AI Glossary: Find and Understand AI Terms

A field guide to the language of AI

Find a term, understand the idea, and follow it further.

Start with a short definition of an AI concept or an industry term you encounter in AI coverage. Linked names open deeper explanations with context and examples.

Find a term · Browse A–Z · New to AI? Start here

An orderly index of ivory and teal cards, with a magnifying glass highlighting an amber card beside connected glass nodes. — Find the concept. Make the connection. A conceptual illustration.

Find a term

Browse all terms by letter — choosing a letter clears the search.

A few places to begin: AI · Machine learning · LLM · Token · RAG · Hallucination. Use the # beside a term to bookmark its definition.

A

Back to finder

Account Reconciliation #: Matching balances, transactions, or supporting records across systems so discrepancies and unresolved exceptions can be found and cleared.
Account Takeover #: When an attacker gains control of a legitimate user's account and begins acting as that user.
Activation Function#: A function applied to a neuron’s input to produce its output. Nonlinear activation functions help neural networks learn complex patterns.
Active Learning #: A training strategy in which the model asks for labels on the most informative examples instead of labeling everything at once.
Active Noise Control (ANC)#: Reducing unwanted sound by generating anti-noise through microphones, speakers, and control algorithms.
Additive Construction #: Using digitally controlled layer-by-layer fabrication to create construction components or structures directly from model data.
ADMET #: The absorption, distribution, metabolism, excretion, and toxicity filters that shape whether a molecule is likely to become a viable drug.
Advanced Driver Assistance Systems (ADAS)#: Vehicle technologies that help a human driver with warnings, braking, steering, and workload reduction without making every car fully autonomous.
Advanced Metering Infrastructure (AMI)#: The metering, communications, and data systems that make interval electricity data and smarter grid interaction possible.
Advanced Process Control (APC)#: A model-based control layer that adjusts process settings using measurements, predictions, and feedback.
Adversarial Attack #: A deliberate attempt to fool a model with input designed to make it fail.
Adversarial Example #: A specially crafted input that looks normal to people but causes a model to make a mistake.
Adversarial Machine Learning #: The field that studies how AI systems are attacked, manipulated, and defended.
Affective Computing #: AI systems that estimate, model, or respond to human affect and emotion from signals such as text, voice, facial expression, or behavior.
After-Call Work (ACW)#: The wrap-up work an agent completes after an interaction, such as notes, summaries, disposition codes, and CRM updates.
Age Assurance #: Estimating or verifying whether someone is above, below, or within an age range so a service can apply the right access and safety rules.
Agent Assist #: Real-time AI support that helps a human agent with knowledge, prompts, summaries, and next-best actions during a live interaction.
Agent-Based Modeling #: Simulating a system through many interacting agents whose behavior and feedback loops shape the larger outcome.
AI Agent #: A software system that can interpret goals, use tools, and take actions with some autonomy.
AI (Artificial Intelligence)#: The broad field of building systems that can perform tasks associated with perception, language, reasoning, and decision-making.
AI Assurance #: The testing, evidence, and review work used to show that an AI system is behaving as claimed and controlled well enough for its context.
AI Content Moderation #: Using AI to review, filter, rank, or escalate content that may violate rules or safety standards.
AI Data Labeling#: The process of tagging data so supervised models can learn from it.
AI Fairness #: The effort to make AI systems behave equitably across people, groups, and contexts.
AI Firewall #: A security layer that inspects AI inputs, actions, and outputs for threats, misuse, or policy violations.
AI-Generated Content (AIGC)#: Text, images, audio, video, or code created by AI systems.
Algorithm#: A set of instructions or rules used to solve a problem or perform a computation.
Algorithmic Bias #: Systematic skew in a system that leads to unfair or distorted outcomes.
Algorithmic Trading #: Using software rules and models to generate, route, or manage orders in financial markets.
Alignment (AI Alignment)#: The effort to make AI systems follow human goals, instructions, and safety expectations.
Ambient Computing #: Computing woven into devices and environments so assistance can appear in context instead of always requiring an explicit app session.
Anomaly Detection #: Finding unusual data points or events that differ sharply from normal patterns.
Architecture (AI Model Architecture)#: The overall design and structure of a model, including its layers and connections.
Archives #: Organized collections of records and materials preserved because they have long-term historical, legal, cultural, or operational value.
Artificial General Intelligence (AGI)#: A hypothetical AI with broad human-like ability across many tasks rather than strength in one narrow domain.
Attention Mechanism#: A way for a model to focus on the most relevant parts of its input when producing an output.
Attribution #: Assigning authorship, origin, source, or responsibility to a work, record, object, or other output.
Audience Segmentation #: Grouping people into useful audience buckets or modeled cohorts for targeting, personalization, or measurement.
Audio Restoration #: Using AI to denoise, declip, inpaint, and rebuild damaged recordings so they become more usable again.
Augmentative and Alternative Communication (AAC)#: Communication methods and tools that support or replace speech when a person cannot rely on spoken language alone.
Authentication #: Confirming that a person, document, object, or piece of content is genuine and really what it claims to be.
Autoencoder#: A neural network trained to compress data into a compact representation and reconstruct it again.
Automated Machine Learning (AutoML)#: Using software to automate parts of model training, tuning, and evaluation.
Automated Valuation Model (AVM)#: A computerized property valuation system that estimates real-estate value from comparable sales, property attributes, and market signals.
Automatic Defect Classification (ADC)#: Using AI to sort detected defect candidates into actionable classes, nuisance, or unknown categories so review teams can respond faster.
Automatic Speech Recognition (ASR)#: Technology that converts spoken language into text.

