Types of ML
As we know Machine Learning is the application of Artificial Intelligence which provides the ability to computers to learn like human and progress from experience by feeding data and information into them without the human aid of getting programmed unequivocally. In machine learning, the machine don’t rely on human programmers for providing explicit instruction, in fact they use machine learning algorithms for training data sets of real-world to deduce models that are additionally accurate and refined. There are three different classes of ML and each of them has a main objective and that is to solve a particular problem. Let us see them:
Supervised Learning:
In Supervised learning, machine learns from the training data that is labeled (now a labeled data is the data having a label feature to be predicted). It is a machine learning model that is able to predict future outcomes by the help of training data from past. In simpler words, we need to supervise the machine while training it to work on its own. Suppose how tough it could be to handle e-mails if Gmail didn’t know what is spam and what isn’t. The supervised learning follows the question and answer pattern. The question and answer are given as input so that machine learns from it and when any new data comes across then the machine is able to make predictions.
For example, if we want to teach the computer to distinguish between different shapes. For this we will run the algorithm on a variety of images of shapes. In order to train the algorithm to learn the procedure of sorting images, we label the pictures as triangles, circles and squares. Once our algorithm learns how to classify images we
now use it on new data and predict labels on new and unseen images. Let us consider the image below for understanding this example.
We can clearly see we train the algorithm on the basis of past data. We feed various shapes along with their respective names. When the algorithm is trained then we pass new data. Now, we can see the machine is capable of determining that a triangle goes into the box of triangle and similarly, the circle and square goes into their respective boxes.
Supervised Learning is further classified into two types. They are:
Classification: In this, we deal with yes or no prediction. It enables the machine to be trained to classify observation into some class. Let us talk about the very significant use case of classification i.e, spam filtering. To a given system we provide spam and not spam mails. Now how is it determined whether a particular e-mail is spam or not. So, it is on the basis of how often
these mails are being received, how many of them have words such as lottery (usually associated with spam) are there and how many of them have been flagged by the users? These all information is given to the computer and it learns from it. Now when it has learned, whenever a new mail comes it is classified as spam or not spam. This sort of classification is done on the basis of Naïve Bayes algorithm.
Regression: In this, we try to predict value for an input on the basis of past data. Now there might be a question that classification and regression sound quite similar. Yes they are similar in terms of usage of past data but the difference between them is- in classification, we classify observations into cupid categories whereas in regression we predict the value of observations on the basis of past data. Now, let us take an example of weather reporting for understanding Regression. We have past data like temperature, speed of wind, humidity. Using these information, we can find out how much rain will be obtained. The system learns from it and when new data comes such as temperature, speed of wind and humidity, the system can tell how much rain we will get. But we need to remember one thing whatever output we get at the end of the regression problem is always quantitative in nature.
Unsupervised Learning:
Unsupervised learning builds model from data without predefined classes and finds hidden pattern in the data. In unsupervised learning we have no correct answers, we only have input data. Let us consider an example of Facebook. Suppose we uploaded a picture on Facebook. When the picture is uploaded then the Facebook is able to recognize who else are in the picture and recommends the person to be tagged in the picture. But how does it happen? It happens by taking our friend’s picture as datasets and encompassing it (picture) as the faces and attributes of the person present in our picture. And hence Facebook identifies who else is in our picture. The unsupervised learning deals with the data. It organizes the data and makes them more readable. It analyzes the data in order to find patterns. Hence, it finds structure and relationships among the input data.
In the above figure, we have provided some data as input. In this learning the system is able to identify, segregate and organize the data by discovering pattern in it. The similar thing occurs on Facebook and on e-commerce sites. It has clustering algorithm.
Clustering: It is a method of dividing sets into subsets known as clusters. The observations inside clusters are similar to each other and are based on some parameter or other. As a result, whole data is partitioned into clusters. Suppose we went to a party and we met new people. Now we cluster them on the basis of age, gender etc. Most of the e-commerce sites use clustering for recommending products to different group of buyers by identifying the hidden pattern in their preferences. It is also used in search engines.
Reinforcement Learning:
This is the type of learning in which the machine learns by seeing the results of the actions performed by it within a specific context to maximize its performance. The results are basically in the form of negative or positive reaction. Let us take an example: Suppose you place a bowl of hot milk. The cat will try to drink the milk without thinking whether it’s hot or not. This time the cat has learned that the milk is hot. Now when next time you place the same bowl in front of your cat, the cat will then drink the milk carefully. Because, cat has learned from its previous action. The same happens in reinforcement learning. In reinforcement the machine will perform some action and on the basis of its result it will determine whether the action was of worth or not. The next time the machine will choose the best potential action. This type of learning is basically iterative learning on the basis of past results.
So, these were the types of machine learning.