Key AI terms

AI Glossary

Plain-language definitions of the terms behind predictive, generative, and agentic AI. New to AI? Start with the key terms.

237 terms

A

35 terms

Activation FunctionDeep Learning: A mathematical function applied to each neuron's output in a neural network that decides how strongly a signal passes forward. Nonlinear activations such as ReLU, sigmoid, and tanh let networks learn complex patterns rather than only straight-line relationships.
Agent BuilderKey termAgentic AI: A tool or environment for creating AI agents by defining their instructions, connected tools, data sources, and workflows. Agent builders often offer visual or low-code interfaces so teams can configure, test, and deploy agents without writing everything from scratch.
Agent EvaluationKey termAgentic AI: The practice of measuring how well an AI agent completes its intended tasks, including accuracy, reliability, safety, cost, and use of tools. Evaluation typically combines test scenarios, automated scoring, and human review before and after deployment.
Agent HarnessKey termAgentic AI: The surrounding software that runs an AI agent: it manages the loop of calling a model, executing tool calls, handling memory and context, and enforcing limits. The harness turns a raw language model into a working agent.
Agent MemoryAgentic AI: The mechanisms that let an AI agent retain and recall information across steps or sessions, such as conversation history, user preferences, and past results. Memory can be short term, within a task, or long term, stored in a database for later retrieval.
Agent MonitoringKey termAgentic AI: The ongoing tracking of AI agents in production to observe their actions, tool calls, outputs, costs, and errors. Monitoring helps teams detect failures, drift, or unsafe behavior quickly and provides an audit trail for review.
Agent OrchestrationKey termAgentic AI: The coordination of one or more AI agents, tools, and data sources so they work together to complete a larger task. Orchestration handles task routing, sequencing, hand-offs between agents, and error recovery.
Agent RuntimeKey termAgentic AI: The execution environment in which AI agents run, providing compute, state management, tool access, security controls, and scaling. A runtime lets agents operate reliably in production rather than only in development or testing.
Agent ToolsKey termAgentic AI: External functions, APIs, and services an AI agent can call to take action or retrieve information, such as searching the web, querying a database, or sending an email. Tools extend an agent beyond generating text.
Agent-to-Agent Communication (A2A)Agentic AI: The exchange of messages, tasks, and results between separate AI agents so they can collaborate on work. A2A also refers to an open protocol that defines a standard way for agents built by different vendors to discover and communicate with each other.
Agentic AIKey termAgentic AI: Agentic AI is a category of artificial intelligence in which systems can autonomously plan, reason, use tools, and take multi-step actions to achieve goals with limited human input.
Agentic GovernanceKey termAI Governance & Compliance: The policies, controls, and oversight processes that define what autonomous AI agents are allowed to do, which data and tools they can access, and how their actions are logged, reviewed, and corrected. It extends AI governance to systems that act independently.
Agentic RAGKey termLLM & GenAI: A retrieval-augmented generation approach in which an AI agent plans its own search steps, chooses which sources or tools to query, and checks results before answering. It suits complex questions that need several rounds of retrieval and reasoning.
AIKey termEnterprise AI Strategy: Artificial intelligence: the field of computer science focused on building systems that perform tasks normally requiring human intelligence, such as understanding language, recognizing patterns, making predictions, and solving problems. It includes machine learning, deep learning, and generative AI.
AI AgentsKey termAgentic AI: Software systems that use AI models to pursue a goal with some autonomy: they plan steps, call tools or APIs, observe results, and adjust. Unlike a simple chatbot, an agent can take actions and complete multistep tasks.
AI AlignmentAI Governance & Compliance: The effort to ensure AI systems behave in line with human intentions, values, and instructions. Alignment work aims to keep models helpful, honest, and safe, and to prevent them from pursuing harmful or unintended goals.
AI ComplianceAI Governance & Compliance: The practice of ensuring AI systems meet applicable laws, regulations, industry standards, and internal policies. It covers areas such as data privacy, fairness, transparency, documentation, and risk assessment throughout the AI life cycle.
AI EngineerEnterprise AI Strategy: A technical role focused on building applications and systems powered by AI models, including integrating language models, designing prompts and retrieval pipelines, building agents, and deploying them to production. The role blends software engineering with applied machine learning.
AI EthicsAI Governance & Compliance: The study and application of moral principles to the design and use of AI, addressing issues such as fairness, bias, privacy, accountability, transparency, and societal impact. It guides decisions about how AI should and should not be used.
AI GovernanceKey termAI Governance & Compliance: The framework of policies, roles, processes, and controls an organization uses to oversee how AI is developed, deployed, and monitored. It aims to manage risk, ensure compliance, and keep AI use accountable and aligned with business goals.
AI in Cloud ComputingMLOps & Deployment: The use of cloud infrastructure and services to build, train, deploy, and run AI models. Organizations rent computing power, storage, and managed AI tools on demand instead of buying and maintaining their own hardware.
AI MaturityKey termEnterprise AI Strategy: A measure of how advanced an organization is in adopting AI, from early experimentation to AI embedded across operations. Maturity models typically assess strategy, data, talent, technology, governance, and measurable business value.
AI ModelsEnterprise AI Strategy: Mathematical systems trained on data to recognize patterns and produce outputs such as predictions, classifications, text, or images. Examples include machine learning models, deep neural networks, and large language models.
AI ObservabilityMLOps & Deployment: The practice of collecting and analyzing data about how AI models and applications behave in production, including inputs, outputs, latency, cost, and errors. It helps teams detect problems, explain behavior, and keep systems reliable.
AI Red TeamingAI Governance & Compliance: A structured practice of deliberately attacking or probing an AI system to uncover vulnerabilities, harmful outputs, bias, or security weaknesses before bad actors do. Findings are used to strengthen safeguards and guardrails.
AI Risk ManagementKey termAI Governance & Compliance: The process of identifying, assessing, mitigating, and monitoring risks from AI systems, such as inaccurate outputs, bias, security threats, privacy violations, and regulatory exposure. Frameworks such as the NIST AI Risk Management Framework provide common guidance.
AI TransformationEnterprise AI Strategy: The organization-wide process of redesigning operations, products, and decision-making around AI. It involves changes to strategy, data infrastructure, skills, processes, and culture, not just the deployment of individual AI tools.
AI WinterEnterprise AI Strategy: A period of reduced funding, interest, and progress in artificial intelligence research, typically following a cycle of inflated expectations. The best known AI winters occurred in the 1970s and late 1980s.
AirgappedKey termMLOps & Deployment: A deployment model in which a computer system or network is physically isolated from the public internet and other unsecured networks. Software, updates, and data must be transferred through controlled means, which reduces exposure to outside threats.
Anomaly DetectionKey termPredictive AI & ML: The practice of identifying data points, events, or patterns that differ significantly from expected behavior. Organizations use it to flag fraud, equipment faults, security intrusions, and data quality problems.
ArtifactsMLOps & Deployment: The files and outputs produced during a machine learning project, such as trained models, datasets, code, configuration files, metrics, and logs. Tracking artifacts lets teams reproduce results, compare experiments, and deploy the right version.
Artificial General IntelligenceKey termLLM & GenAI: A hypothetical form of AI that could understand, learn, and perform any intellectual task a human can, across domains, without being built for a specific job. No existing system is widely accepted as having reached this level.
Attention MechanismDeep Learning: A neural network technique that lets a model weight the parts of its input by relevance when producing each output, instead of compressing the whole input into one fixed vector. It first improved machine translation and is the basis of transformers.
AUC-ROCPredictive AI & ML: The area under the receiver operating characteristic (ROC) curve, which plots a classifier's true positive rate against its false positive rate at every decision threshold. A score of 1.0 means every positive example is ranked above every negative one, and 0.5 is no better than chance.Also: ROC AUC, AUROC
Automated Machine LearningKey termPredictive AI & ML: The use of software to automate repetitive steps in building machine learning models, such as data preparation, feature engineering, algorithm selection, and hyperparameter tuning. It helps teams produce accurate models faster with less manual effort.Also: AutoML