B

Back to finder

Backpropagation#: The training process that moves error information backward through a neural network so its weights can be updated.
BACnet #: A widely used communications standard that helps building automation devices exchange data and commands across vendors.
Baggage Reconciliation #: Confirming that a checked bag stays correctly matched to the right passenger, flight, custody state, and handling decision as airport operations evolve.
Battery Management System (BMS)#: The control system that monitors an EV battery pack and manages charging, power limits, safety, and battery-health estimation.
Beamforming #: Using an array of microphones or other sensors to focus on sound from one direction or location while reducing interference from others.
Behavioral Biometrics #: Authentication and fraud-detection methods that identify people by how they type, swipe, move, or otherwise behave.
Benchmarks#: Standard tests used to compare models on common tasks or datasets.
BERT (Bidirectional Encoder Representations from Transformers)#: An influential transformer-based language model designed for understanding text with bidirectional context.
Berth Allocation #: Assigning vessels to berth positions and time windows so crane plans, yard readiness, and port-side workflows stay coordinated.
Beyond Visual Line of Sight (BVLOS)#: Flying a drone beyond the range where the remote pilot can directly see it, which raises the need for stronger sensing, procedures, and airspace integration.
Bias #: Systematic skew or error in data, modeling, or decisions that can distort results or create unfair outcomes.
Bias Mitigation #: Methods for identifying, reducing, and monitoring unfair bias in AI systems.
Bias-Variance Tradeoff#: The balance between a model that is too simple to capture patterns and one that is too sensitive to the training data.
Binary Classification#: A task in which each example must be assigned to one of two classes.
Black Box Model#: A model whose internal reasoning is hard for humans to inspect or explain directly.
Borescope Inspection #: Inspecting the inside of an engine or other hard-to-reach component with a camera probe so condition can be assessed without full disassembly.
Brand Lift #: Measuring whether advertising changed awareness, recall, favorability, consideration, or related brand outcomes compared with a control group.
Brand Safety #: Keeping ads and generated creative away from harmful, unsuitable, misleading, or policy-violating content and contexts.
Building Information Modeling (BIM)#: A structured digital representation of a building and its components that supports coordination, simulation, estimating, construction, and operations.

C

Back to finder

Calibration #: The degree to which a model's confidence matches what really happens.
Call Deflection #: Resolving routine issues through self-service or alternate channels so avoidable live-agent calls never have to enter the voice queue.
Candidate Generation #: The retrieval stage that narrows a huge pool of possible items into a smaller set worth ranking in detail.
Cataloging #: The structured process of describing an item so it can be identified, found, and managed later.
Causal Inference #: Estimating what changed because of an intervention rather than what merely moved alongside it.
Chain of Thought (CoT)#: A prompting style that encourages a model to work through intermediate reasoning steps.
Change Detection #: Comparing observations across time to determine what changed, where it changed, and how much it changed.
Chatbot#: A system that interacts with users in natural language through text or voice.
Classification#: The task of assigning an input to one of several categories.
Clinical Decision Support #: Software that uses patient data, rules, models, or retrieved evidence to help clinicians make safer and better-informed decisions.
Clustering #: Grouping similar items together without using predefined labels.
Cognitive Accessibility #: Designing content and interfaces so people can understand, remember, navigate, and complete tasks with less mental effort.
Cognitive Radar #: A radar approach that uses feedback and adaptive sensing so waveform, beam, or scheduling choices can change based on what the system just observed.
Cold Start #: The recommendation problem that appears when a new user, item, or context has too little history for confident prediction.
Collaborative Robot (Cobot)#: A robot designed to work more safely near people and to make more mixed, flexible automation tasks practical.
Collections Management #: The ongoing work of organizing, tracking, preserving, and governing collections over time.
COLREGs #: The international maritime rules of the road that govern how vessels determine right of way, avoid collisions, and signal intent at sea.
Combined Heat and Power (CHP)#: Generating electricity and useful heat from the same fuel source so less energy is wasted overall.
Computational Aesthetics #: Using AI to estimate formal visual qualities such as composition, style, balance, and perceived appeal in images or artworks.
Computational Fluid Dynamics (CFD)#: Using numerical simulation to estimate how fluids move, interact with surfaces, and create forces, heat, pressure, or mixing effects.
Computer Vision #: The branch of AI that helps systems interpret images and video.
Confidence #: The system's stated or implied degree of certainty about an output, prediction, or match.
Confusion Matrix#: A table that shows how often a classifier makes each kind of right and wrong prediction.
Conservation #: The active care and protection of objects, records, and heritage materials to slow deterioration and maintain their integrity.
Contextual Targeting #: Showing ads or recommendations based on the surrounding content or moment rather than mainly on long-term user identity.
Context Window #: The maximum amount of information a model can process in one interaction, measured in tokens. The space available for input and output depends on the model and system.
Continuous Authentication #: Reassessing identity during a session instead of trusting one successful login forever.
Continuous Controls Monitoring (CCM)#: Checking live controls and current evidence on an ongoing basis so drift, exceptions, and broken compliance workflows are found before the next audit cycle.
Contract Lifecycle Management (CLM)#: Managing the full contract workflow from request and drafting through negotiation, approval, execution, obligations, renewals, and amendments.
Conversational Commerce #: Shopping flows that use natural language, recommendations, and live product data to guide product discovery and buying.
Conversation Intelligence #: Using AI to turn calls, meetings, and other conversations into searchable structure, topics, sentiment, and workflow signals.
Convolutional Neural Network (CNN)#: A neural network architecture especially useful for images and other grid-like data.
Cover Crops #: Crops grown mainly to protect and improve the soil system between or alongside cash crops rather than only for direct harvest.
Creative Fatigue #: The decline in ad performance that happens when audiences see the same or too-similar creative too often.
Cross-Lingual Information Retrieval (CLIR)#: Finding relevant documents in one language when the search query is written in another.
Cross-Validation#: A method of testing a model by training and validating it across multiple different splits of the data.
Cycle Counting #: Regularly verifying selected inventory locations or SKUs so stock records stay accurate without shutting down for a full physical inventory.

