• Category
  • >Machine Learning

Types of Machine Learning

  • Ritesh Pathak
  • Jan 05, 2021
Types of Machine Learning title banner

We have been discussing the latest technologies in our blogs. Artificial Intelligence, Machine Learning, Deep Learning, Natural Language Processing, the Internet of Things, and many other technologies are being leveraged by businesses. All these technologies work together to make our lives easier. Isn’t it fascinating when we can switch off our lights without needing to step out of the bed? This is just one of the many conveniences which have been made possible with the help of technology. Machines are on their way to become more intelligent and efficient. 

 

Also Read: Information Technology, its functions and why is it important?

 

Machine learning is the technology that is concerned with teaching computers different algorithms to perform different tasks, and making machines capable of taking care of themselves. Different ideas are framed and fed to machines. There are mainly three recognized categories of framing ideas, which we reckon as the three types of machine learning. 

 

 

 

Machine Learning

 

Machine Learning (ML), is simply the field of study that deals with teaching computer programs and algorithms to keep improving on a particular task. Machines make use of insights extracted from data. In a world where machines complete most of the tasks, they need to learn how things are done and also anticipate. This is where machine learning steps in. It teaches machines to learn on their own and make predictions based on previous insights. 

 

“Machine Learning is the science of getting computers to learn and act like humans do, and improve their learning over time in an autonomous fashion, by feeding them data and information in the form of observations and real-world interactions.”

- Emerj, The AI Research and Advisory company


 

Machine Learning is a part of artificial intelligence that aims at feeding computers or machine learning systems knowledge through data, observations, and interactions with the surroundings. There are different ways of doing it that we will explore in this blog.

 

 

3 types of Machine Learning

 

In times of excessive use of artificial intelligence and machine learning, it becomes necessary to differentiate the types of machine learning. As everyone perceives everything differently, for an average computer user, this can be about the exhibition of these different types of ML in several applications. 

 

While for a programmer, who is creating such applications, it is essential to know about the different types of ML so that they can create a proper learning environment, and also understand the purpose of creating such applications. 

 

The three major recognized categories of machine learning are: supervised learning, unsupervised learning, and reinforcement learning. 


The image shows the 3 types of machine learning that are supervised learning, unsupervised learning, and reinforcement learning.

3 types of machine learning


Recommended Blog: What is Confusion Matrix

 

 

1. Supervised Learning

 

Supervised learning is reckoned as the most popular and typical example of machine learning. It is the easiest and simplest form of machine learning that is easy to understand. It is like teaching a child to recognize things with the help of flashcards.

 

Algorithms are taught like a child to identify the given data. For an instance, let us consider the given data as examples with labels. So, what is done in supervised learning is that the algorithms are presented with example-label pairs one by one, allowing the algorithm to predict the label for each example. 

 

The person feeding these example-labels to the algorithms gives feedback on every prediction, whether it was correct or not. This practice is repeated over time until the algorithms start predicting the exact nature of the relationship between the examples and their respective labels. When fully-trained, the supervised learning algorithm will be able to observe a new, never-before-seen example and predict a good label for it.

 

Supervised learning is often described as task-oriented as it needs to perform a task several times before it is accurate. This is the learning we are likely to encounter most of the time.

 

Some common applications of this are: 

 

  1. Advertisement Popularity: Selecting an advertisement that will gain more popularity is often decided by supervised learning. The ads we see while surfing the internet are there because some learning algorithm predicted that they could gain reasonable popularity and clickability. We often encounter some ads on a specific platform or a certain site and when we search for a query. This is just because a learned algorithm suggested that the matching between that ad and placement will be effective. 

 

  1. Spam Classification: Spam mails were a big stress for users. However, they no longer bother us. Modern email systems like Gmail have a spam filter. This spam filter is nothing else but a supervised learning system. These systems are fed email examples and labels (spam/not spam) and are taught to differentiate them. These supervised learning systems learn to pre-emptively filter out spam and malicious emails. Many of these systems also allow its user to provide new labels so that it can learn user preference. 

 

  1. Face Recognition: Facebook is able to recognize faces and suggest us to tag them. How is it made possible? Most likely our faces are used in a supervised learning algorithm that is trained to recognize our faces. A supervised system is able to find faces in a photo, recognize them and suggest us to tag them. Google Photos also uses this supervised system. If you have used it you can remember the application suggesting to share the picture with the people in it. 


 

Recommended Read: What is an Algorithm? Types, Applications, and Characteristics


 

2. Unsupervised Learning

 

Unsupervised learning is totally opposite to supervised learning. There are no labels used in unsupervised learning. 

