Machine Learning (ML)

Machine learning (ML) is the branch of artificial intelligence (AI) in which computer systems learn patterns from data instead of following rules written by a programmer. A spam filter built with ML is never told which words signal spam; it studies thousands of emails labeled spam or not spam and learns the patterns itself, then applies them to new messages. Arthur Samuel coined the term in 1959 while building a checkers program that improved by playing against itself, and Tom Mitchell later gave a widely cited definition: a program learns when its performance at a task, measured by some metric, improves with experience.

How Machine Learning Works#

An ML model is a function with adjustable parameters, from the few coefficients of a linear regression to the millions of weights in a neural network. Training searches for the parameter values that make the model's predictions on the training data as accurate as possible, usually by minimizing a loss function. The real test is generalization: the model is then scored on held-out validation data and test data it never saw during training. A model that scores well on training data but poorly on new data is overfitting, memorizing examples instead of learning the underlying pattern.

This scikit-learn example trains a random forest to read handwritten digits and scores it on digits held out from training:

from sklearn.datasets import load_digits
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split

# 1,797 labeled 8x8 images of handwritten digits bundled with scikit-learn
X, y = load_digits(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.25, random_state=0)

# Fit learns decision rules from the labeled examples; no rules are hand-written
model = RandomForestClassifier(random_state=0)
model.fit(X_train, y_train)

# Score on digits the model never saw during training
print(f"Test accuracy: {model.score(X_test, y_test):.1%}")

The model reaches 97.6% accuracy on the 450 test digits, using only patterns it learned from the other 1,347 labeled examples.

Types of Machine Learning#

• Supervised learning: learns from labeled examples, each input paired with the correct output. Classification, regression, and object detection are supervised tasks.
• Unsupervised learning: finds structure in unlabeled data, such as customer segments found by k-means clustering.
• Self-supervised learning: creates its own labels from the data, for example by predicting the next word in a sentence or a hidden patch of an image. It is how large language models (LLMs) and other foundation models are pretrained.
• Reinforcement learning: an agent learns by trial and error, taking actions in an environment and adjusting its behavior to maximize a reward.

Machine Learning vs. AI, Deep Learning, and Data Science#

Machine learning sits between AI, the broad goal of building intelligent machines, and deep learning (DL), the subset of ML that uses many-layered neural networks and learns its own features from raw data:

Classical algorithms such as random forests and XGBoost remain strong on tabular data, where each column, such as price or age, is already a meaningful feature. Data science is a neighboring discipline that uses ML as one tool among many, alongside statistics, data visualization, and domain analysis, to answer questions from data.

The Machine Learning Lifecycle#

A model in production is never finished. Teams collect and label data, train and validate a model, deploy it, and monitor its predictions, returning to the data whenever accuracy falls short or the real world changes. This loop is the subject of machine learning operations (MLOps):

For computer vision, Ultralytics Platform runs this whole loop in one place: teams upload and annotate datasets, train Ultralytics YOLO26 models on cloud GPUs, deploy them, and monitor their performance in production.

Real-World Applications#

• Computer vision: models detect defects on production lines, read medical scans, and guide autonomous vehicles.
• Natural language processing (NLP): translation, search, chatbots, and spam filtering all rely on models trained on text.
• Recommendation systems: streaming and e-commerce platforms rank content for each user from their past behavior.
• Time series forecasting: models predict product demand, energy load, and equipment failures from historical data.
• Anomaly detection: banks score every card transaction in milliseconds with models trained on past fraud, flagging the ones that look unusual.