D

Back to finder

Data Augmentation#: Expanding a dataset by creating modified versions of existing examples.
Data Clean Room #: A controlled environment where multiple parties can analyze combined data with privacy rules and aggregation safeguards.
Data Drift #: A change in the input data over time that can hurt model performance.
Data Governance #: The policies and controls that determine how data is collected, managed, protected, and used.
Data Labeling#: Adding tags or annotations to data so a model can learn the target output.
Data Preprocessing#: Cleaning and transforming raw data before training or inference.
Decoder#: The part of a model that generates output from an internal representation or from encoded input.
Deepfake #: AI-generated or AI-altered media designed to convincingly imitate a real person's appearance or voice.
Deep Learning #: A branch of machine learning based on multi-layer neural networks.
De-Identification #: Removing or transforming identifying details so data can be analyzed or shared with lower privacy risk.
Demand Response #: Reducing or shifting electricity use in response to grid conditions, prices, or utility signals.
Device Fingerprinting #: Using device, browser, network, and environment clues to estimate whether an access event fits a known pattern.
Differential Privacy #: A formal way to reduce how much any one person's data can be inferred from a result or model.
Diffusion Models #: Generative models that start from noise and gradually turn it into structured output.
Digital Accessibility #: Designing software and digital content so people with different abilities, devices, and assistive needs can use it effectively.
Digital Identity #: The reusable credentials, proofs, and trust layer that lets a person or organization prove who they are across online services.
Digital Product Passport (DPP)#: A machine-readable lifecycle record tied to a specific product so provenance, materials, repairs, resale, and compliance data can travel with it.
Digital Thread #: A connected flow of lifecycle data that links design, production, operation, and service information.
Digital Twin #: A live digital representation of a physical asset or process that stays connected to operational data.
Digitization #: Converting physical or analog material into digital form so it can be stored, searched, and reused more effectively.
Direct Indexing #: Owning the underlying securities of an index so a portfolio can be customized, rebalanced, and tax-managed more precisely.
Dissolved Oxygen #: The amount of oxygen mixed into water and available to aquatic life, which strongly shapes fish stress, respiration, and closed-system stability.
Document AI #: AI systems that read, classify, extract, and route information from documents; often called intelligent document processing (IDP).
Driver Monitoring System #: A vehicle system that checks whether the human driver is attentive, alert, and ready to supervise or take over.
Dynamic Creative Optimization (DCO)#: Automatically combining, testing, and promoting ad assets so different viewers and placements get more effective creative variants.
Dynamic Pricing #: Updating prices as demand, inventory, competition, or context changes instead of relying only on static price lists.

E

Back to finder

Early Intervention System (EIS)#: A supervisory workflow that flags when an officer may need coaching, support, training, or closer review before risk escalates.
Earthquake Early Warning (EEW)#: Detecting an earthquake just after it begins and issuing alerts before the strongest shaking arrives.
Edge Computing #: Processing data and running software close to where it is generated instead of sending every decision to a faraway cloud.
E-Discovery #: Using AI and review workflows to collect, prioritize, analyze, and validate large populations of electronic documents for litigation or investigations.
Electronic Health Record (EHR)#: A digital longitudinal record of patient care data such as notes, medications, orders, labs, and encounter history.
Embedding #: A numeric vector representation that captures semantic similarity between items such as words, images, or documents.
Encoder#: The part of a model that transforms input into a useful internal representation.
Entity Extraction and Linking #: Identifying important entities in text and connecting each mention to the correct real-world or database entry.
Entity Resolution #: Deciding when different records really refer to the same person, company, account, or other real-world entity.
Epoch#: One full pass through the training dataset during model training.
Ethical AI #: Building and using AI in ways that respect fairness, accountability, privacy, and human values.
Evidence #: The records, measurements, retrieved sources, or other signals that support a claim, decision, or model output.
Explainability #: The broader practice of making a system's outputs, reasoning, evidence, and limits understandable to people.
Explainable AI (XAI)#: The field of making model behavior and outputs easier for people to understand.

F

Back to finder

F1 Score #: A metric that combines precision and recall into one balanced measure.
Face Identification #: Searching one face against many enrolled identities to find likely candidates or a match above threshold.
Face Verification #: Checking whether one presented face matches one claimed or enrolled identity.
Facings #: The number of visible units of a product shown to the shopper on the shelf.
Factor Investing #: Building or analyzing portfolios through systematic exposures such as value, momentum, quality, size, or low volatility.
Fault Detection and Diagnostics (FDD)#: Using rules, models, and live operational data to find faults in equipment or controls and explain what is likely going wrong.
Feature Engineering#: Creating or refining input variables so a model can learn more effectively.
Federated Learning #: Training a shared model across many devices or organizations without pooling all raw data in one place.
Feed Ranking #: The process of ordering eligible posts, videos, or updates so a personalized feed shows the most relevant items first.
FHIR #: Fast Healthcare Interoperability Resources, a standard for structuring and exchanging healthcare data through modern APIs.
Fine-Tuning #: Adapting a pretrained model to a narrower task or domain with additional training.
Flue Gas Cleaning #: The systems that remove particulates, acid gases, metals, and other pollutants from combustion exhaust before release.
Forgery #: A deceptive imitation or alteration meant to pass as a genuine object, document, identity, or piece of media.
Fraud Detection #: Using analytics and AI to identify suspicious behavior, transactions, or impersonation.
Function Calling #: A way for a model to produce structured arguments for a tool or function instead of free text alone.