B

5 terms

Back PropagationDeep Learning: An algorithm for training neural networks that calculates how much each weight contributed to the prediction error. It passes the error backward through the network's layers so the weights can be adjusted to reduce future errors.
BERTLLM & GenAI: A language model introduced by Google in 2018, short for Bidirectional Encoder Representations from Transformers. It reads text in both directions at once to understand context and is widely used for classification, search, and question answering.
Bias-Variance TradeoffPredictive AI & ML: The balance between a model too simple to capture the real pattern (high bias, underfitting) and one so flexible it fits noise in the training data (high variance, overfitting). The goal is the complexity that gives the lowest error on new data.
Big DataEnterprise AI Strategy: Datasets too large, fast-moving, or varied to be handled with traditional database tools, often described by volume, velocity, and variety. Big data technologies store, process, and analyze this information to uncover patterns and insights.
Binary ClassificationPredictive AI & ML: A classification task with exactly two possible outcomes, such as fraud or not fraud. Models usually output a probability, and a threshold turns it into the final yes or no.

C

16 terms

Chain-of-Thought PromptingLLM & GenAI: Chain-of-thought (CoT) prompting is a prompt-design pattern that asks a large language model to write out intermediate reasoning steps before giving a final answer.
Citizen Data ScientistEnterprise AI Strategy: A business user without formal data science training who builds analytical or predictive models using accessible tools such as automated machine learning and low-code platforms. The role helps organizations extend analytics beyond specialist teams.
ClassificationKey termPredictive AI & ML: A supervised machine learning task that assigns each input to one of a set of predefined categories. Common examples include flagging emails as spam or not spam and predicting whether a customer will churn.
Classification Token ([CLS])LLM & GenAI: A special token placed at the start of every BERT input. Its final hidden state is used as the representation of the whole sequence for classification tasks, typically after fine-tuning.Also: CLS Token, Classify Token
CloudKey termMLOps & Deployment: A model for delivering computing resources such as servers, storage, databases, and software over the internet on demand. Organizations typically pay for what they use rather than owning and operating the underlying hardware.
ClusteringKey termPredictive AI & ML: An unsupervised machine learning technique that groups data points so that items in the same group are more similar to each other than to items in other groups. It is often used for customer segmentation and pattern discovery.
Coding AgentKey termAgentic AI: An AI agent that writes, edits, tests, and debugs software with a degree of autonomy. It can read a codebase, run commands, and make changes across files based on natural language instructions from a developer.
Computational LinguisticsLLM & GenAI: The study of human language with computational methods, combining linguistics and computer science. It overlaps heavily with natural language processing, which focuses more on building applications such as translation, parsing, and speech recognition.
Computer VisionDeep Learning: A field of artificial intelligence that enables computers to interpret and understand images and video. Common tasks include image classification, object detection, facial recognition, and reading text from documents.
Confusion MatrixPredictive AI & ML: A table that summarizes the performance of a classification model by comparing predicted labels with actual labels. It shows counts of true positives, true negatives, false positives, and false negatives.
ContainersMLOps & Deployment: Lightweight, portable software packages that bundle an application with its code, libraries, and settings so it runs the same way in any environment. Containers share the host operating system, making them faster to start than virtual machines.
Context EngineeringLLM & GenAI: The practice of deciding what information a language model receives with each request, such as instructions, retrieved documents, conversation history, and tool results. Good context engineering improves accuracy, relevance, and cost by giving the model exactly what it needs.
Context WindowLLM & GenAI: The maximum amount of text, measured in tokens, that a language model can consider at one time, including both the input and its response. Content beyond this limit is cut off or must be summarized or retrieved selectively.
Conversational AIKey termLLM & GenAI: Technology that lets software hold natural dialogue with people through text or voice, as in chatbots and virtual assistants. It combines language understanding, dialogue management, and language generation to interpret requests and respond appropriately.
Convolutional Neural NetworksDeep Learning: A type of deep learning model designed to process grid-like data such as images. It uses filters that scan across the input to detect features like edges, shapes, and textures, building up to recognize whole objects.
Cross-ValidationPredictive AI & ML: A technique for estimating how well a model will perform on new data by repeatedly splitting the dataset into training and testing portions. Averaging results across the splits gives a more reliable performance estimate than a single split.

