Types of Machine Learning

Machine learning is a simple study of teaching a computer program or algorithm which enables one to gradually improve upon a set task that is provided at a high level. On the side of research of things, machine learning can be seen through the theoretical lens and mathematical modelling of how this process works. More clearly, it is the study of building applications that displays this iterative improvement. There are multiple ways in which this idea can be framed, but mainly it has three major known categories: Reinforcement learning, Supervised learning as well as Unsupervised learning.

In a world soaked by artificial intelligence, machine learning, and over-zealous talk about both, it is fascinating to learn how to figure out and identify the multiple types of machine learning we may face. For an average computer operator, this can take the form of understanding the different machine learning and how they may reveal themselves in applications we operate.

Types of Machine Learning

For the professionals who make these applications, it is necessary to know about the types of machine learning so that for any designated task, you may face.

1. Supervised Learning

Supervised learning is the most famous pattern for machine learning. It is effortless to understand, and it is the easiest way to implement it. This method is quite similar to teaching a student with the use of flashcards.

Provided stats are the examples with labels; we can sustain a learning algorithm with these examples- label one by one, this allows the algorithm to judge the label for each example, and providing it with feedback to check whether it guessed the right answer or wrong. Over time, the algorithm will master to approximate the definite nature of the relationship between the examples and their labels. 

It is commonly defined as task-oriented because of this. It is immensely focused on a single task, giving more and more examples to the algorithm until it can perfectly perform that task. It is the type of learning that you will most likely face, as it is shown in many of the following applications: https://hackr.io/blog/real-world-machine-learning-applications

1.1. Advertisement Popularity

A selection of advertisements that are likely to perform is made by a supervised learning task. The various ads you see while surfing on the internet are placed there because a learning algorithm claimed that they were of reasonable popularity. Besides, its placement is connected on a definite site or with a definite query (if you find yourself using a search engine), it is mainly due to a learned algorithm claiming that the matching between the ad and placement will be useful.

1.2. Spam Classification

If you work on a modern email system, it increases the chances that you have encountered a spam filter. That spam filter is a type of supervised learning system. Fed email and labels (spam/not spam), these systems understand and learn how to prevent themselves from malicious emails so that the user is not disturbed by them. Many of these work in such a way that a user can give new labels to the system and help to learn user preference.

1.3. Face Recognition

Do you use Facebook? If yes then it is more likely possible that your face has been used in a supervised learning algorithm, which is taught to recognize your face. A supervised process is a system that takes photos, finds faces, and guesses who in the photo. It consists of multiple layers to it, which helps in finding faces and identifying them, but it is still supervised.

2. Unsupervised learning

It is opposite to supervised learning. No features are labeled; instead, the algorithm feeds on a lot of data and provides tools to understand the property of the given dataset. With the help of this, it can learn how to group, cluster, or organize the data in such a way that a human can contribute to it and make sense of the newly organized data.

The thing which makes unsupervised learning so fascinating is that an overwhelming majority of data in this world is unlabeled. Using intelligent algorithms that can carry our terabytes and terabytes of unlabeled data and make sense of it can eventually become a huge source of potential profit for many industries. It alone can increase productivity in many fields.

For example, imagine we have a huge database of every research paper ever published, and we had an unsupervised learning algorithm that is aware of how to group these in a way that you are every time aware of the current progression within a particular domain of research. Now, you take up a research project yourself, attaching your work into this network so that the algorithm can see it. As you write down your work and carry your notes, the algorithm suggests you about the related work, works you want to cite, and works, which might help you to push the domain of research. With the help of this tool, one can surely boost up their productivity.

 As the data and its properties are the main basis of unsupervised learning, we can conclude that unsupervised learning is data-driven. The outcomes we get from an unsupervised learning task are controlled by its data as well as the way it is formatted. Some of the areas you might see in unsupervised learning are:

2.1. Recommender systems

If you have ever used youtube or Netflix, you have most probably faced a video recommendation system. In the unsupervised domain, these systems are often used at times. We all are quite familiar with the composition of the videos. We also have information about the watch history of many operators. Looking into the account of other users who have seen similar videos like you and now are enjoying other videos that you have yet not seen, a recommender system can judge the connection in the data and prompt you with similar suggestions.

2.2. Buying Habits

There is a full possibility that your buying habits are detected in a database somewhere, and that data is being processed and sold actively at this time. Unsupervised learning algorithms use these buying habits to group customers into the same purchasing segments. It is helpful to the company's market to resemble recommender systems. For example, if you are recently browsing on the internet for online purchase of mobile phones than the very next time you log in, you will start getting more options and opportunities for mobile phones from different websites that offer competitive prices and homogenous product features.