G

Back to finder

Gaze Tracking #: Using eye position and eye movement to estimate where a person is looking so an interface can respond.
Generative Adversarial Network (GAN)#: A generative architecture in which one network creates outputs and another tries to detect whether they are fake.
Generative Artificial Intelligence (GenAI)#: AI systems that create new content such as text, images, audio, video, or code.
Geofencing #: Using virtual boundaries around places, assets, or moving work zones so software can trigger alerts, rules, or actions when something enters, exits, or gets too close.
Geographic Information System (GIS)#: Software and data systems for storing, visualizing, analyzing, and managing spatial information.
Geoparsing #: Extracting place names from text and linking them to the correct real-world locations.
Gesture Recognition #: Using AI and sensors to interpret hand, body, or motion signals as interface commands.
GPT (Generative Pre-trained Transformer)#: A family of transformer-based language models trained to predict the next token in text.
Graph Neural Network (GNN)#: A neural network built to learn from graph-structured data such as molecules, transaction networks, and knowledge graphs.
GraphRAG #: A retrieval-augmented generation pattern that uses graph structure to retrieve, connect, and organize evidence more coherently for grounded answers.
Grounding #: Connecting a model's output to trusted sources, retrieved evidence, tools, or real-world state.
Ground Truth #: The verified real-world state or trusted label set used as the reference point for training, evaluation, or operational measurement.
Guardrails #: Filters, rules, and runtime checks that keep an AI system within desired safety or workflow boundaries.

H

Back to finder

Hallucination #: When an AI system produces plausible-sounding information that is false, unsupported, or ungrounded.
Handwriting Recognition #: Using AI to read handwritten notes, forms, and manuscripts and convert them into machine-readable text.
Hazard Analysis and Critical Control Point (HACCP)#: A preventive food-safety system that identifies hazards, defines critical controls, and verifies that food stays safe as it moves through production and handling.
Human in the Loop #: A workflow in which people remain involved in review, correction, approval, or decision-making instead of leaving everything to the model.
Hydraulic Model Calibration #: Tuning a water network model so simulated flows, pressures, and states stay aligned with field measurements.
Hydroponics #: Growing plants in nutrient solution, often with inert media, so water, chemistry, and root-zone conditions can be controlled more precisely.
Hyperparameter#: A setting chosen by the developer, such as learning rate or batch size, rather than learned by the model itself.
Hyperparameter Tuning#: The process of searching for better hyperparameter settings to improve model performance.
Hyperspectral Imaging #: Capturing many narrow wavelength bands so materials can be identified by their spectral signatures rather than by ordinary color alone.

I

Back to finder

Identity Proofing #: Establishing that a person is who they claim to be during enrollment before later authentication begins.
Image Classification #: The task of assigning an image to one or more categories based on what it contains.
Image Generation#: Creating new images with AI, often from text prompts or other visual input.
Incrementality #: Measuring what effect a campaign or intervention caused beyond what would likely have happened anyway.
Inference#: Using a trained model to produce output on new input data.
Infrasound #: Low-frequency sound below normal human hearing that can reveal eruptions, explosions, storms, and other energetic events over long distances.
Inpainting #: Filling, repairing, replacing, or extending missing and masked parts of an image or other media so they match the surrounding context.
InSAR #: Using repeated radar images to measure subtle ground deformation across earthquakes, volcanoes, subsidence, and other Earth-surface change.
In-Situ Monitoring #: Watching a process while it is happening so deviations can be detected, interpreted, and sometimes corrected before the job is finished.
Instruction Tuning #: Post-training that teaches a model to follow user requests more reliably.
Intelligent Tutoring System (ITS)#: Educational software that models a learner's progress and provides hints, explanations, and practice that adapt to the student over time.
Intent Recognition #: Identifying what a user is trying to accomplish so a conversational system can route the interaction toward the right next step.
Interoperability #: The ability of different systems to exchange data and use it consistently without losing meaning.
Interpretability #: The degree to which humans can understand how and why a model behaves the way it does.
Inventory Visibility #: Knowing what inventory exists, where it is, and how available it really is across stores, warehouses, and channels.
Inverse Design #: Working backward from a target behavior to propose structures, parameters, or recipes that can produce it.
Itinerary Optimization #: Turning destinations, timing, routing, budget, and traveler constraints into a feasible trip plan that can adapt when conditions change.

J

Back to finder

Jailbreaking #: Manipulating a language model so it bypasses intended restrictions or safety rules.

K

Back to finder

Knowledge Distillation #: Training a smaller model to imitate a larger one so it becomes cheaper to deploy.
Knowledge Graph #: A structured network of entities and relationships that makes connected facts easier to query and reason over.
Knowledge Tracing #: Estimating how a learner's mastery changes across sequences of answers, hints, retries, and practice events over time.

