So the first question is “What is Machine Learning ?”
Machine Learning can be defined in a variety of ways. Some of the popular definitions of machine learning are as follows:-
- Machine learning is the scientific study of algorithms and statistical models that computer systems use in order to perform a specific task effectively without using explicit instructions, relying on patterns and inference instead. It is seen as a subset of artificial intelligence. Source: https://en.wikipedia.org/wiki/Machine_learning
- Machine learning is an application of artificial intelligence (AI) that provides systems with the ability to automatically learn and improve from experience without being explicitly programmed. Machine learning focuses on the development of computer programs that can access data and use it to learn for themselves. Source: https://www.expertsystem.com/machine-learning-definition/
The above definitions are absolutely correct but they lack simplicity. I would like to define machine learning as the process of teaching the computers to do a specific task by programming it to learn by itself. My definition might not sit correctly with all the norms of machine learning but it’s perfect for someone who doesn’t know what patterns, algorithms or statistical analysis mean.
How humans perform Machine Learning in their everyday lives?
Suppose you want to start reading/writing in English. So this is your specific task. Let this task be called T. To accomplish this task you need to understand the alphabets in English by reading preliminary English books or by watching English videos. Once you start studying you will gain experience in the field of English. Let this experience be called E. Now after a few days, you try to test yourself in English grammar by taking some test. As this was your first test you were not able to perform properly since you were not experienced in successfully passing a test or simply lacked in fundamentals. This cyclic process of studying, gaining experience and testing yourself goes on for a few months. After 1 month you see that sometimes you were able to crack the test and sometimes you failed in the test. Slowly you start improving and your chances of cracking the test increases very rapidly every day. Let P be the performance in test.
Now as we know what T, E, and P we will be able to understand the modern definition of machine learning. Tom Mitchell, the modern definition which is “A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at tasks in T, as measured by P, improves with experience E.” Now as we know what T, E and P we will be able to understand the modern definition of machine learning.
Types of Machine Learning:-
- Supervised Learning
- Unsupervised Learning
- Reinforcement Learning
You already know what supervision means right! , so understanding supervised learning wouldn’t be hard at all. Remember when you were a kid, your parents used a fruit identification book to teach you the different types of fruits. Each fruit had a specific color, shape and a label under its name. So you already know all the correct answers. Here you are the machine and the fruit identification book is the training data set. After teaching you everything from the book under the supervision of your parents, you are given to identify a fruit from the book. Since you as a machine already has the knowledge of previous data you will think wisely. You will first judge the fruit shape, color and other qualitative factors and then put the fruit in the appropriate category. This is what supervised learning means. Easy right! , now let’s dive a bit deeper into supervised learning.
What are the supervised learning techniques?
What is Classification?
Classification is generally used when the problem has discrete outputs like ‘1’ or ‘0’, ’Yes’ and ‘No’. A classification model is generally used to draw some conclusions.
What is Regression?
Regression is generally used to solve prediction problems which have a continuous-valued output such as salary or weight of a person based on some independent factors.
You might be thinking why am I not diving into the more technical details of these techniques and the reason is that if I just jump into the hard math and programming right now you might get frustrated with the topic and might be disinterested in even learning it. I promise you that I will go into deeper aspects of machine learning slowly in my upcoming blogs and this would also share my insights and provide you with resources and good problems to solve.
Unsupervised Learning as the name suggests is the process of learning without any supervision. You might be thinking, “Wait, what does that even mean?”. So this type of learning in which we don’t direct or guide the machine and tell what our data is or in other words, our data don’t have labels. So you might be thinking then how does it even learn? well, that’s the fun part so keep reading. In unsupervised learning the machine groups unsorted information according to some similarities, patterns and differences that it finds in data. We don’t train the machine, unlike the supervised learning.
What are unsupervised learning techniques?
Suppose you are given data of 1000 children who are suffering from different diseases and told to group them without providing any more details. What to do is look at the age of each child, what disease they are suffering from, in which region they are located, which children are facing the kind of symptoms, etc. You try to group them by automatically finding variables to categorize them. This is known as clustering.
Association is used for problems where you have large groups of data and you try to find the trends between the data. What you actually do is associate one data to similar data and group them. This is known as an association.
Reinforcement Learning is similar to supervised learning except in a few ways. Reinforcement learning works on the principle of maximizing the reward for each suitable action. Suppose you are trying to solve a math problem and your reward for solving the problem is candies. First, you need to understand the question then proceed through the right process. Every mistake you make in trying to solve the problem would reduce your number of candies by 1 and every correct process you choose would increase your reward by 1. The total reward would be calculated after you solve the problem. And the natural part of this type of learning is that the learning agent would always want to maximize its reward by minimizing its mistakes.
I think by now you have a basic understanding of what machine learning really is and how it is similar to our learning process.
This is all for this blog. In the upcoming blogs, I would explain each process and steps of machine learning in details. Be sure to check them out. Have a good day, bye.
Originally published at http://computingbytes.wordpress.com on July 3, 2019.