Documentation
What is Machine Learning?

What Is Machine Learning?

Machine learning is a way to build software from examples. Instead of writing every rule by hand, you give an algorithm data and a goal; training finds a model that can make useful predictions on new data.

For example, rows of past support messages and their correct departments can train a classifier. After training, a new message goes into the saved model and a predicted department comes out. The important question is not whether the model remembers the training rows; it is whether it works on representative rows it did not see during training.

Five useful terms

TermPlain-language meaningExample
ExampleOne case the model can learn from or predictOne customer, image or message
FeatureInformation available when the prediction is madeAccount age, words in a message
Target or labelThe answer to learn during supervised trainingwill_churn, billing_question
ModelThe learned relationship between inputs and outputsA CatBoost classifier
InferenceUsing a trained model on new inputPredicting the next support message

The main types of machine learning

Supervised learning

Supervised learning uses examples with known answers. Classification predicts a category; regression predicts a number. Most AnyLearning training projects (including image classification, object detection, keypoints, Tabular AI and Text AI classification) are supervised.

Unsupervised learning

Unsupervised learning looks for structure without a target label. Clustering, dimensionality reduction and anomaly discovery are common examples. They can be useful for exploration, but a cluster is not automatically a business category and an outlier is not automatically an error. AnyLearning does not currently train an unsupervised model.

Reinforcement learning

Reinforcement learning learns a policy by taking actions and receiving rewards over time. It is used in areas such as control, games and sequential decisions. It needs a carefully designed environment and reward, not just a spreadsheet of independent rows, and is outside AnyLearning's current project types.

Where generative AI and LLMs fit

A large language model is a model trained to generate or transform sequences of text. It is machine learning, but machine learning is much broader than LLMs. AnyLearning's Text AI classifier is classical supervised NLP; its search is a lexical baseline; its response evaluator measures saved output from any source. Those workflows do not load or train a generative LLM.

A model is one part of a system

Useful machine learning also needs representative data, an honest evaluation, a review path, and a plan for mistakes. Start with the decision the prediction will support, then choose the workflow, not the other way around.

Continue with Choosing a machine-learning task, then learn how to train, validate and test without leakage.