the logic of addition. If the kid passes the tests, well then it means he is trained and ready for every new problem on addition. Otherwise, if the kid fails teacher is again assigned the duty of teaching (training) while adopting new methods.
This is how supervised learning works. In machine learning, labels associated with the training data are the “teachers” and they will help the machine to learn effectively.
Applications Of Supervised Learning
Supervised learning is mainly used to solve two types of problems:
1) Classification: A classification problem is when the output variable is a category, such as “blue” or “red” or “disease” or “no disease”.
To solve the classification problems, we use algorithms like Logistic Regression, K-Nearest Neighbor (KNN), Support Vector Machine (SVM), Naïve Bayes, Decision Tree Classification, Random Forest Classification, etc.
2) Regression: When the expected output is a real value, such as “temperature” or “price”, you have a regression problem. And, to solve the regression problems, developers usually use algorithms like Linear Regression, Multiple Linear Regression, Polynomial Regression, Support Vector Regression, Decision Tree Regression, Random Forest Regression, etc.
Check out this article to learn more about the applications of supervised learning. There are several advantages and disadvantages when we apply supervised machine learning. You can read this article to learn more about its pros and cons.
Unsupervised Learning
Now, let’s see what unsupervised learning is.
In unsupervised learning, the training data provided to the machines are unlabeled, which means we only have the input data, and we don’t have the corresponding output variables.
Let’s look at an example to understand the concept of unsupervised learning.
An Example Of Unsupervised Learning
A simple example to understand would be by taking an example from the movie called “PK”. A humanoid alien who lands upon the Earth from another planet on a research mission loses the remote control of his spaceship, and he doesn’t know anything about the earth.
Even though he has no prior knowledge, he judges what he is seeing, and based on the physical features of humans, he seems to cluster or make groups of people like males, females, kids, etc.
It is not true that all categories he created would be perfect but on the first level, he made those distinctions or clusters, which simply serve as the input, where he doesn’t know further about the outputs. And what happens is, he slowly understands things and more importantly, develops learning and reaches a stage of correctness in outputs.
Applications Of Unsupervised Learning
Unsupervised learning is used to solve many types of problems. Here are some of its main applications.
1) Clustering: When it comes to unsupervised learning, clustering is a crucial term. It is primarily concerned with identifying a pattern in a set of uncategorized data. If natural clusters (groups) exist in the data, clustering algorithms can process it and find them.
Customer segmentation is a widely known example of clustering, where each customer can be categorized into different categories like Price sensitive customers, Enquiry customers, Lazy customers, etc. For this, a dataset of customers is provided to the model to train, identify, and learn features of customers to cluster them.
This is mainly utilized in retail industries when the interests of customers are identified and products can be recommended when they come. You can see that this kind of recommendation system is used by online shopping companies as well.
K-Means Clustering and Hierarchical Clustering are some algorithms used for clustering.
2) Dimensionality Reduction: Techniques for reducing the number of input variables in training data are referred to as dimensionality reduction. The term “high-dimensionality” can refer to a large number of input variables.
Fewer input dimensions usually imply fewer parameters in the machine learning model, also known as degrees of freedom. A model with too many degrees of freedom is likely to overfit the training dataset, resulting in poor performance on the test dataset.
Check out this article to learn more about the real-life applications of unsupervised learning. This type of machine learning has several advantages as well as disadvantages. Check out this article to learn more about its pros and cons.
Reinforcement Learning
Now let’s learn about reinforcement learning.
Reinforcement learning is the process in which the machine makes mistakes and learns from them. If the machine makes wrong decisions, it will get penalties, and if it makes the right decisions, it will get rewards. The main objective of reinforcement learning is to maximize long-term rewards and minimize penalties.
Reinforcement learning is actually based on reward and policy. Let’s consider the humanoid alien from the previous example. Now, let’s say he is performing some action on the environment, and based on that, he either receives a penalty or a reward and thus the state changes. What happens is, considering these state changes, he is making a policy.
In reinforcement learning, a software agent makes observations and performs actions within an environment, and in return, it gets rewards.
An Example Of Reinforcement Learning
The alien gives a currency note which has a photo of Mahatma Gandhi to a local shopkeeper and he gets dozens of carrots in return. He assumes that he got the carrots because of the image of Mahatma Gandhi. Therefore, he goes on to collect as many images as he can and returns to the shopkeeper to get carrots.
However, this time, the shopkeeper refuses to accept those plain pictures, thereby also imparting a valuable message that the importance of a picture of Gandhi is only on the currency note, whose value could not be compared to that of any other printed paper.
This is a well-suited example of reinforcement learning where he is making policies based on rewards and penalties while learning simultaneously through making adequate changes in every circumstance.
Applications Of Reinforcement Learning
Reinforcement learning has a lot of applications in the real world. Here are some of them:
1) Game Playing: Reinforcement learning can be very useful in the best moves in games considering various factors. A machine can even beat world champions in games, and we have seen that many times with games like Chess and Go.
In games, when the machine takes a particular step, it either gets a positive or negative point. Based on that, the machine decides whether to take that step in future games or not. After several cycles of trial and error, the machine can become an expert in that particular game.
2) Robotics:
Reinforcement learning is widely used in robotics, where the robot is trained to do certain tasks by itself with human supervision. Based on the rewards and penalties it gets, it improves the policies and becomes good at doing tasks even without the help of humans.
This type of machine learning has several advantages as well as disadvantages. Check out this article to learn more about the pros and cons of reinforcement learning.
Final Thoughts
We have seen the different types of machine learning, some real-world examples of each, and their use-cases and applications. Supervised learning, unsupervised learning, and reinforcement learning are the three major types of machine learning we have. There are some combinations of these three types as well(for example, semi-supervised learning, which is a combination of supervised and unsupervised learning).
I hope that you’ll find this article helpful. Do write a comment if you have something more to add. Thanks for reading!
