If documentation, privacy notices or product settings leave you wondering what an AI or web-data term means, this glossary translates the most common vocabulary into practical language. It separates related concepts—such as a dataset and a model, or a cookie and browser storage—so you can see what each one describes, where it operates and what it produces.
Definitions can change with context. The descriptions below use Google’s Machine Learning Glossary, Google’s Privacy & Terms key-terms glossary, the MDN Web Docs glossary and standards-oriented terminology compiled in ITU-T Y Supplement 97 (November 2025). Google’s privacy explanations are published guidance, not legal or universal definitions; standards terms retain their stated source and scope.
As an Amazon Associate I earn from qualifying purchases.
AI foundations
Artificial intelligence (AI)
Artificial intelligence is the broad area of computing concerned with systems that perform tasks associated with human intelligence, such as recognizing patterns, understanding language, making decisions or generating content. “AI” is an umbrella term rather than one single method. A rules-based fraud check, a speech recognizer and a text generator can all be described as AI even though they work differently.
Free tools Windows power users keep installed
One-click scans. No signup required.
Google’s Machine Learning Glossary describes machine learning as a subfield of AI. In everyday product language, however, organizations sometimes use “AI” and “machine learning” interchangeably. When precision matters, ask which technique, data and output are actually involved.
Machine learning (ML)
Machine learning is an approach in which a system uses data and an algorithmic procedure to build or adjust a model, rather than relying only on hand-written rules. The resulting model can apply what it learned to new inputs. ML is therefore narrower than AI in the formal relationship used by Google’s glossary, although the labels overlap in marketing and general conversation.
Algorithm
An algorithm is a defined process for carrying out a task. Google’s Privacy & Terms key-terms glossary defines it as “a process or set of rules followed by a computer in performing problem-solving operations.” An algorithm can sort records, calculate a score, train a model or decide which web resources to retrieve. The word describes the procedure, not necessarily the data or the result.
Automation, AI system and application
Automation means a task is carried out with limited human action after it has been configured. An AI system is an application that uses one or more AI techniques; an application is the broader software product in which that system operates. A customer-support application might contain an ML model, an algorithm for routing requests and ordinary code for authentication and billing.
Data, datasets and models
Data
Data is recorded information that a program can store or process. It may be text, numbers, images, audio, video, events or metadata. “Raw data” means the material before it has been cleaned, transformed or summarized for a particular use.
Dataset
A dataset is a collection of raw data. It can be a spreadsheet of transactions, a folder of photographs, a table of sensor readings or a set of text documents. A dataset may be divided into subsets for different stages of work, but the term itself refers to the collected material—not the mechanism that learns from it.
Model
A model is a structure and a set of parameters used to process inputs and produce an output such as a prediction. In machine learning, training adjusts parameters so that the model captures useful relationships in its training data. A model is not the same thing as the dataset: the dataset is the material supplied to the process, while the model is the resulting computational mechanism.
| Term | What it describes | Typical role |
|---|---|---|
| Dataset | Collected examples or records | Supplies inputs for analysis or training |
| Model | Structure plus parameters | Transforms an input into a prediction, classification or other output |
| Algorithm | Process or rules | Specifies how a task is performed, including how a model may be trained |
| Output | Result produced by a procedure or model | Can be a score, label, prediction, generated text or another value |
Feature
A feature is an input characteristic used by a model. For a house-price model, floor area or number of bedrooms could be features. In an image or language system, features may be represented internally rather than entered as a neat column. The useful question is what information the model receives and how that information is encoded.
Recommended Free Tools
Label and ground truth
A label is the target value associated with an example, such as “spam” for an email or the known temperature at a measurement time. In supervised learning, the model uses examples paired with labels. “Ground truth” usually means the reference answer used to assess or train a system; its accuracy depends on how that reference was obtained.
Training
Training is the process of adjusting a model with data so it performs a desired task. A training procedure typically evaluates the model’s current output against a target or objective and updates parameters. Training data is the portion used for this adjustment. Training can be computationally expensive, but the word alone does not specify a particular model size, hardware setup or duration.