L

Back to finder

Language Model (LM)#: A model that learns patterns in language and predicts likely token sequences.
Large Language Model (LLM)#: A neural network trained on large amounts of text to generate and work with language.
Latent Space#: A compressed internal representation in which models capture the underlying structure of data.
Layout Analysis #: Identifying the structural parts of a page so AI can preserve reading order, tables, fields, and other document context.
Learning Rate#: A training setting that controls how large each update step is during optimization.
Link Prediction #: Estimating which relationships are likely missing from a graph based on the structure already present.
Live-Agent Handoff #: Transferring a conversation from a bot to a human agent with enough context that the customer does not have to start over.
Liveness Detection #: Anti-spoofing checks that try to confirm a real person is present instead of a photo, replay, mask, or deepfake.
Logical Qubit #: A qubit encoded across multiple physical qubits so errors can be detected and corrected before the stored quantum information is lost.
LoRA (Low-Rank Adaptation)#: A parameter-efficient way to adapt a large model without retraining all of its weights.
Loss Function#: A formula that measures how wrong a model's predictions are during training.
Loudness Normalization #: Measuring and adjusting audio so episodes, tracks, ads, or clips play back at a more consistent perceived volume.

M

Back to finder

Machine Translation #: Using AI to translate meaning from one language into another.
Marine Energy #: Generating usable power from tides, waves, currents, thermal gradients, or salinity differences in oceans and waterways.
Marketing Mix Modeling (MMM)#: Using aggregated data and statistical models to estimate how channels and external factors affect marketing outcomes over time.
Market Microstructure #: How order books, spreads, queues, and venue rules shape the way real trades are executed.
Matchmaking #: Pairing people, teams, or other entities using mutual fit, constraints, and context rather than one-sided recommendation alone.
Material Recovery Facility (MRF)#: A recycling facility that receives mixed material, separates it into usable commodities, and prepares those streams for sale or further processing.
Materials Informatics #: Using data, models, and experiments together to discover and optimize materials faster.
Matter #: A smart-home interoperability standard that helps accessories from different brands work together more cleanly inside the same home.
Medication Verification #: Checking that the medication selected, labeled, or dispensed truly matches the intended drug, dose, form, patient, and order context before it moves forward.
Metadata Enrichment #: Adding useful tags, descriptions, relationships, and context to content so it is easier to search, organize, connect, and reuse.
Metagenomics #: Sequencing mixed microbial communities directly and using computation to infer which organisms, genes, and functions may be present.
Microgrid #: A local energy system that can coordinate generation, storage, and loads together and sometimes operate independently from the wider grid.
ML (Machine Learning)#: The branch of AI in which systems learn patterns from data instead of relying only on fixed hand-written rules.
Model Card #: A document that explains a model's purpose, evaluation, limitations, and intended use.
Model Compression #: Techniques that make models smaller, faster, or cheaper to run.
Model Drift #: A decline or change in model behavior over time as conditions shift.
Model Evaluation #: Testing a model to understand how well it performs and where it fails.
Model Explainability #: The ability to communicate why a model produced a particular result.
Model Fairness #: Whether a specific model behaves equitably across different groups or contexts.
Model Monitoring #: Tracking a deployed model so drift, degradation, and unusual behavior are caught after launch.
Model Parameters#: The values a model learns during training, such as neural network weights.
Model Predictive Control (MPC)#: A control approach that uses a model of the system to predict future behavior and choose actions that work best under constraints.
Multilevel Regression and Poststratification (MRP)#: A survey-estimation method that combines multilevel modeling with population weighting to produce more reliable local or subgroup estimates from thin samples.
Multimodal Large Language Models (MLLMs)#: Models that combine language with images, audio, video, or other input types.
Multimodal Learning #: Learning from or generating across more than one kind of data, such as text plus images.
Multiscale Modeling #: Connecting material, process, part, and system behavior so decisions can reflect more than one physical scale at once.
Myoelectric Control #: Using electrical activity from muscles to control a prosthetic or assistive device in a way that reflects user intent.

N

Back to finder

Named Entity Recognition (NER)#: Identifying references to people, organizations, places, dates, and other key entities in text.
Nanofabrication #: Building micro- and nano-scale structures through patterning, deposition, etch, imprint, assembly, and measurement-aware process control.
Natural Language Processing (NLP)#: The branch of AI focused on understanding and generating human language.
Neural Networks #: Computing systems inspired by layered neuron-like structures that learn patterns from data.
Nondestructive Testing (NDT)#: Inspecting a weld, structure, or component for flaws without cutting it apart or destroying it.
Non-Manual Signals #: The facial expressions, head movements, mouth shapes, and upper-body cues that carry grammar and meaning in signed languages.
Nowcasting #: Estimating the current state of the economy before the official statistics are fully available.

O

Back to finder

Object Detection #: Identifying both what objects appear in an image and where they are located.
Onboard Autonomy #: Letting a spacecraft, rover, or other remote system make limited decisions locally when waiting for human commands is too slow.
On-Device AI #: Running AI features locally on a phone, laptop, vehicle, or other device instead of sending every request to a remote server.
Ontology #: A formal description of the important concepts, categories, and relationships in a domain.
Open Banking #: User-permissioned access to bank and financial account data so apps and services can help with budgeting, payments, and account aggregation.
Operational Design Domain (ODD)#: The specific conditions under which an autonomous or assisted system is designed and validated to operate.
Optical Character Recognition (OCR)#: Converting text in scans, images, or PDFs into machine-readable text.
Optical Sorting #: Using cameras and other sensors to identify materials on a moving line and separate them automatically.
Orthomosaic #: A stitched and georeferenced aerial image corrected so it can be measured and compared like a map.
Overfitting #: When a model learns the training data too specifically and performs poorly on new examples.