D

16 terms

Data GovernanceKey termAI Governance & Compliance: The policies, roles, standards, and processes that ensure an organization's data is accurate, secure, consistent, and used appropriately. It covers data ownership, quality, access control, privacy, lineage, and compliance.
Data PoisoningAI Governance & Compliance: An attack in which someone deliberately corrupts the data used to train or update an AI model, causing it to learn incorrect behavior, produce biased results, or contain hidden backdoors triggered by specific inputs.
Data PreparationEnterprise AI Strategy: The process of collecting, cleaning, transforming, and organizing raw data so it can be used for analysis or machine learning. Typical steps include handling missing values, fixing errors, joining sources, and formatting features.
Data ProfilingEnterprise AI Strategy: The process of examining a dataset to understand its structure, content, and quality, such as value distributions, missing values, data types, and relationships. Profiling helps teams spot issues before analysis or model training.
Data ScienceEnterprise AI Strategy: An interdisciplinary field that uses statistics, programming, and domain knowledge to extract insights and predictions from data. Data scientists collect and prepare data, build models, and communicate findings to support decisions.
Data Science Life CycleKey termEnterprise AI Strategy: The sequence of stages in a data science project, typically problem definition, data collection, preparation, exploration, modeling, evaluation, deployment, and monitoring. The cycle is iterative, with teams revisiting earlier steps as they learn.
Decision TreePredictive AI & ML: A model that makes predictions by following a series of if-then rules arranged in a branching, tree-like structure. Each internal node tests a feature, and each leaf holds a predicted class or value.
Deep LearningKey termDeep Learning: A subset of machine learning that uses neural networks with many layers to learn patterns directly from large amounts of data. It powers applications such as image recognition, speech recognition, and language models.Also: Neural Networking and Deep Learning
Deep Learning CloudDeep Learning: Cloud computing resources and services used to train and run deep learning models. They typically provide on-demand access to GPUs or other accelerators, storage, and frameworks, so teams can scale workloads without buying specialized hardware.
Deep Learning Use CasesDeep Learning: The practical business applications of deep learning, such as image and speech recognition, document processing, fraud detection, recommendation systems, medical imaging analysis, and natural language tasks like translation and chat.
Deep ResearchKey termAgentic AI: An AI capability in which an agent conducts multistep research on a question: it searches many sources, reads and compares them, and produces a synthesized report with citations. Several AI providers offer features under this name.
Differentiable ProgrammingDeep Learning: A programming approach in which programs are written so their outputs can be automatically differentiated with respect to their parameters. This lets gradient-based optimization tune whole programs, extending the techniques used to train neural networks.
Diffusion ModelDeep Learning: A type of generative AI model that learns to create data, such as images, by gradually removing noise from random input. It is trained by adding noise to real examples and learning to reverse that process.
Document AILLM & GenAI: A capability of Enterprise h2oGPTe that reads documents with large language models and returns the fields you ask for as schema-validated JSON, without templates, labeling, or training.
Document ExtractKey termLLM & GenAI: The step that turns an unstructured document into structured data; in Enterprise h2oGPTe an extractor applies a JSON schema and an LLM of your choice to each document and returns the same fields every time.Also: Extractor, Extractors, Document Extraction
Document UnderstandingLLM & GenAI: The ability of an AI system to read a document the way a person would, recognizing its text, tables and layout and the meaning of each value.Also: Document intelligence

E

4 terms

EmbeddingLLM & GenAI: A list of numbers that represents a piece of data, such as a word, sentence, or image, so that items with similar meaning sit close together. Embeddings power semantic search, recommendations, clustering, and retrieval for language models.
EvaluationKey termMLOps & Deployment: The process of measuring how well a machine learning model or AI system performs against defined criteria, such as accuracy, reliability, fairness, or safety. Evaluation guides decisions about whether a model is ready to deploy.
Explainable AIAI Governance & Compliance: Methods and techniques that make the outputs of AI models understandable to humans, showing which factors drove a prediction or decision. Explainable AI supports trust, debugging, fairness review, and regulatory requirements.
Exponential SmoothingPredictive AI & ML: A time series forecasting method that predicts the next value as a weighted average of past observations, with weights that shrink exponentially for older data. Extensions such as Holt-Winters add trend and seasonality.

F

11 terms

Feature EngineeringKey termPredictive AI & ML: The process of creating, transforming, or combining raw data variables into inputs that help a machine learning model learn more effectively. Examples include extracting the day of week from a date or computing ratios between fields.
Feature SelectionPredictive AI & ML: The process of choosing the most relevant input variables for a machine learning model and discarding the rest. It can improve accuracy, reduce training time, and make models easier to interpret.
Federated LearningMLOps & Deployment: A machine learning approach in which a model is trained across many devices or servers that each keep their data locally. Only model updates are shared and combined, which helps protect data privacy.
FedRAMP HighKey termAI Governance & Compliance: The highest baseline of the US Federal Risk and Authorization Management Program, which sets security requirements for cloud services used by federal agencies. It applies to systems handling data where a breach could have severe or catastrophic impact.
Few-Shot PromptingLLM & GenAI: A prompting technique in which a language model is given a small number of worked examples inside the prompt before the actual task. The examples show the desired format and reasoning, helping the model respond correctly without additional training.
Fine Tuning EngineKey termLLM & GenAI: A software component or service that manages the process of fine-tuning pretrained models on an organization's own data. It typically handles data preparation, training runs, hyperparameter settings, evaluation, and deployment of the customized model.
Fine-TuningLLM & GenAI: The process of further training a pretrained model on a smaller, task-specific or domain-specific dataset. Fine-tuning adapts general capabilities to a particular use case, style, or vocabulary without building a model from scratch.
ForecastingKey termPredictive AI & ML: The practice of predicting future values based on historical data and identified patterns, usually over time. Organizations use it to estimate demand, sales, inventory needs, and resource requirements.
Forward PropagationDeep Learning: The process by which input data moves through a neural network, layer by layer, to produce an output or prediction. Each layer applies weights, biases, and activation functions to the values it receives.
Foundation ModelLLM & GenAI: A large AI model trained on broad, diverse data that can be adapted to many downstream tasks, such as writing, summarization, coding, or image analysis. Large language models are the most common example.
Function CallingAgentic AI: A capability that lets a language model produce structured requests to call external functions or APIs, including the function name and arguments. The application executes the call and returns the result to the model.

G

11 terms

GDPR / EU AI Act / ISO 27001Key termAI Governance & Compliance: Three key frameworks for data and AI: the EU General Data Protection Regulation governs personal data privacy, the EU AI Act regulates AI systems by risk level, and ISO/IEC 27001 is an international standard for information security management.
Generalized Linear ModelsPredictive AI & ML: A family of statistical models that extends linear regression to handle target variables with different distributions, such as binary outcomes or counts. Logistic regression and Poisson regression are common examples.
Generative Adversarial NetworkDeep Learning: A type of generative model made of two neural networks trained against each other: a generator that creates synthetic data and a discriminator that tries to tell it apart from real data. The competition improves realism over time.
Generative AIKey termLLM & GenAI: A category of artificial intelligence that creates new content, such as text, images, audio, code, or video, based on patterns learned from training data. It responds to prompts rather than only classifying or predicting from existing data.
Generative Pre-trained TransformerKey termLLM & GenAI: A family of large language models, known as GPT, that are pretrained on large amounts of text to predict the next word and then generate new text. The name describes the approach: generative output, pretraining, and the transformer architecture.Also: GPT
GLUE BenchmarkLLM & GenAI: The General Language Understanding Evaluation benchmark: nine English tasks, such as sentiment, paraphrase, and inference, used to compare natural language understanding models. Its harder successor is SuperGLUE.Also: General Language Understanding Evaluation
GPUKey termDeep Learning: A graphics processing unit: a processor built to perform many calculations in parallel. Originally designed for rendering graphics, GPUs are now widely used to train and run machine learning and deep learning models.
Gradient DescentDeep Learning: An optimization algorithm that trains machine learning models by repeatedly adjusting parameters in the direction that most reduces error. Each step follows the slope, or gradient, of the error function toward a lower value.
Grid SearchPredictive AI & ML: A hyperparameter tuning method that trains and evaluates a model for every combination of values in a predefined set of candidate settings. The combination with the best validation performance is then selected.
Grounded GenerationKey termLLM & GenAI: A method of producing AI responses that are tied to specific, verifiable source material, such as documents or databases, rather than relying only on what the model learned in training. It reduces unsupported answers and often includes citations.
GuardrailsAI Governance & Compliance: Controls placed around an AI system to keep its inputs and outputs within safe, accurate, and approved limits. Examples include filtering harmful content, blocking sensitive data, restricting topics, and validating responses before they reach users.

