Machine learning. Business and tech spheres are inundated with this buzzword that is thrown around with the intent of driving forward concepts and ideas, but they frequently miss the mark when it comes to demonstrating actual understanding of the topic. And while a generalized definition can be taken from the two words as they are, it may be useful to establish a more straightforward idea of what machine learning actually is and explore the differences between the various subsets of machine learning and how they are categorized.

The term machine learning was first coined by Arthur Samuel, who largely spearheaded the field of artificial intelligence starting in the late ‘40s. By his definition, “Machine learning is a field of computer science that gives computers the ability to learn without being explicitly programmed.” More specifically, it is the field where algorithms are designed and constructed to take data and predict specific outcomes through the use of various modeling techniques. These algorithms develop these predictions by creating a mathematical model derived from the data provided to them. As the quantity of input data increases, the aim is for the accuracy of the algorithms to improve as well. 

The process of breaking down data and drawing conclusions can be generalized into two broad categories: supervised versus unsupervised learning. With supervised learning, the input data initially comes in the form of training data (a sample data set that is selected to fit the parameters of the model). From this data, the algorithm generates an inferred function that can help map predictions regarding the output values. These output predictions generated by the algorithm can then be compared to the correct, intended output which helps direct any modifications to the model. Ultimately, the specific, preset parameters of the training data is what helps differentiate supervised versus unsupervised learning. Some basic examples of supervised learning are classification, where the algorithm is given the correct classification, and regression, where the algorithm is given the correct output value.

Unsupervised learning is when the algorithm generates an inferred function from unlabeled data. The idea is that the function will describe hidden structures within the unclassified data. The predictions themselves aren’t judged on an accuracy metric. Rather, the goal is to allow the algorithm to identify structures within the data that might not have originally been discernable. An example of unsupervised learning is cluster analysis, where the general task is to group a cluster of objects so that objects more closely related to each other are placed nearer to one another.

A third category, semi-supervised learning, is a sub-class of supervised learning. As the name implies, the input data is a combination of both labeled and unlabeled training data. The vast majority of the data is unlabeled, with a small percentage of the labeled data thrown into the mix.

Machine learning is a concept that has been relevant for quite a few decades. However, with the advent of improved computational power and the availability of enriched data sources, we are seeing a resurgence of the implementation of machine learning in various business, technological, and data-driven industries. We live in a world where every aspect of their lives is saturated with metrics and data. With machine learning, much of this data can be mapped into useful information that will further improve efficiency in business and technology and the quality of human life.