The practical AI and ML glossary.
Definitions written for people making decisions about software, data, machine learning, and AI—not for people trying to win a jargon contest.
ARIMA
A family of statistical time-series models that uses autoregression, differencing, and moving-average errors to forecast future values.
Bayesian regression
Regression that represents model parameters as probability distributions, combining prior information with observed data.
Clustering
Unsupervised methods that group observations by similarity so analysts can explore structure and design different actions.
Embeddings
Numeric representations that place semantically similar items near one another for search, recommendation, and classification.
Forecast backtesting
Repeatedly training on historical cutoffs and evaluating later periods to estimate how a forecasting system will behave in production.
Incrementality
The change caused by an intervention compared with what would have happened without it.
Large language model
A probabilistic model trained on large text collections to predict and generate sequences of tokens.
Marketing mix model
A statistical model that estimates how media and business drivers contribute to an outcome over time.
Monte Carlo simulation
Repeated random sampling used to estimate a range of possible outcomes and the uncertainty around them.
Prophet
An additive forecasting approach designed for business time series with trend, seasonal, holiday, and event effects.
Retrieval-augmented generation
A pattern that retrieves approved source material and supplies it to a generative model at request time.
Recommendation system
A system that ranks products, content, or actions for a user or context using behavior, attributes, and feedback.
Synthetic data
Artificially generated records designed to preserve useful statistical patterns without directly reproducing source records.