H

3 terms

HallucinationsKey termLLM & GenAI: An AI hallucination is when a language model generates content that sounds plausible but is factually wrong, made up, or not supported by the source data provided to it.Also: AI Hallucination
HIPAAKey termAI Governance & Compliance: The Health Insurance Portability and Accountability Act, a US law that sets standards for protecting the privacy and security of individually identifiable health information. It applies to healthcare providers, health plans, clearinghouses, and their business associates.
Hyperparameter OptimizationPredictive AI & ML: The process of finding the best configuration settings for a machine learning algorithm, such as learning rate or tree depth, which are set before training rather than learned from data. Common methods include grid search, random search, and Bayesian optimization.

I

3 terms

Information RetrievalKey termLLM & GenAI: The field and practice of finding relevant documents or data from a large collection in response to a query. Search engines are the best-known example, and retrieval is a core step in systems that feed information to language models.
Intelligent Character Recognition (ICR)LLM & GenAI: Reading with understanding: LLMs and vision-language models recognize the characters, handwriting and layout on a page and understand what each value means, so differently labeled fields land in the same place.
Intelligent Document Processing (IDP)LLM & GenAI: The category of software that turns documents such as forms, claims, invoices and contracts into structured data a business system can use.

K

3 terms

Kaggle GrandmasterKey termEnterprise AI Strategy: The highest performance tier on Kaggle, an online data science competition platform. The title is earned by achieving top results across multiple competitions and is widely recognized as a mark of elite machine learning skill.
Knowledge GraphKey termAgentic AI: A structured representation of information as entities, such as people, products, or places, connected by labeled relationships. Knowledge graphs help AI systems reason over connected facts and ground answers in reliable data.
KubernetesKey termMLOps & Deployment: An open source system for automating the deployment, scaling, and management of containerized applications across clusters of machines. It schedules workloads, restarts failed containers, and balances traffic.

L

10 terms

Language ModelingLLM & GenAI: The task of predicting the next token, or a missing token, in a sequence of text. Large language models learn grammar, facts, and style this way during pretraining, before any fine-tuning.
Large Language ModelsLLM & GenAI: AI models trained on vast amounts of text to understand and generate human language. Built on the transformer architecture with billions of parameters, they can answer questions, summarize, translate, write, and reason across many tasks.
Layer NormalizationDeep Learning: A technique that normalizes each example's values across a layer's features to zero mean and unit variance, then applies a learned scale and shift. It helps deep networks such as transformers train faster and more reliably.Also: LayerNorm
LGPD / Argentina PDPA / Colombia PDPL / Mexico LFPDPPPKey termAI Governance & Compliance: Data protection laws in Brazil (LGPD), Argentina, Colombia, and Mexico that govern how organizations collect, use, store, and share personal information. They generally require a lawful basis for processing, security safeguards, and rights for individuals to access and correct their data.
Linear RegressionPredictive AI & ML: A modeling technique that predicts a numeric outcome as a weighted sum of input variables, fitting a straight line or plane through the data. It is one of the simplest and most interpretable regression methods.
LLM EvaluationKey termMLOps & Deployment: The process of measuring how well a large language model performs on tasks such as answering questions, summarizing, or following instructions. It assesses qualities like accuracy, relevance, safety, and consistency using benchmarks, human review, or automated scoring.
LLMOpsKey termMLOps & Deployment: The set of practices and tools for deploying, monitoring, and maintaining large language model applications in production. It covers prompt management, evaluation, fine-tuning, cost control, and safety, extending MLOps to the needs of language models.
Logistic RegressionPredictive AI & ML: A modeling technique used for classification that estimates the probability of an outcome, such as yes or no, from input variables. It applies a logistic function to a weighted sum of inputs to produce values between 0 and 1.
LoRA (Low-Rank Adaptation)LLM & GenAI: A technique for fine-tuning large models efficiently by training a small set of added parameters instead of updating all of the original weights. It cuts the memory and compute needed to customize a model while keeping quality close to full fine-tuning.
Low CodeKey termEnterprise AI Strategy: An approach to building software using visual interfaces, drag-and-drop components, and configuration instead of extensive hand coding. Low-code platforms let business users and developers create applications and workflows faster.

M

22 terms

Machine LearningKey termPredictive AI & ML: A branch of artificial intelligence in which computer systems learn patterns from data to make predictions or decisions without being explicitly programmed for each task. Performance typically improves as the system is exposed to more relevant data.
Machine Learning AlgorithmsPredictive AI & ML: The mathematical procedures a computer uses to learn patterns from data and produce a model. Examples include linear regression, decision trees, gradient boosting, k-means clustering, and neural networks.
Machine Learning ExperimentPredictive AI & ML: A structured attempt to build and evaluate a machine learning model using a specific dataset, algorithm, and set of parameters. Teams run and compare many experiments to find the model that performs best.
Machine Learning LifecycleMLOps & Deployment: The end-to-end process of developing and maintaining a machine learning model. Stages typically include defining the problem, collecting and preparing data, training, evaluating, deploying, monitoring, and retraining the model as conditions change.
Machine Learning ModelsPredictive AI & ML: The output of training a machine learning algorithm on data: a learned set of patterns and parameters that can generate predictions or decisions on new inputs. Models are evaluated, deployed, and monitored over time.
Machine Learning TrainingPredictive AI & ML: The process of feeding data to a machine learning algorithm so it can adjust its internal parameters to learn patterns. The goal is a model that makes accurate predictions on new, unseen data.
Mask Token ([MASK])LLM & GenAI: A special token that replaces a random share of input tokens during masked language model pretraining, as in BERT. The model learns by predicting the original token at each masked position from the surrounding context.
MCP RunnerKey termAgentic AI: A component that hosts and runs Model Context Protocol servers, which expose tools and data to AI applications through a standard interface. It manages starting, connecting, and securing those servers so agents can use them.
Mixture of Experts (MoE)LLM & GenAI: A model architecture that splits a network into many specialized subnetworks, called experts, and uses a routing mechanism to activate only a few of them for each input. This lets models grow very large while keeping computation per request manageable.
ML InterpretabilityKey termAI Governance & Compliance: The degree to which humans can understand how a machine learning model reaches its predictions. Interpretability techniques reveal which inputs matter and how, helping teams validate models, detect bias, and explain decisions.
MLOpsKey termMLOps & Deployment: MLOps (Machine Learning Operations) is the set of engineering practices for deploying, monitoring, retraining, and governing machine learning models in production reliably and at scale.Also: Machine Learning Operations
Model Context Protocol (MCP)Key termAgentic AI: The Model Context Protocol (MCP) is an open standard that defines how large language models and AI agents connect to external tools, data sources, and other agents.
Model DistillationLLM & GenAI: A technique for training a smaller model, called the student, to reproduce the behavior of a larger model, called the teacher. The result is a faster, cheaper model that keeps much of the original's capability for a given task.
Model DriftKey termMLOps & Deployment: A decline in a model's predictive performance over time as real-world data or relationships change from what the model learned during training. Common forms include data drift and concept drift.
Model FittingPredictive AI & ML: The process of adjusting a model's parameters so that it best matches the patterns in training data. A well fitted model captures real relationships without memorizing noise, avoiding both underfitting and overfitting.
Model HubKey termMLOps & Deployment: A central repository where teams store, share, discover, and manage machine learning models, often with versions, documentation, and metadata. It makes it easier to reuse existing models and deploy them consistently.
Model MonitoringKey termMLOps & Deployment: The ongoing tracking of a deployed machine learning model's performance and behavior in production. It watches metrics such as accuracy, drift, latency, and data quality so teams can catch problems and retrain when needed.
Model Scoring / InferenceKey termMLOps & Deployment: The process of using a trained machine learning model to make predictions on new data. Scoring can happen in real time for individual requests or in batches for large datasets.
MonitoringKey termMLOps & Deployment: The continuous collection and review of metrics, logs, and alerts to track the health and performance of systems and applications. Monitoring helps teams detect failures, slowdowns, and unusual behavior quickly.
Multiclass ClassificationPredictive AI & ML: A classification task in which each input is assigned to one of three or more possible categories. Examples include identifying the type of product in an image or routing a support ticket to one of several teams.
Multilayer PerceptronDeep Learning: A basic type of feedforward neural network made of an input layer, one or more hidden layers, and an output layer. Each neuron connects to every neuron in the next layer, letting the network learn nonlinear patterns.
Multimodal AILLM & GenAI: AI systems that can process and combine more than one type of data, such as text, images, audio, and video. A multimodal model might answer a question about a chart or describe what appears in a photo.