P

Back to finder

Parametric Design #: Defining geometry and behavior through editable parameters, rules, and relationships so many design variations can be explored systematically.
Passkey #: A phishing-resistant sign-in credential that often uses device biometrics or a PIN for local user verification.
Path Planning #: Choosing a route or movement sequence that helps a robot reach its goal safely and efficiently.
Personally Identifiable Information (PII)#: Information that can identify a specific person and therefore requires careful handling and protection.
Pharmacogenomics #: Studying how genetic variation changes drug response so treatment choice and dose can be tailored more safely and effectively.
Phenotyping #: Identifying meaningful patient traits, disease patterns, or clinical subgroups from health data.
Photobioreactor #: A controlled vessel or tubing system for growing light-dependent organisms such as microalgae under managed illumination, gas exchange, mixing, and water chemistry.
Physical AI #: AI systems that act in the physical world through robots, vehicles, sensors, and control systems rather than only through software interfaces.
Pillar Two #: The global minimum-tax framework that pushes large multinationals toward jurisdiction-level effective-tax-rate analysis and top-up tax calculations.
Planogram #: A structured retail shelf layout that specifies where products should go, how much space they get, and how the shelf should be merchandised.
Planogram Compliance #: Checking whether products, facings, prices, and placement match the intended retail merchandising plan.
Plant Phenotyping #: Measuring crop traits such as canopy structure, vigor, fruit load, and stress from imagery, spectra, geometry, and field sensors.
Player Modeling #: Estimating how a player behaves, learns, prefers, or struggles so content, pacing, and challenge can adapt more intelligently.
Post-Quantum Cryptography #: Cryptographic methods designed to remain secure even if large-scale quantum computers become practical.
Precision #: The share of positive predictions that are actually correct.
Precision Aquaculture #: Using sensors, imaging, telemetry, and predictive models to manage fish or shellfish production with earlier warnings and more targeted health interventions.
Precision Beekeeping #: Using sensors, imaging, acoustics, and models to manage bee colonies with earlier warnings and more targeted interventions.
Predictive Analytics #: Using data and models to estimate likely future outcomes, risks, or trends.
Predictive Maintenance #: Using data and models to estimate when equipment is likely to degrade or fail so maintenance can happen before an outage.
Presence-Based Automation #: Automations that respond to whether people are home, away, arriving, or moving through a space instead of relying only on fixed schedules.
Preservation #: The long-term work of keeping information, media, artifacts, or records safe, usable, and accessible over time.
Pre-trained Model#: A model that has already been trained on a large dataset before being adapted to a narrower task.
Pretraining#: The large-scale initial training stage that teaches a model broad patterns before task-specific adaptation.
Price Elasticity #: How strongly demand changes when price changes, often varying by product, shopper, channel, and context rather than staying fixed.
Privacy-Enhancing Technologies (PETs)#: Technical methods that reduce exposure of personal data while still allowing some useful processing, sharing, or analysis.
Process Mining #: Using event logs to reconstruct how a business process actually runs so teams can find bottlenecks, rework, and automation opportunities.
Product Lifecycle Management (PLM)#: Managing product data, changes, releases, and traceability from design through manufacturing, service, and retirement.
Product Tagging #: Using AI to assign structured categories, attributes, and descriptors to products at catalog scale.
Prompt #: The text, instructions, or input given to a model to guide its output.
Prompt Engineering #: Designing prompts so a model produces more useful, accurate, or structured responses.
Prompt Injection #: A security attack in which malicious instructions hidden in content try to override a model's intended behavior.
Pronunciation Assessment #: Using structured scoring to judge how closely a spoken sound, word, or phrase matches a target production.
Prosody #: The rhythm, pitch, stress, pacing, and intonation patterns in speech that carry meaning beyond the transcript alone.
Provenance #: The documented origin and ownership or custody history of an object, record, or other item of value.

Q

Back to finder

Quantization#: Reducing numeric precision in a model to save memory and speed up inference.

