Section 5.1 An Overview of Machine Learning

Machine learning was created in order to replicate aspects of this capability. As a major subfield of artificial intelligence (AI), the overarching goal of machine learning is to understand the structure of data and fit that data into models that can be understood, interpreted, and utilized by people.

The field of machine learning is concerned with the question of how to construct computer programs that automatically improve with experience. This definition captures the essence of the field: rather than programming every possible scenario, we design systems that learn from data and experience.

Machine learning is one of the most exciting areas of computer science because it enables computers to perform tasks that would be impossible to program manually. For example, how would a programmer write explicit rules to recognize a cat in any possible photograph? The cat might be sitting, standing, sleeping, or hiding. The lighting might be bright, dim, or shadowed. The cat might be seen from the front, side, or back. Writing rules for every possible scenario is impossible. Machine learning solves this problem by letting the computer discover the patterns on its own.

The idea of machines that can learn has been around since the early days of computing. In 1959, Arthur Samuel defined machine learning as a field of study that gives computers the ability to learn without being explicitly programmed. Samuel created a program that learned to play checkers by playing thousands of games against itself—an early example of what we now call reinforcement learning.

Since then, the field has grown enormously. Early machine learning systems were limited by the amount of available data and computing power. Today, we have access to vast amounts of data and powerful computers, which has enabled breakthroughs in areas like image recognition, language translation, and medical diagnosis.

Although machine learning is a core field within computer science, it differs fundamentally from traditional computing. In traditional computing, algorithms are sets of explicitly programmed instructions written by human developers to calculate solutions or perform specific tasks. Machine learning algorithms, by contrast, allow computers to train on data inputs and use statistical analysis to output predictions or decisions. Rather than requiring explicit instructions for every edge case, machine learning enables computers to build models from sample data in order to automate decision-making processes.

At the heart of every machine learning system is data. Machine learning algorithms learn from examples. The quality and quantity of the data used for training directly affect how well the algorithm performs. This is often summarized as "garbage in, garbage out"—if the training data is incomplete, biased, or incorrect, the resulting model will reflect those problems.

This is why data collection and preparation are such important steps in any machine learning project. Data scientists spend a significant amount of time cleaning and organizing data before they even begin training models.

Any technology user today benefits daily from machine learning applications:

There are many different problem classes in machine learning, categorized primarily by the type of data provided and the specific conclusions to be drawn from that data. In the following sections of this chapter, we will explore three foundational paradigms: supervised learning, unsupervised learning, and reinforcement learning.