N

10 terms

Naive BayesPredictive AI & ML: A family of simple probabilistic classification algorithms based on Bayes' theorem that assume input features are independent of each other. They are fast to train and often used for text classification tasks like spam filtering.
Named Entity RecognitionKey termLLM & GenAI: A natural language processing task that finds and labels specific items in text, such as names of people, organizations, locations, dates, and amounts. It is widely used to extract structured information from documents, emails, and records.
Natural Language GenerationKey termLLM & GenAI: The branch of AI that produces human-readable text from data or other inputs. Examples include writing report summaries from figures, drafting emails, and generating chatbot replies.Also: Text Generation
Natural Language ProcessingKey termLLM & GenAI: The field of AI concerned with enabling computers to read, interpret, and produce human language. It covers tasks such as translation, classification, summarization, sentiment analysis, and question answering.
Natural Language UnderstandingKey termLLM & GenAI: The part of natural language processing focused on interpreting the meaning and intent behind text or speech. It helps systems work out what a user is asking, including context, entities, and sentiment, so they can respond correctly.
Neural NetworkKey termDeep Learning: A machine learning model loosely inspired by the brain, made of layers of connected nodes called neurons. Each connection has a weight that is adjusted during training so the network learns to map inputs to outputs.
Neural Network ArchitectureDeep Learning: The overall design of a neural network, including the number and type of layers, how they connect, and how data flows through them. Common architectures include convolutional, recurrent, and transformer networks.
NLP PredictionKey termLLM & GenAI: The use of natural language processing models to make predictions from text, such as assigning a category, detecting sentiment, extracting entities, or estimating an outcome. It turns unstructured language into labels or scores that support decisions.
NTreesPredictive AI & ML: A model parameter that sets the number of trees built in a tree ensemble, such as a random forest or gradient boosting model. More trees can improve accuracy but increase training time and may lead to overfitting in boosting.
NumPyDeep Learning: An open source Python library for numerical computing. It provides fast multidimensional arrays and mathematical functions and serves as a foundation for many data science and machine learning tools.

O

8 terms

ObservabilityKey termMLOps & Deployment: The ability to understand a system's internal state from the data it produces, such as logs, metrics, and traces. Observability helps teams investigate why problems occur, not just detect that something is wrong.
On-premKey termMLOps & Deployment: Short for on-premises: a deployment model in which software and hardware run in an organization's own data centers or facilities rather than in a public cloud. The organization manages the infrastructure directly.
OntologyKey termAgentic AI: A formal definition of the concepts in a domain, their properties, and the relationships between them. Ontologies give data and AI systems a shared vocabulary, making information consistent and easier to connect and reason over.
Open Source LLMsKey termLLM & GenAI: Large language models whose weights, and sometimes training code and data, are publicly released for anyone to download, run, and modify under a license. Organizations can host them on their own infrastructure and customize them freely.
Operationalizing AIMLOps & Deployment: The process of moving AI models from experiments into reliable, everyday use in business operations. It includes deployment, integration with existing systems, monitoring, governance, and ongoing maintenance.
Optical Character RecognitionLLM & GenAI: Technology that converts images of printed or handwritten text, such as scanned documents, photos, or PDFs, into machine-readable text. It is a common first step in automating document processing.
OrchestrationKey termAgentic AI: The automated coordination of multiple tasks, services, or systems so they run in the right order and work together as a single workflow. In AI, it includes managing models, data pipelines, tools, and agents.
OverfittingPredictive AI & ML: A modeling problem in which a model learns the training data too closely, including its noise and random quirks. The model performs well on training data but poorly on new, unseen data.

P

12 terms

PDPA / PIPL / Privacy Act / DPDP Act / MAS GuidelinesKey termAI Governance & Compliance: Data protection laws and regulatory guidance across Asia Pacific, including privacy laws in Singapore, China, Australia, and India and guidelines from the Monetary Authority of Singapore. They govern how organizations collect, use, transfer, and secure personal data.
PII Detection and RedactionAI Governance & Compliance: Finding personally identifiable information such as names, account numbers and addresses in text or documents and masking it before a model or a person sees it.Also: PII redaction, Personally identifiable information
Precision and RecallPredictive AI & ML: Two measures of a classifier's positive predictions. Precision is the share of predicted positives that are correct, and recall is the share of actual positives the model finds; raising one usually lowers the other.
PredictionPredictive AI & ML: The output a trained model produces when given new input data, such as a forecast value, a category, or a probability. Predictions are used to support decisions like approving a loan or scheduling maintenance.
Predictive AIKey termPredictive AI & ML: The use of machine learning and statistical techniques to analyze historical data and forecast future outcomes or behaviors. Common applications include demand forecasting, fraud detection, churn prediction, and risk scoring.
Probing ClassifierLLM & GenAI: A simple classifier trained on a model's internal representations to test what information they encode, such as part of speech or syntax. Probes are a common way to study what language models have learned.Also: Probing Classifiers
Prompt CatalogKey termLLM & GenAI: A shared library of tested, reusable prompts that teams can browse, reuse, and maintain. It helps organizations standardize how they instruct language models, share what works, and keep prompt versions consistent across applications.
Prompt EngineeringKey termLLM & GenAI: The practice of designing and refining the instructions given to a language model to get accurate, useful, and consistent results. It involves choosing wording, examples, structure, and constraints that guide the model's output.
Prompt InjectionAI Governance & Compliance: An attack in which malicious instructions are hidden in input text, documents, or web content to manipulate a language model into ignoring its original instructions, leaking data, or taking unintended actions.
Proprietary LLMsKey termLLM & GenAI: Large language models owned and controlled by a company that does not publicly release the model weights. Access is usually provided through a paid API or product, and users cannot inspect or modify the underlying model.
Python AutoMLPredictive AI & ML: The use of automated machine learning tools through Python, the most widely used programming language for data science. Python libraries and interfaces let practitioners automate model building, tuning, and evaluation within their existing code and notebooks.
PytorchDeep Learning: An open source machine learning framework for Python used to build and train neural networks. It is popular in research and production for its flexible approach to defining and debugging models.