R

Back to finder

Random Forest#: A machine learning method that combines many decision trees to improve prediction stability.
Real-World Evidence (RWE)#: Clinical evidence developed from routine-care data such as EHRs, claims, registries, and other non-trial sources.
Recall #: The share of real positive cases that the model successfully catches.
Recommender System #: An AI system that ranks and suggests items a particular user is likely to value.
Redlining #: Marking proposed edits in a contract so negotiating parties can see exactly what changed and why.
Red Teaming #: Structured adversarial testing used to uncover safety, security, bias, and reliability failures.
Reduced-Order Modeling #: Compressing a complex physical system into a smaller model that still preserves the dominant behavior needed for fast analysis or control.
Regression#: A task in which a model predicts a numeric value rather than a category.
Regularization#: Training techniques that reduce overfitting and help a model generalize better.
Regulatory Impact Assessment (RIA)#: A structured review of how a proposed law or regulation may affect people, markets, government operations, and compliance before adoption.
Reinforcement Learning from Human Feedback (RLHF)#: A method that uses human preferences to shape model behavior after pretraining.
Reinforcement Learning (RL)#: A learning approach in which an agent improves through rewards and penalties from its environment.
Remote ID #: The broadcast identity and location layer that helps drones become more visible to airspace managers, nearby operators, and authorized responders.
Remote Sensing #: Collecting imagery or other measurements from a distance so AI systems can analyze the Earth, oceans, atmosphere, or planetary surfaces.
Replenishment #: Restocking inventory so products are available where and when they are needed without carrying too much excess stock.
Reranking #: Rescoring an initial result set with richer signals so the most useful retrieved items rise to the top.
Responsible AI #: Building AI systems that are useful, safe, fair, accountable, and governable.
Responsive Search Ads (RSA)#: Search ads that mix and match multiple headlines and descriptions to find stronger combinations for different queries and contexts.
Restoration #: Repairing or reconstructing damaged, degraded, faded, or incomplete material so it becomes more legible, usable, or understandable.
Retail Media #: Advertising inside retailer-owned shopping environments and data ecosystems using commerce signals close to purchase.
Retrieval Augmented Generation (RAG)#: A pattern that combines a model with retrieved external information so answers stay fresher and better grounded.
Retro-Commissioning #: A systematic process of testing and tuning an existing building so its systems perform closer to intended design and operational goals.
Retrosynthesis #: Planning how to make a target molecule by reasoning backward from the product to simpler precursors.
Revenue Management #: Matching price, availability, restrictions, and channel mix to demand so a business can earn more from limited capacity.
RFID (Radio-Frequency Identification)#: Using radio tags and readers to identify physical items automatically as they move through stores, stockrooms, and fitting rooms.
Risk-Based Authentication #: Adjusting authentication requirements based on how risky the current sign-in or action appears to be.
Risk-Based Monitoring (RBM)#: Targeting oversight toward the trial sites, subjects, data streams, and processes most likely to affect participant safety or study reliability.
Robo-Adviser #: An automated digital investment service that uses questionnaires, algorithms, and portfolio rules to guide or manage investing.
Robustness #: The ability of a system to keep working under noise, shift, unusual input, or active attack.

S

Back to finder

Sanctions Screening #: Checking customers, counterparties, and transactions against sanctions and watchlists with enough precision to be useful and enough evidence to be auditable.
Self-Supervised Learning #: Learning from data that provides its own training signal instead of relying only on manual labels.
Semantic Search #: Search that finds results by meaning and intent, not only exact keyword matches.
Send-Time Optimization #: Using data and models to decide when a message is most likely to be opened, read, or acted on.
Sensor Fusion #: Combining signals from multiple sensors into one more reliable estimate of what is happening.
Sentiment Analysis #: Using AI to identify whether language expresses positive, negative, neutral, or mixed attitude.
Shared Autonomy #: Dividing control between a person and an intelligent system so automation can assist while the human remains responsible for supervision and takeover.
Shelf Intelligence #: Using AI and computer vision to understand stock levels, placement, pricing, and other real shelf conditions in stores.
Skill-Based Matchmaking (SBMM)#: Matching players or teams using estimated skill so multiplayer games stay competitive, less lopsided, and healthier over time.
SLAM (Simultaneous Localization and Mapping)#: A robotics method for figuring out where the robot is while building a map of the space around it.
Slippage #: The gap between the price a trader expected and the price the market actually delivered.
Slotting Optimization #: Assigning products to warehouse locations so travel, replenishment effort, and space use stay efficient together.
Smart Charging #: Using software to decide when, where, and how fast an electric vehicle should charge.
Smart Grid #: An electricity system that uses sensing, communication, software, and automation to manage supply, demand, and reliability more intelligently.
SOAR #: Security orchestration, automation, and response systems that connect tools and run playbooks faster.
Social Listening #: Using AI to monitor, organize, and interpret social posts, comments, reviews, and other public audience signals at scale.
Source Separation #: Splitting a mixed recording into more isolated components such as vocals, drums, speech, or background sound.
Spaced Repetition #: Scheduling review so information returns at intervals that improve long-term memory instead of fading away between study sessions.
Space Planning #: Organizing rooms, furnishings, storage, and circulation so the space works for real use instead of only looking good in a concept image.
Spatial Computing #: Computing that understands physical space, surfaces, and position so digital content can behave as if it belongs in the real world.
Speaker Diarization #: Figuring out who spoke when in a recording so transcripts preserve conversational structure instead of collapsing everyone together.
Speech Biofeedback #: Using visual, acoustic, or sensor-based feedback to help a speaker see or hear aspects of speech production that are otherwise hard to monitor directly.
Stable Diffusion #: A widely known family of diffusion-based image generation models.
Stress Testing #: Testing how a system, market, or portfolio behaves under difficult but plausible conditions.
Structural Break #: A change in the underlying relationships that a model relied on before.
Structural Health Monitoring #: Using inspection data, sensors, and models to track whether a structure is staying sound or drifting toward damage.
Supervised Learning #: Learning from labeled examples where the correct target is already known.
Support Vector Machine (SVM)#: A classical machine learning method often used for classification and sometimes regression.
Surface Code #: A topological quantum error-correcting code that protects logical qubits by measuring local parity checks on a two-dimensional lattice of physical qubits.
Surrogate Model #: A simplified fast-running model that approximates a more complex simulation or physical process.
Swarm Intelligence #: Coordinating many agents through local rules, shared signals, and bounded autonomy so the group can do more collectively than any one unit could alone.
Syndromic Surveillance #: Monitoring symptom and encounter patterns in near real time so public-health teams can detect unusual change before confirmed diagnoses fully accumulate.
Synthetic Data #: Artificially generated data used to train, test, or evaluate models.
Synthetic Identity Fraud #: Creating a fake but plausible identity by mixing real personal data with fabricated details so weak onboarding and fraud controls treat it as legitimate.
System Prompt #: A high-priority instruction that sets a model's role, behavior, or constraints for an interaction.