2.3. Grouping User Logs

It undergoes minimum user-facing, but still, there is maximum relevance, unsupervised learning can be used to group user logs and issues.  The issues their customer faces can be identified by this as well as it can also be corrected by improving the product or designing an FAQ to rectify and correct the common issues. It is a method that is actively followed to fix a bug. If you have ever registered an issue with a product or registered a bug report, it's possible to feed as an unsupervised learning algorithm to bunch it with other same issues or bugs.

3. Reinforcement Learning

Reinforcement learning is impartially different when it is contrasted to supervised and unsupervised learning. As the relationship between the supervised and unsupervised is visible, on the other hand, the relationship to reinforcement learning is a little foggy. At some point in time, few people try to bind reinforcement learning closely with the two by defining it as a type of learning that is dependent on a time sequence of labels, but in my view, it makes things more confusing.

I consider reinforcement learning as learning from the mistakes. A lot of mistakes are made in the beginning when a reinforcement learning algorithm is placed in any environment. As long as we provide some kind of sign to the algorithm that identifies good behavior with a positive signal and bad behavior with a negative signal, we can reinforce our algorithm to increase good behaviors over bad ones.  Eventually, our learning algorithm learns to make a few mistakes than it used to before.

Reinforcement learning is very behavior-driven. Fields of neuroscience and psychology have got a great influence over it. If you are aware of Pavlov's dog, then you may know about the idea of reinforcing an agent.

Nonetheless, to properly understand reinforcement learning, let's go through an example. Let's study about teaching an agent to play the game called Mario.

For any problem faced during reinforcement learning, we need an agent and an environment as well as a proper way to combine the two with the help of a feedback loop. A set of actions are given, which are taken to the environment to combine or connect the agent to the environment. For connecting the agent and the environment, we have to continuously send two signals to the agent: a reward and an updated state.

Elaborating the early example in the game of Mario, the learning algorithm is the agent, and the game itself is our environment. The agent consists of a set of actions. These are referred to as our button states. Our updated state is each game frame as time passes, and there will be a change in our scores due to our reward signals. We will have set up of reinforcement learning scenario to play the game Mario but the only then when we connect all these components.

Example of reinforcement learning in the real world

1. Video games

Video games are one of the most common things to look at reinforcement learning is in mastering to play games. Google's reinforcement learning applications like AlphaZero and AlphaGo are the applications which mastered to play the game Go. The example we discussed before of the game, Mario is also a common example. I can imagine that this learning will soon be introduced and will be an interesting option for game devs to employ, but currently, there is no known production-grade game that has reinforcement learning.

2. Resource management

Reinforcement learning is great for navigating complex environments. Certain requirements can be handled with the help of resource management. Reinforcement learning was used to balance the need to satisfy the power requirements of Google's data centers. But it has to be done as efficiently as possible. How is the average person affected by this?  It has less impact on the environment we all share and provides us with cheaper data storage costs. 

Conclusion

The three various categories of machine learning that we have discussed, it is necessary to note that many of the lines between these types of learning are quite blur or vague. Nonetheless, there are a variety of tasks that can easily be phrased as one type of learning and then remold it into another paradigm.

For example:  taking the recommender system. Earlier it was discussed as an unsupervised learning task. It can also be remolded as a supervised task. Provided with a group of users watch histories, guess whether a certain film should be recommended or not recommended. In the end, all the learning remains learning. It simply depends on how we phrase the problem statement. Some of the problems are more easily phrased in a way or another.

Another fascinating idea is also highlighted in which we can mix these types of learning, designing the components one way or another, but combine it in a huge algorithm.

• Why is not supervised learning ability to recognize and label enemies provided to an agent playing game Mario?
• Why is the system that classifies sentences not given the ability to capitalize on a representation of the meaning of a sentence learned through an unsupervised process?
• Why is the reinforcement process used to group people in a social network into key segments and social groups?

Hence, in my view, it is very necessary that we all find out about a bit of machine learning, if we even do not construct a machine learning system ourselves. Machine learning is becoming increasingly more frequent in everything we use on our daily basis. The world is excessively changing. Finding out the elementary will indeed assist us in navigating this world easily. It will allow us to the world of better reasoning about the technology that we use. 

People are also reading:

• Machine Learning Certification
• Machine Learning Books
• Machine learning Interview Questions
• How to become a Machine Learning Engineer
• Types of Machine Learning
• Decision Tree in Machine Learning
• Machine Learning Algorithm
• Difference between Data Science vs Machine Learning
• Difference between Machine Learning and Deep Learning
• Difference between Artificial Intelligence vs Machine Learning