Q

2 terms

QuantizationLLM & GenAI: A technique that reduces the numerical precision of a model's weights, for example from 16-bit to 4-bit numbers. It shrinks memory use and speeds up inference, allowing large models to run on smaller or cheaper hardware with modest accuracy loss.
Query Data (Text2SQL)Key termLLM & GenAI: A capability that converts a question written in everyday language into a SQL query, runs it against a database, and returns the result. It lets business users get answers from structured data without writing code.

R

14 terms

RAG EvaluationKey termMLOps & Deployment: The process of measuring how well a retrieval-augmented generation system performs. It assesses both retrieval, whether the right source documents are found, and generation, whether answers are accurate, relevant, and grounded in those sources.
Random ForestPredictive AI & ML: An ensemble machine learning method that builds many decision trees on random subsets of the data and features, then combines their outputs. Averaging or voting across trees improves accuracy and reduces overfitting compared with a single tree.
Recommendation SystemEnterprise AI Strategy: A system that predicts and suggests items a user is likely to want, such as products, content, or next actions, based on past behavior, similar users, and item attributes. Common techniques include collaborative filtering and machine learning.
Recurrent NetworkDeep Learning: A type of neural network designed for sequential data such as text, speech, or time series. It keeps an internal memory of earlier inputs, so each output depends on both the current input and what came before.
RegressionKey termPredictive AI & ML: A supervised machine learning task that predicts a continuous numeric value, such as price, revenue, or temperature, from input variables. It also describes the family of methods used to model relationships between variables.
Regression TreesPredictive AI & ML: Decision trees designed to predict continuous numeric values rather than categories. The data is split into branches based on feature values, and each leaf predicts a number, typically the average of the training examples it contains.
Regularization Parameter (Lambda)Predictive AI & ML: A setting that controls how strongly a model is penalized for complexity during training, typically by shrinking coefficient sizes. Higher values produce simpler models that are less likely to overfit, while lower values allow closer fitting to the training data.
Reinforcement LearningDeep Learning: A type of machine learning in which an agent learns to make decisions by taking actions in an environment and receiving rewards or penalties. Over time it learns strategies that maximize total reward.
Reinforcement Learning from Human Feedback (RLHF)LLM & GenAI: A training method in which people rank or rate a model's responses, and those judgments are used to train a reward signal that guides further learning. It helps align language models with human preferences for helpfulness and safety.
Responsible AIKey termAI Governance & Compliance: An approach to designing, building, and deploying AI that is fair, transparent, accountable, safe, and respectful of privacy. It turns ethical principles into concrete practices, reviews, and controls across the AI life cycle.
Retrieval PipelineLLM & GenAI: The sequence of steps that finds and prepares relevant information for an AI system, typically ingesting documents, splitting them into chunks, creating embeddings, searching an index, and ranking results. It supplies the context used in retrieval-augmented generation.
Retrieval-Augmented Generation (RAG)LLM & GenAI: Retrieval-Augmented Generation (RAG) is an AI architecture that improves large language model outputs by retrieving relevant passages from an external knowledge source and adding them to the prompt before generation.Also: Graph RAG, Agentic RAG
Risk Governance FrameworkAI Governance & Compliance: A structured set of policies, roles, processes, and controls an organization uses to identify, assess, manage, and oversee risk. It defines accountability, risk appetite, and reporting so decisions about risk are consistent and transparent.
RuleFitPredictive AI & ML: An interpretable machine learning method that extracts decision rules from an ensemble of trees and combines them with the original features in a sparse linear model. The result is a set of readable rules with weights.

S

22 terms

SaaSKey termMLOps & Deployment: Software as a service: a delivery model in which software is hosted by a provider and accessed by users over the internet, usually through a subscription. The provider handles infrastructure, maintenance, and updates.
Self-AttentionDeep Learning: An attention mechanism in which each element of a sequence, such as each word in a sentence, is compared with the elements of that same sequence to decide how much each one should shape its new representation. In text generation models, each element looks only at earlier ones. It is the core operation of transformer models.
Semantic SearchLLM & GenAI: A search method that finds results based on meaning and intent rather than exact keyword matches. It typically compares embeddings of the query and documents, so a search for "car repair" can also find "auto maintenance."
Sentiment AnalysisLLM & GenAI: A natural language processing technique that determines the emotional tone of text, such as positive, negative, or neutral. It is commonly used to analyze customer reviews, support tickets, survey responses, and social media posts.
Separator Token ([SEP])LLM & GenAI: A special token that BERT and similar models place at the end of each input segment, marking the boundary between two segments, such as a question and the passage that answers it.Also: SEP Token, Separate Token
Shapley ValuesAI Governance & Compliance: A method from cooperative game theory used to explain machine learning predictions by assigning each input feature a fair share of its contribution to the result. SHAP is a popular approach built on Shapley values.
SLM DevelopmentKey termLLM & GenAI: The process of building, training, or adapting small language models for specific tasks or environments. It often involves distillation, fine-tuning on domain data, and optimization so the model runs efficiently on limited hardware.
Small Language Model (SLM)Key termLLM & GenAI: A language model with far fewer parameters than a large language model, designed to be cheaper and faster to run. Small language models are often tuned for specific tasks and can run on local devices or modest servers.
SOC 2Key termAI Governance & Compliance: An auditing framework developed by the American Institute of Certified Public Accountants that evaluates a service organization's controls for security, availability, processing integrity, confidentiality, and privacy. A SOC 2 report is issued by an independent auditor.
Sovereign AIKey termMLOps & Deployment: AI that an organization owns and controls: its agents, models, data, and results stay under its control, and it can run them in any cloud, on premises, or air-gapped.
Speech-to-TextLLM & GenAI: Technology that converts spoken audio into written text, also called automatic speech recognition. It is used for transcription, voice assistants, call center analytics, and captioning.
SQuADLLM & GenAI: The Stanford Question Answering Dataset, a reading comprehension benchmark of questions about Wikipedia articles where each answer is a span of the passage. Version 2.0 adds questions the passage cannot answer.Also: Stanford Question Answering Dataset
Stack EnsemblePredictive AI & ML: A modeling technique that combines predictions from several different base models by training a second model, called a metalearner, to learn how best to weight them. It often achieves higher accuracy than any single base model.
Stochastic Gradient DescentPredictive AI & ML: A version of gradient descent that updates a model's parameters from one example or a small batch at a time instead of the full dataset. Each update is far cheaper to compute, which is why it and its variants, such as Adam, train most neural networks.Also: SGD
Structured OutputLLM & GenAI: Model output that follows a defined, machine-readable format, such as JSON matching a schema, rather than free-form text. It makes language model responses reliable to parse and easy to pass into other software.
Structured vs Unstructured DataEnterprise AI Strategy: The distinction between data organized in a fixed format, such as rows and columns in a database, and data without a predefined model, such as text, images, audio, and video.
Supervised Machine LearningKey termPredictive AI & ML: An approach to machine learning in which models learn from labeled examples, where each input is paired with the correct output. Classification and regression are the two main types of supervised learning tasks.Also: Supervised Learning
Support Vector MachinePredictive AI & ML: A supervised learning algorithm that classifies data by finding the boundary that separates categories with the widest possible margin. Using kernel functions, it can also handle complex, nonlinear relationships and regression tasks.
System of ContentKey termEnterprise AI Strategy: A platform that stores, manages, and governs an organization's documents and other unstructured content, making it searchable and reusable. The term is used by analogy to a system of record, which manages structured transactional data.
System of IntelligenceKey termEnterprise AI Strategy: A layer of software that sits on top of systems of record and applies analytics and AI to their combined data to generate insights, predictions, and recommended actions. It connects data across applications to support decisions.
System of RecordKey termEnterprise AI Strategy: The authoritative source for a specific type of business data, such as customer records in a CRM or financial transactions in an ERP. Other systems rely on it as the single trusted version of that data.
System PromptKey termLLM & GenAI: A set of instructions given to a language model, separate from user messages, that defines its role, tone, rules, and boundaries for a conversation. It shapes how the model behaves across every response in a session.

