Machine learning
Machine learning (ML) is a subfield of artificial intelligence focused on systems that improve their performance on a specific task through experience with data, without being explicitly programmed for every rule. Rather than relying on fixed, hand-coded instructions, machine learning algorithms build statistical models from training data and apply them to make predictions or decisions on new, unseen data. This capacity to generalize from examples distinguishes ML from traditional rule-based programming and underpins its widespread use across scientific, industrial, and consumer applications.
Machine learning approaches are commonly grouped into several categories depending on the nature of the learning signal available or supposed to be present. Supervised learning algorithms learn a mapping from inputs to known outputs from labeled training data; unsupervised learning discovers hidden structure or patterns in unlabeled data; and reinforcement learning trains an agent to make sequential decisions by maximizing a cumulative reward signal. Within these broad categories, a wide range of techniques — including decision trees, support-vector machines, and deep learning architectures — address different kinds of problems.
Supervised and unsupervised learning
In supervised learning, the training dataset consists of input–output pairs. The algorithm learns to predict the output for new inputs by finding a function that minimizes error on the known examples. Common supervised tasks include classification, where the output is a discrete category, and regression, where the output is a continuous value. Supervised learning underpins many practical systems, from email spam filters to image-recognition pipelines.
Unsupervised learning operates on data without labeled answers. The algorithm must infer the natural structure present in the inputs, such as grouping similar instances into clusters or reducing the dimensionality of the data while preserving its essential variation. Applications include customer segmentation, anomaly detection in sensor streams, and feature extraction for downstream tasks.
Reinforcement learning
In reinforcement learning (RL), a learning agent interacts with an environment by taking actions and receiving feedback in the form of rewards or penalties. The agent's objective is to learn a policy — a mapping from states to actions — that maximizes the expected cumulative reward over time. RL is particularly suited to problems involving sequential decision-making, such as game-playing systems, resource allocation, and the control of robotic systems where an agent must learn to perform complex motor tasks or navigate uncertain environments.
Generalization and applications
A central concern in machine learning is generalization: the ability of a model trained on a finite dataset to perform well on previously unseen data. Overfitting — when a model memorizes training examples rather than learning the underlying patterns — is a common challenge addressed through techniques such as regularization, cross-validation, and the careful partitioning of data into training, validation, and test sets.
Machine learning has become integral to robotics, where it enables robots to learn motor skills, recognize objects and scenes, and adapt to dynamic environments. It is also widely used in computer vision, natural-language processing, speech recognition, and predictive analytics across healthcare, finance, and manufacturing sectors.