 

In unsupervised learning, the algorithm is given a lot of unorganized data and the tools to identify the properties of the data. The algorithm then leverages these tools to group, cluster, and organize the given data in a way that any intelligent algorithm or a human can make sense of the output i.e. the newly organized data. 

 

The ability to organize massive amounts of unorganized and unlabeled data makes unsupervised learning a demanding and interesting area. This is so because there is an overwhelming majority of unlabeled data present around us. If we can make anything sensible out of this data, it can prove highly beneficial. Unsupervised learning algorithms make it possible and bring huge profits. 


The image shows the working of unsupervised learning.

Working of unsupervised learning


Since unsupervised learning makes use of data and its properties, we can call it data-driven. The outcomes of unsupervised learning tasks depend on data and its formatting. Some applications of unsupervised learning are: 

 

  1. Recommender Systems: In the times of binging shows on Netflix and other such OTT platforms, unsupervised learning is being utilized subtly. When we watch and wishlist our favorite shows on these platforms, we provide data to the learning system. These platforms have a video recommendation system. 

 

The unsupervised learning system takes into account the uncategorized data in the form of our watch history, genres of shows, their length, and organizes this data. It then matches with other shows available and prepares a list of such shows that a user can be interested in. YouTube also uses this kind of unsupervised learning system. 

 

Also Read: Review-based Recommendations System


 

  1. Buying Habits: We are now almost used to shop online and we all have our shopping preferences. Some people prefer shopping on a specific platform. We wishlist items and also have purchasing history. This is all data. It is possible that all our buying habits are contained in a database and it is being actively traded at the moment you read. These buying habits are used in unsupervised algorithms to group customers into similar purchasing segments. This is used by companies to market specific products among suitable segments. 


 

  1. Grouping User Logs: Unsupervised learning can also be utilized to group users’ logs and issues. Unsupervised learning is considered as less user facing but it is still relevant enough to be utilized. Companies use this as a tool to understand the central theme of issues faced by their users and then work on it to rectify such issues. It can also be used in the designing of a product and preparing FAQs. When you report an issue or a bug, you may have possibly fed the data to an unsupervised learning algorithm which then clusters it with other similar issues. 

 

 

3. Reinforcement Learning

 

Reinforcement learning is distinct in many ways when compared to supervised and unsupervised learning. We can differentiate supervised and unsupervised learning on the basis of labeled and unlabeled examples. However, reinforcement learning uses no such labels. The relationship to reinforcement learning is a bit murkier. Some people try to make unnecessary ties by calling it a type of learning that relies on a time-dependent sequence of labels.

 

Reinforcement learning is very much behavior-driven. It has some impact from the fields of neuroscience and psychology on it. In psychology, we are taught about Pavlov’s dog. It gives us the idea about reinforcing an agent. Therefore, we can also look at reinforcement learning as the one that learns from it mistakes. When a reinforcement algorithm is placed in any environment, it makes a lot of mistakes in the beginning. It starts improving the moment some sort of signal to the algorithm, that associates good behaviors with a positive signal and bad behaviors with a negative one, is provided. Over time, it learns to make less mistakes.

 

Some of the applications of reinforcement learning are: 

 

  1. Video Games: One of the most common places where reinforcement learning is video games. Examples include Google's reinforcement learning application, AlphaZero and AlphaGo which learned to play the game Go. The game of Mario is a prime example of reinforcement learning application. 

 

In the game, the agent is learning algorithms and the game is the environment. The agent has some set of actions. There will be button states and every new game frame behaves as the updated status. The change in the score is our reward signal. So as long as we keep connecting all these components together, we will keep forming a reinforcement learning scenario.  

 

 

  1. Industrial Simulation: In the industries where robots are being utilized to perform different tasks, it becomes vital to make them capable of completing their tasks without having to monitor. It is a cheaper and efficient option and more than that it reduces the chances of failure. The machines can be programmed to consume less electricity and hence reduce costs. 


 

  1. Resource Management:  Google’s data centers use reinforcement learning to balance the need to satisfy our power requirements, but do it as efficiently as possible, cutting major costs. How does this affect us and the average person? Cheaper data storage costs for us as well and less of an impact on the environment we all share.

 

 

 

Conclusion

 

Now, since we have discussed the three different types of machine learning, it is important to note that sometimes the difference between them may not be clear or sometimes even they may seem the same. For instance, take a recommender system. We know it as an unsupervised learning task. It can also easily be rephrased as a supervised task. We would just need to label the data. 

 

The all three types of machine learning aim to teach computers algorithms that make them capable of performing tasks with more efficiency. 

Latest Comments