Plain-language definitions of the vocabulary used across machine learning, model architectures, tooling and AI safety. Every entry opens with a one-sentence definition, then explains where the term is used.
Multi-Agent Reinforcement Learning is a branch of machine learning where multiple agents learn and interact within a shared environment to achieve individual or collective goals through trial and error.
Multi-Agent System is a network of autonomous agents that interact and collaborate to solve complex problems or perform tasks.
Multi-agent systems are AI systems where multiple autonomous agents interact, cooperate, or compete to achieve individual or shared goals.
A Multi-Armed Bandit is a problem in probability theory and machine learning where a fixed limited set of resources are allocated between competing choices to maximize their expected gain.
Multi-Factor Authentication is a security process that requires users to provide two or more distinct forms of identification before gaining access to a system or account.
Multi-frame processing is a technique that combines data from multiple consecutive frames to enhance image quality and extract more information.
Multi-Head Attention is a mechanism used in neural networks to improve performance by allowing the model to focus on different parts of the input sequence simultaneously.
Multi-Instance GPU (MIG) is a technology that partitions a single physical GPU into multiple isolated instances, allowing simultaneous, independent workloads.
Multi-Modal AI is artificial intelligence that processes and integrates multiple types of data inputs, such as text, images, and audio, to improve understanding and decision-making.
Multi-model AI is an artificial intelligence system designed to process and integrate multiple types of data such as text, images, and audio to generate comprehensive insights or outputs.
Multi-Object Tracking (MOT) is a computer vision process that involves simultaneously tracking and identifying multiple objects within a video sequence or real-time environment.
Multi-Speaker Recognition is the process of identifying and distinguishing multiple individual speakers within an audio recording or live conversation.
Multi-turn conversation is an interactive dialogue where multiple exchanges occur between a user and a system to achieve a goal or clarify information.
A Multi-Word Expression is a combination of two or more words that function as a single unit of meaning.
Multicollinearity is a statistical phenomenon where two or more predictor variables in a multiple regression model are highly correlated.
Multilingual AI is artificial intelligence designed to understand, process, and generate content in multiple languages seamlessly.
Multilingual BERT is a pre-trained language model designed to understand and process multiple languages simultaneously.
A multilingual chatbot is an AI-powered conversational agent designed to understand and respond in multiple languages to serve diverse user groups.
A Multilingual Model is an artificial intelligence system designed to understand and generate text in multiple languages.
Multilingual NLP is the field of natural language processing focused on building AI systems that understand, interpret, and generate multiple human languages using shared or language-specific models.
Multimedia embedding is the process of integrating various types of media such as videos, audio, images, or interactive content directly into digital platforms or web pages.
Multimodal AI is a type of artificial intelligence that can understand, process, and combine multiple forms of data such as text, images, audio, and video.
Multivariate testing is a method of simultaneously testing multiple variables on a webpage to determine which combination performs best for user engagement and conversions.
Muse is a source of inspiration, often a person or idea, that sparks creativity and drives artistic or intellectual work.
Page 10 of 11
The TiorAI AI Glossary is a free reference of 3,885 artificial intelligence terms, each defined in plain English. Every entry opens with a one-sentence definition that stands on its own, then explains where the term is actually used and how it sits next to the ideas around it. There is nothing to sign up for and nothing to download: each definition has its own page and its own URL, so a term you needed once can be linked to and found again.
They combine, and none of them cancels the others. The A to Z strip narrows the index to a single initial, with the # key collecting the entries that begin with a digit or a symbol and that no letter would reach. The category filter cuts the glossary to one of sixteen subject areas, among them Machine Learning, Deep Learning and Neural Networks, Natural Language Processing, Computer Vision, Generative AI, Reinforcement Learning, AI Agents and Systems, MLOps and Infrastructure, Data and Analytics, and AI Ethics. The search field matches names and definitions, so a concept you can only half-name still surfaces. Every combination has its own URL, which means a narrowed view can be bookmarked or sent to a colleague.
The vocabulary of modern AI, at the three depths people arrive needing. The fundamentals a first week runs into — model, parameter, training, inference, overfitting. The architecture behind current systems — transformers, attention, embeddings, diffusion, retrieval-augmented generation. And the operational and governance language that turns up in a procurement review or a policy document — fine-tuning, hallucination, alignment, model drift, bias. Terms are defined as they are used in practice rather than as they appear in a paper abstract.
Each entry leads with the answer. The opening sentence says what the term means without assuming the rest of the page, which is what makes an entry useful to someone who arrived from a search with exactly one question. The explanation underneath is for the reader who wants more than that: what the idea is commonly confused with, what it looks like in practice, and which neighbouring terms are worth reading next. The card in the index above shows the opening of that definition, so you can tell from the results whether an entry is the one you want.
A glossary answers what something means; it is not the place that tells you what to do about it. Where a concept has products behind it, the AI tools directory is where they are listed, and where it has a working instruction behind it, the prompt library is where that lives. Entries link out to those when the link earns its place, and the glossary stays what a glossary is — the definitions, ordered so they can be browsed, and written so a single one can be read on its own.
Alphabetically by term name, which is the order a reference work is browsed in and the order every page of this archive has always used. The sort menu can switch it to reverse alphabetical, or to recently added and recently updated if you want to see what has changed.
It narrows the index to terms whose name begins with that character, and it keeps whatever else you have set — a category filter, a search, a sort — so you can move between letters without starting again. The # entry collects the terms that begin with a digit or a symbol, which no letter would reach.
Entries that span several areas, or that predate the current category set, sit under “Other”. It holds a large share of the glossary, so the filter is most useful for narrowing to a named topic rather than for excluding one.
A short definition in plain language, then a longer explanation of where the term is used and how it relates to neighbouring concepts. The card above shows the opening of that definition; the entry itself also links to related terms and, where relevant, to the AI tools that implement the idea.
The field at the top matches term names and definitions, so a concept you can only half-name will usually surface. Search combines with the category filter and the letter strip, and every combination has its own URL, so a narrowed view can be bookmarked or shared.
