What is Machine Learning?

Artificial intelligence (AI) is a term that is getting a lot of buzz. It is a field that aims to build intelligent machines (computers) that can act and think humanly and rationally [1]. For a computer to be considered an intelligent entity, it must possess one of the four intelligent capabilities which is machine learning. Machine learning is a science of making computers to act without being explicitly programmed. How do computers act without being programmed? We, human beings learn to do things from experience. Similarly, we can make computers learn from data. So machine learning is a study that provides the computers to learn from data and make predictions or decisions without being explicitly programmed. For example, to distinguish spam emails from legitimate emails, we can compile thousands of examples of legitimate and spam emails. Then, we “teach” the computer by providing it the examples to learn what constitutes spam emails. Specifically, the computer automatically learns to detect spam messages by recognizing the patterns in the data that represent legitimate and spam emails. In other words, we do not have to know build the algorithm to process the input in order produce the output. Machine learning deals with the data and automatically build the algorithm that maps the inputs to outputs.

In typical machine learning problems, we are given a dataset that consists of a set of input-output pairs. The aim is to predict the outputs given the input values. This exercise is known as supervised learning or predictive analytic. The inputs or often called attributes, features or covariates can be anything, but typically in the form of d-dimensional vector of numbers, representing characteristics of an entity such as word count (in the above example), height or weight. Similarly the output or often called response may be in any form, but in general the output is either categorical or nominal variable (yes or no, or types of vehicles) or numeric value where some values are smaller or bigger than others. When the output is categorical, the problem is known as classification and when the output is numeric value, problem is known as regression. Figure 1 and Figure 2 illustrate examples of classification and regression problems.

Figure 1: A classification example in two dimensions. The output is categorical variable (PURPLE=0, YELLOW=1) and the attributes are sepal length and sepal width.

 

Figure 2: A regression example where the output is numeric values (house price) and the attribute is number of rooms.

The other type of machine learning is known as unsupervised learning or descriptive analytic. In contrast to supervised learning, here we are only given the inputs and the aim is to discover the underlying structure in the input space. The underlying structure refers to the interesting patterns that occurs in the data that tells what generally happens and what does not. A method to discover patterns in data is known as clustering where the aim is to find clusters or grouping of inputs. Figure 3 illustrates an example of clustering problem.

Figure 3: A clustering example in two dimensions. Two groups of samples can be clearly seen that may represent two distinct classes.

[1] S. J. Russell and P. Norvig, Artificial intelligence: a modern approach. Malaysia; Pearson Education Limited, 2016.

Leave a Reply