Inference
Inference is a trained model’s use of new input to produce an output. Asking a text model to answer a question, sending a photograph to an image classifier or scoring a transaction are inference operations. Inference is distinct from training even when both use the same software service.
Validation and test data
Validation data helps choose settings or compare versions during development. Test data is held back for a final estimate of performance on examples not used to fit the model. Keeping these roles separate helps reveal whether a model has learned a general pattern rather than simply memorized its training material.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →Overfitting and generalization
Overfitting occurs when a model matches its training examples too closely and performs poorly on new cases. Generalization is the ability to remain useful on those new cases. A reported score is meaningful only when you know which data was used, how it was labeled and whether the evaluation set was kept separate.
Common machine-learning approaches
Supervised learning
Supervised learning trains from examples that include a target label or value. Classification predicts a category, such as a document type; regression predicts a numerical value, such as demand. The approach depends on having labels that are sufficiently accurate for the intended task.
Unsupervised learning
Unsupervised learning works with data without a supplied target. A method may group similar records, identify structure or compress representations. The discovered groups are not automatically meaningful; a person still has to interpret what a cluster represents.
Reinforcement learning
Reinforcement learning trains an agent through interactions in which actions receive rewards or penalties. The objective is to learn a policy for choosing actions over time. A reward is a design signal, not a guarantee that the resulting behavior matches every human preference.
Neural network
A neural network is a model built from connected computational units arranged in layers. During training, its parameters are adjusted so the network maps inputs to useful outputs. “Neural” describes the mathematical architecture; it does not mean the software is a biological brain.
Deep learning
Deep learning generally refers to neural networks with many processing layers. The additional layers can learn successive representations, which is useful for tasks involving images, audio, language and other complex inputs. The term does not by itself state how reliable, large or current a system is.
Generative AI
Generative AI produces new content—such as text, images, audio, video or code—rather than only assigning a label to an existing item. A generated result is an output of a model and should not be assumed to be a verified fact. The quality and appropriate safeguards depend on the model, its data and the application around it.
Large language model (LLM)
A large language model is a model designed to process and generate language. It can estimate likely sequences of tokens and use that capability to produce passages, summaries or code. “Large” is relative to a particular model family and does not, on its own, establish accuracy, recency or reasoning ability.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Web technology terms
MDN’s Web Docs glossary explains jargon, acronyms and abbreviations encountered in web documentation and related contexts; its glossary page reported a last modification date of July 11, 2025. The following terms describe different parts of a web request, page or browser environment.
Browser
A browser is software that requests web resources and presents documents or applications to a person. It can send information with a request, run scripts, maintain local data and enforce security controls. “Browser” may mean the application or, informally, the device session in which it runs.
HTTP request and response
An HTTP request is a message a browser or other client sends to a server for a resource or action. The server returns an HTTP response containing a status, headers and, when appropriate, content. Cookies, referrer information and caching instructions can appear in request or response headers.
URL
A URL (Uniform Resource Locator) identifies where a resource can be accessed, such as a page, image or API endpoint. A URL can include a scheme, host, path and query parameters. The complete URL may reveal information supplied by a link or search operation, so it can be personal or sensitive in some contexts.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Repair Windows errors before they cause bigger problems3Fix the driver behind crashes, sound loss and screen glitchesDomain and server
A domain is a human-readable name used to locate an internet service. A server is a computer or software service that receives requests and returns resources or performs an operation. One domain can use multiple servers, and one server can host resources for multiple domains.
API
An application programming interface (API) is a defined way for software components to request data or actions from one another. A web API commonly uses HTTP and structured responses. Documentation specifies the available operations, inputs, authentication requirements and response format.
