Machine learning is a branch of artificial intelligence that allows computers to learn patterns from data and make predictions or decisions without being explicitly programmed for every task. It relies on algorithms that adjust internal parameters to minimize errors on training examples, then generalize to new, unseen data.

Supervised learning uses labeled data to learn a mapping from inputs to outputs. This includes regression for predicting continuous values and classification for assigning discrete categories. Unsupervised learning works with unlabeled data to uncover structure, such as clusters or low dimensional representations. Reinforcement learning focuses on agents that learn to act in an environment by maximizing cumulative reward through trial and error.

Modern machine learning often uses neural networks, including deep architectures with many layers that can automatically discover complex features. Convolutional neural networks excel at image tasks, while recurrent and transformer based models handle sequences like text and speech. Training these models typically requires large datasets and significant computational resources.

Key steps in a typical workflow include collecting and cleaning data, selecting and engineering features, choosing an appropriate model, training and validating it, then evaluating performance with metrics suited to the problem. Overfitting is a central concern and is mitigated with techniques such as regularization, dropout, and cross validation.

Machine learning has wide applications in computer vision, natural language processing, recommendation systems, medical diagnosis, finance, and scientific discovery. At the same time, it raises issues around transparency, bias, and robustness, motivating research on interpretability, fairness, and methods that can provide reliable uncertainty estimates.