Introduction to machine learning

In iOS applications, we use CoreML to incorporate machine learning in iOS applications. It is Apple's framework to use pre-trained models in iOS applications. In this tutorial, we will discuss what machine learning is, different types of it, including some real-life examples of machine learning.

Machine learning is the field of study that allows computers to learn without being explicitly programmed. Using machine learning, we don't need to provide explicit instructions to Computers for reacting to some special situations. We need to provide training to the computers to find real-time solutions for the specific problems. The chess game is a famous example where machine learning is being used to play chess. The code lets the machine learn and optimizes itself over repeated games.

Machine Learning is broadly classified into two main categories.

  1. Supervised Machine Learning
  2. Unsupervised Machine Learning

Supervised Learning

Supervised learning is similar to having a trainer or teacher who supervises all the machines' reactions and tells step-by-step solutions to specific problems. It's like a hand-holding way of teaching the computers what to do. One real-time example of supervised learning is recognizing different types of images using computers. Being humans, we also learn by this model as we are taught to recognize different objects like a car by repeat exposures. In the same way, machines are taught.

We feed a different set of some specific images into a machine where each image has a specific identifier to identify the type of the image. However, computers are taught so that every time the particular blend of pixels comes in front of the computer; it can recognize the type of image loaded into the model dataset.

Supervised learning works in a way that the computer can learn through the previous exposures; for example, if a computer sees a car object and recognizes it like a car, then next time, it should be able to identify any different image of car object by identifying a lot of features that are similar to previously identified images of Car.

When we train a machine learning model for image recognition, we present many images where every single image is attached to a label so that the data can be clearly labeled and gets stored in a machine learning model. Once we complete the training, we should present an object's image that is not part of the training data, and the machine should be able to identify it by classifying all its previous learnings.

This is the most fundamental type of Supervised learning, which is called Classification. Out machine learning model must be able to classify the different bunch of images. It must be programmed so that the different objects can be recognized according to their unique characteristics. However, we can create a generic classifier so that it is not dependent on learning data. We don't need to recode the entire model on changing the training data.

Unsupervised Learning

In Supervised learning, the specific kind of dataset is loaded into computers to learn through the repeat exposures to the dataset. In this section of the article, we will discuss Unsupervised learning. It is one of the other major types of machine learning. Instead of providing training data where every piece of data is clearly labeled, we provide the unstructured training data in unsupervised learning. We want the model to sense the dataset so that it learns to find the structure in unstructured data.

In other words, we can say, in unsupervised learning, we don't tell computers the kind of data. Instead, we want the computers to see the structure in the data by observing which way the data is being organized.

One type of Unsupervised learning is called clustering, in which the computer looks at the dataset and its features and can figure out the separate clusters in which the data is maintained.

Reinforcement Learning

We have covered supervised learning in which we have loaded the labeled training data in the machine learning models so that the computers can classify the data and performs regressions to identify the dataset. We have also covered Unsupervised learning. We have loaded the unstructured unlabeled dataset grouped in separate clusters, and we want computers to be smart enough to identify the separate clusters.

In this section of the article, we will discuss Reinforcement Learning. As humans, we are much experienced in reinforcement learning. We tend to learn through reinforcement. For example, if driving through a route that is full of traffic jam, we'll ignore going through the same route on other days. There are two kinds of reinforcements we generally come through, 1. Positive, 2. Negative Reinforcement.

The same way machine works in the case of Reinforcement Algorithms. One of the real-time examples of reinforcement learning is a Chess game where the Computer with the Reinforcement learning algorithm calculates the winning probability with every move. The computer might come through positive as well negative reinforcement with every single move. However, through many cycles of training and by practicing more and more games, the computers will learn about which moves in which situations will lead to an increase in its winning percentage.


Next TopicWhat is CoreML




Latest Courses