Suppose you are invited by your friend to join a party. Everyone in the party is a stranger to you. Your friend who invited you is busy talking to someone else. You determine to make new friends here. Since you have no idea about anyone, one possibility is just to grab any random person to start the conversation. However, that idea might make you feel shy and too risky. You may come up with some characteristics of the right person in the room to make friends, which can be on the basis of gender, age group, dressing, etc.
In this context, the person that you are going to talk to represents unlabeled data. You are in the process of classifying data points, which are people in the room, into different groups. The process of classifying unlabeled data is unsupervised machine learning, which is a machine learning problem.
So what is machine learning? Machine learning examines the study and construction of algorithms that can learn from and make predictions on data. The ultimate goal of machine learning is to learn a target function that can be used to predict the values of another unknown or complex function.
Suppose we have three basic components:
- Set of tasks (T)
- A performance measure (P)
- Data describing some experience (E)
A computer program learns if its performance at tasks in T, as measured by P, improves based on E.
From: Tom M. Mitchell, Machine Learning (1997)
What does it mean to learn?
Let’s take another example. Anne is taking a course on machine learning. She knows that by the end of the course, she should have “learned” about all topics covered. But how? She will be given an exam to measure her knowledge on the course. But what should the exam be? If the exam covers math or political science topics, Anne’s performance on that exam would not be representative of her learning. On the other hands, if the exam only covers the exact questions that were already answered during the lectures, that exam is not representative either. So what is considered a good exam? It should make Anne think of specific examples that she learned through the course, but to answer new and related questions on the exam. The test gauges Anne’s ability to generalize.
Machine learning provides systems the ability to automatically learn and improve from experience, without being explicitly programmed. It focuses on the development of computer programs that can learn from data by themselves.
Although there are various definitions and explanations of machine learning, hopefully you get a broad sense of what machine learning is. We will deep-dive into some specific use cases and machine learning problems to fully understand its applications.
Image credit: Dassault Systems
Data Science How To 