T

15 terms

Tabular Foundation ModelKey termPredictive AI & ML: A large pretrained model designed to work with structured data organized in rows and columns, such as spreadsheets and database tables. It can make predictions on new tabular datasets with little or no task-specific training.
Target LeakagePredictive AI & ML: A modeling error that occurs when training data includes information that would not be available at prediction time, often because it is derived from the outcome itself. It makes models look accurate in testing but fail in production.
Target VariablePredictive AI & ML: The outcome a supervised machine learning model is trained to predict, such as whether a customer churns or the price of a house. It is also called the label, response, or dependent variable.
Telemetry & OpsKey termMLOps & Deployment: The collection of operational data, such as metrics, logs, events, and usage, from running systems, combined with the practices used to operate them. Teams use telemetry to monitor health, troubleshoot issues, and manage performance.
Text ClassificationLLM & GenAI: Assigning a label to a piece of text, such as its sentiment, topic, intent, or whether it is spam. It is one of the most common natural language processing tasks and can use classic machine learning or language models.
Time Series DataPredictive AI & ML: Data points collected or recorded in time order, usually at regular intervals, such as daily sales, hourly sensor readings, or monthly website traffic. Its order and time dependencies must be preserved during analysis.
Time Series ModelsKey termPredictive AI & ML: Models designed to analyze data ordered over time and forecast future values based on past observations, trends, and seasonal patterns. Examples include ARIMA, exponential smoothing, and machine learning approaches that use lagged features.
TokenKey termLLM & GenAI: A unit of text that a language model processes, often a word, part of a word, or a punctuation mark. Models measure input limits, output length, and usage costs in tokens.
TokenizationLLM & GenAI: The process of breaking text into smaller units, called tokens, that a language model can process. Each token is mapped to a number, and the method used affects how efficiently a model handles different languages and vocabulary.
TokenomicsKey termLLM & GenAI: In generative AI, the economics of token usage: how the number of input and output tokens drives cost, speed, and capacity when running language models. Managing it involves choosing models, prompt length, and caching to control spend.
Tool BuilderKey termAgentic AI: A capability for creating custom tools that AI agents can call, typically by wrapping an API, database query, or script with a defined name, description, and inputs. It lets teams extend agents with organization-specific actions.
Tracing MatrixKey termMLOps & Deployment: A structured view that maps traces, the recorded steps of individual requests or AI interactions, against dimensions such as components, models, or evaluation criteria. It helps teams locate where errors, delays, or quality issues occur.
Training SetsPredictive AI & ML: The portion of a dataset used to teach a machine learning model by letting it learn patterns and adjust its parameters. It is kept separate from validation and test data used to evaluate performance.
Transfer LearningDeep Learning: A machine learning technique in which a model trained on one task is reused as the starting point for a related task. It reduces the data and computing power needed to build accurate models.
Transformer ArchitectureKey termDeep Learning: A neural network design that uses a mechanism called attention to weigh the relationships between all parts of an input at once. Transformers are the foundation of most modern large language models.Also: Deep Learning Transformer

U

2 terms

UnderfittingPredictive AI & ML: A modeling problem in which a model is too simple to capture the underlying patterns in the data. It performs poorly on both training data and new data, often because of too few features or too little complexity.
Unsupervised Machine LearningKey termPredictive AI & ML: An approach to machine learning in which models find patterns, groupings, or structure in data that has no labeled outcomes. Common tasks include clustering, anomaly detection, and dimensionality reduction.Also: Unsupervised Learning

V

7 terms

Validation SetsPredictive AI & ML: A portion of data held out from training and used to evaluate a model during development, tune hyperparameters, and compare candidate models. It helps detect overfitting before final testing on unseen data.
VectorLLM & GenAI: An ordered list of numbers that represents data in a mathematical space. In AI, vectors such as embeddings capture the features or meaning of text, images, or other data so they can be compared by distance.
Vector DatabaseLLM & GenAI: A database designed to store and search vectors, such as embeddings, by similarity rather than exact match. It lets applications quickly find the most related documents, images, or records and is a core component of retrieval-augmented generation.
Vertical AgentsKey termAgentic AI: AI agents designed for a specific industry or business function, such as healthcare, banking, or insurance claims. They combine domain knowledge, specialized data, and relevant workflows to perform tasks more accurately than general-purpose agents.
Vision ModelingKey termDeep Learning: The process of building machine learning models that interpret images or video. These models perform tasks such as classifying images, detecting and locating objects, segmenting scenes, and extracting information from visual documents.
Vision-Language Model (VLM)LLM & GenAI: A model that takes images and text together as input, so it can read a scanned page, a chart or a photo and answer or extract information from it.
VPCKey termMLOps & Deployment: A virtual private cloud: a logically isolated section of a public cloud where an organization runs resources in its own private network. It controls IP addressing, subnets, routing, and access rules.