HTML, CSS and JavaScript
HTML describes the structure and meaning of a web document. CSS controls presentation such as layout, color and typography. JavaScript is a programming language commonly used to add behavior in the browser, including responding to events and making background requests. A site can use all three, but they are separate technologies.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Cookies and browser-side storage
Cookie
Google’s Privacy & Terms key-terms glossary defines a cookie as: “A cookie is a small file containing a string of characters that is sent to your computer when you visit a website.” A site can send a cookie to a browser, which may return it on a later visit so the site can recognize a browser or preserve a setting. Cookies can support sessions, preferences or measurement; their exact purpose depends on how a site uses them.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchFirst-party and third-party context
“First-party” generally refers to data associated with the site a person is visiting, while “third-party” refers to another domain involved in the page or request. The distinction describes the relationship between the browser and domains in that context, not a universal judgment about whether a cookie is safe or harmful. Browser policies and privacy controls can limit third-party behavior.
Browser web storage
Browser web storage is data saved by a site in the browser, commonly through mechanisms such as local storage or session storage. It is separate from cookies: storage entries are not simply cookie files, and they are accessed through different browser interfaces. Storage can preserve preferences or application state, but the details of lifetime, access and transmission depend on the mechanism.
Cache
A cache is a temporary copy of resources kept so they can be reused without fetching them again. Browser caches can improve speed and reduce repeat downloads. Cached content is different from a cookie or web-storage entry: a cache primarily stores a resource copy, not a small identifier intended to be sent as a site-specific value.
Privacy, measurement and tracking terms
Pixel tag
A pixel tag is a small, often invisible resource or piece of code embedded in a page, message or application. When it loads, it can signal that the content was viewed and may send associated information to a service. A pixel tag is not the same mechanism as a cookie; it is a way to trigger a request or record an event, while a cookie is data stored and potentially returned by the browser.
Referrer URL
A referrer URL is the address of the page that led a browser to request another resource. It can help a site understand navigation or diagnose links. Depending on browser and site policy, the referrer may be reduced or omitted, and a URL can contain identifying or sensitive query information.
Server log
A server log is a record created by a server while handling requests or other events. Entries may include times, requested paths, response statuses, technical identifiers and diagnostic details. Logs are records kept on the server side; they are not browser cookies, web-storage entries or pixel tags, although a single visit can generate all of them.
Analytics
Analytics is the collection and interpretation of information about activity, such as page requests, feature use or errors. An analytics system may combine server logs, cookies, pixel requests, browser storage and application events. “Analytics” describes the purpose or practice, not one particular technology or a guarantee that the data is anonymous.
Personal data and pseudonymous data
Personal-data definitions depend on the applicable law and context. In ordinary technical discussion, information is personal when it relates to an identifiable person; pseudonymous data has identifiers replaced or separated but may still be linkable. Do not treat a technical label as a legal conclusion. A privacy notice should explain what a particular service collects, why, how long it keeps it and with whom it shares it.
Consent and privacy controls
Consent is a person’s permission for a stated processing purpose, where the relevant legal framework requires it. A browser setting, cookie banner choice or account preference may control some collection, but the effect depends on the site and jurisdiction. Read the site’s explanation of which technologies are optional, which are required for a requested feature and how to change the choice.
Standards and source context
Why definitions differ
A term can have an everyday meaning, a product-specific meaning and a standards-defined meaning. Google’s Machine Learning Glossary is a broad technical reference; MDN focuses on web documentation; Google’s privacy glossary presents accessible explanations of privacy and web-data terms. ITU-T Y Supplement 97, published in November 2025, compiles machine-learning terminology and points to definitions from external standards, including ISO/IEC. When a standards-based definition matters, identify that standard instead of presenting one wording as universal.
Quick Recap
How to read a glossary entry safely
- Identify what the term describes: a collection of data, a procedure, a model, a browser mechanism or a server record.
- Check where it operates: inside a model, in the browser, in a web request or on a server.
- Ask what goes in and what comes out: examples, identifiers, predictions, content or event records.
- Look for the source, date and jurisdiction when the term affects privacy, compliance or a technical standard.
- Separate a capability claim from a performance claim; a system can generate an output without that output being correct.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