T

Back to finder

Tax-Loss Harvesting #: Selling positions at a loss to offset taxable gains while keeping the portfolio aligned to its long-term strategy.
TEFCA #: The Trusted Exchange Framework and Common Agreement, a U.S. framework for nationwide health information exchange.
Telematics #: Using connected-vehicle data such as location, diagnostics, battery state, and driver signals to manage fleets and mobility operations more intelligently.
Telemetry #: Operational signals such as events, metrics, logs, traces, and state changes that show what a device or system is doing over time.
Teleoperation #: Remote control or supervision of a robot by a human operator, often blended with autonomy.
Test Set#: A held-out dataset used to measure how well a model generalizes after training.
Text Summarization #: Condensing a longer document or conversation into a shorter version that preserves the main ideas.
Time Series Forecasting #: Predicting future values from time-ordered data such as demand, occupancy, emissions, or sensor readings.
Token #: A unit of text a language model processes, such as a word piece, symbol, or short sequence of characters.
Tokenization #: Breaking text into the tokens a model can process.
Tool Use #: Letting a model call external tools, APIs, or services as part of solving a task.
Toxicity #: Harmful, abusive, or hostile content that AI systems may generate, amplify, or help detect.
Training Set #: The portion of data used to fit a model's parameters during learning.
Trajectory Prediction #: Estimating where an aircraft, vehicle, drone, or other moving system is likely to go next based on current state, intent, and environment.
Transaction Cost Analysis (TCA)#: Measuring how much value is lost or preserved as orders move from investment intent to real market execution.
Transaction Monitoring #: Reviewing payments and account activity for patterns that may indicate suspicious, fraudulent, or non-compliant behavior.
Transfer Learning #: Reusing knowledge from one task or dataset to improve performance on another.
Transfer Pricing #: Pricing related-party transactions so profits, documentation, and tax outcomes stay supportable across jurisdictions.
Transformer #: A neural network architecture built around attention that powers many modern language and multimodal models.
Translation Memory #: Reusing previously approved translated segments so multilingual content can stay more consistent across projects, releases, and recurring text.
Trend Forecasting #: Using AI to detect weak signals and estimate which styles, products, topics, or behaviors are likely to rise next.
Trust and Safety #: The people, policies, systems, and workflows used to keep digital services safe, lawful, and usable in the face of abuse and other harms.

U

Back to finder

Uncertainty #: The degree to which a model, dataset, or decision remains ambiguous, incomplete, or not fully known.
Underfitting#: When a model is too simple to capture meaningful structure in the data.
Underwriting #: Evaluating a risk to decide whether to offer coverage, on what terms, and at what price, with evidence, rules, and human review where needed.
Unsupervised Learning #: Learning patterns from unlabeled data without predefined target labels.

V

Back to finder

Validation Set#: A dataset used during model development to compare choices and tune settings before final testing.
Variational Quantum Algorithm (VQA)#: A hybrid quantum-classical method that repeatedly tunes quantum circuit parameters using classical optimization around measured quantum results.
Vector Database #: A database optimized for storing and searching embeddings.
Vector Search #: Finding items by semantic similarity in embedding space rather than exact keyword matching.
Vehicle-to-Grid (V2G)#: Using bidirectional EV charging so vehicles can supply power back to a building or the electric grid when it makes sense.
Verification #: Checking whether a claim, identity, document, output, or piece of media is correct, genuine, or supported by evidence.
Virtual Commissioning #: Testing automation and control logic in simulation before real equipment goes live.
Virtual Metrology #: Estimating measurement results from process and equipment data instead of physically measuring every unit.
Virtual Power Plant (VPP)#: A coordinated network of distributed energy resources that can be managed as a flexible power system even though the assets are spread across many locations.
Virtual Try-On #: Using AI and visual overlays to preview how makeup, accessories, or apparel may look before buying.
Visual Search #: Searching with an image or camera input so the system finds visually similar objects, scenes, or products.
Voice Biometrics #: Using voice characteristics as an identity signal for personalization, verification, or low-friction access control.

W

Back to finder

Wake-Word Detection #: The lightweight speech model that listens for an activation phrase so a voice device knows when to start paying closer attention.
Waste-to-Energy #: Converting residual waste into usable energy through controlled thermal, biological, or other recovery processes.
Weak Supervision #: Using rules, heuristics, prompts, or other noisy signals to create useful draft labels faster than full manual annotation.
Workflow Orchestration #: Coordinating the sequence of models, rules, tools, approvals, and human review steps around an AI-driven process.

Z

Back to finder

Zero-Knowledge Proof (ZKP)#: Proving that a statement is true without revealing the private data or secret behind it.
Zero-Shot Learning#: Handling a task or class without labeled training examples for that specific task or class. In language-model prompting, zero-shot means giving an instruction without worked examples in the prompt.
Zero Trust #: A security model that assumes no user, device, or network location should be trusted by default.

Keep exploring

For a guided introduction, read Understanding AI or the introduction to large language models. Explore Yenra’s AI library for longer guides.

For a more technical reference, Google’s Machine Learning Glossary expands on model, training, and evaluation terminology, including activation functions, context windows, and zero-shot prompting. Industry terms here provide context for AI applications; follow their linked explanations for domain-specific detail.