W

3 terms

Weights and BiasesDeep Learning: The adjustable parameters of a neural network. Weights set the strength of connections between neurons, and biases shift each neuron's output; both are tuned during training so the network makes accurate predictions.
Word2VecLLM & GenAI: A technique introduced by Google researchers in 2013 that learns vector representations of words from large text collections. Words used in similar contexts end up with similar vectors, capturing relationships such as meaning and analogy.
WordPieceLLM & GenAI: A subword tokenization algorithm, used by BERT, that splits words into frequent pieces, such as "unaffable" into "un", "##aff", and "##able". It keeps the vocabulary to a fixed size while still representing rare or unseen words.

X

1 term

XGBoostPredictive AI & ML: A widely used open source library that implements gradient boosted decision trees, building trees one after another so each corrects the errors of the previous ones. It is known for speed and strong performance on tabular data.

Z

2 terms

Zero-Shot LearningDeep Learning: A machine learning approach in which a model performs a task or recognizes categories it was not explicitly trained on. It relies on general knowledge or descriptions learned during training rather than labeled examples.
Zero-Shot PromptingLLM & GenAI: A prompting technique in which a language model is asked to perform a task with only instructions and no examples. It relies on knowledge the model gained during training to understand and complete the request.

See what H2O AI can do on your data

Request a demo

Key AI terms

AI Glossary

Plain-language definitions of the terms behind predictive, generative, and agentic AI. New to AI? Start with the key terms.

237 terms

A

35 terms

Activation FunctionDeep Learning: A mathematical function applied to each neuron's output in a neural network that decides how strongly a signal passes forward. Nonlinear activations such as ReLU, sigmoid, and tanh let networks learn complex patterns rather than only straight-line relationships.
Agent BuilderKey termAgentic AI: A tool or environment for creating AI agents by defining their instructions, connected tools, data sources, and workflows. Agent builders often offer visual or low-code interfaces so teams can configure, test, and deploy agents without writing everything from scratch.
Agent EvaluationKey termAgentic AI: The practice of measuring how well an AI agent completes its intended tasks, including accuracy, reliability, safety, cost, and use of tools. Evaluation typically combines test scenarios, automated scoring, and human review before and after deployment.
Agent HarnessKey termAgentic AI: The surrounding software that runs an AI agent: it manages the loop of calling a model, executing tool calls, handling memory and context, and enforcing limits. The harness turns a raw language model into a working agent.
Agent MemoryAgentic AI: The mechanisms that let an AI agent retain and recall information across steps or sessions, such as conversation history, user preferences, and past results. Memory can be short term, within a task, or long term, stored in a database for later retrieval.
Agent MonitoringKey termAgentic AI: The ongoing tracking of AI agents in production to observe their actions, tool calls, outputs, costs, and errors. Monitoring helps teams detect failures, drift, or unsafe behavior quickly and provides an audit trail for review.
Agent OrchestrationKey termAgentic AI: The coordination of one or more AI agents, tools, and data sources so they work together to complete a larger task. Orchestration handles task routing, sequencing, hand-offs between agents, and error recovery.
Agent RuntimeKey termAgentic AI: The execution environment in which AI agents run, providing compute, state management, tool access, security controls, and scaling. A runtime lets agents operate reliably in production rather than only in development or testing.
Agent ToolsKey termAgentic AI: External functions, APIs, and services an AI agent can call to take action or retrieve information, such as searching the web, querying a database, or sending an email. Tools extend an agent beyond generating text.
Agent-to-Agent Communication (A2A)Agentic AI: The exchange of messages, tasks, and results between separate AI agents so they can collaborate on work. A2A also refers to an open protocol that defines a standard way for agents built by different vendors to discover and communicate with each other.
Agentic AIKey termAgentic AI: Agentic AI is a category of artificial intelligence in which systems can autonomously plan, reason, use tools, and take multi-step actions to achieve goals with limited human input.
Agentic GovernanceKey termAI Governance & Compliance: The policies, controls, and oversight processes that define what autonomous AI agents are allowed to do, which data and tools they can access, and how their actions are logged, reviewed, and corrected. It extends AI governance to systems that act independently.
Agentic RAGKey termLLM & GenAI: A retrieval-augmented generation approach in which an AI agent plans its own search steps, chooses which sources or tools to query, and checks results before answering. It suits complex questions that need several rounds of retrieval and reasoning.
AIKey termEnterprise AI Strategy: Artificial intelligence: the field of computer science focused on building systems that perform tasks normally requiring human intelligence, such as understanding language, recognizing patterns, making predictions, and solving problems. It includes machine learning, deep learning, and generative AI.
AI AgentsKey termAgentic AI: Software systems that use AI models to pursue a goal with some autonomy: they plan steps, call tools or APIs, observe results, and adjust. Unlike a simple chatbot, an agent can take actions and complete multistep tasks.
AI AlignmentAI Governance & Compliance: The effort to ensure AI systems behave in line with human intentions, values, and instructions. Alignment work aims to keep models helpful, honest, and safe, and to prevent them from pursuing harmful or unintended goals.
AI ComplianceAI Governance & Compliance: The practice of ensuring AI systems meet applicable laws, regulations, industry standards, and internal policies. It covers areas such as data privacy, fairness, transparency, documentation, and risk assessment throughout the AI life cycle.
AI EngineerEnterprise AI Strategy: A technical role focused on building applications and systems powered by AI models, including integrating language models, designing prompts and retrieval pipelines, building agents, and deploying them to production. The role blends software engineering with applied machine learning.
AI EthicsAI Governance & Compliance: The study and application of moral principles to the design and use of AI, addressing issues such as fairness, bias, privacy, accountability, transparency, and societal impact. It guides decisions about how AI should and should not be used.
AI GovernanceKey termAI Governance & Compliance: The framework of policies, roles, processes, and controls an organization uses to oversee how AI is developed, deployed, and monitored. It aims to manage risk, ensure compliance, and keep AI use accountable and aligned with business goals.
AI in Cloud ComputingMLOps & Deployment: The use of cloud infrastructure and services to build, train, deploy, and run AI models. Organizations rent computing power, storage, and managed AI tools on demand instead of buying and maintaining their own hardware.
AI MaturityKey termEnterprise AI Strategy: A measure of how advanced an organization is in adopting AI, from early experimentation to AI embedded across operations. Maturity models typically assess strategy, data, talent, technology, governance, and measurable business value.
AI ModelsEnterprise AI Strategy: Mathematical systems trained on data to recognize patterns and produce outputs such as predictions, classifications, text, or images. Examples include machine learning models, deep neural networks, and large language models.
AI ObservabilityMLOps & Deployment: The practice of collecting and analyzing data about how AI models and applications behave in production, including inputs, outputs, latency, cost, and errors. It helps teams detect problems, explain behavior, and keep systems reliable.
AI Red TeamingAI Governance & Compliance: A structured practice of deliberately attacking or probing an AI system to uncover vulnerabilities, harmful outputs, bias, or security weaknesses before bad actors do. Findings are used to strengthen safeguards and guardrails.
AI Risk ManagementKey termAI Governance & Compliance: The process of identifying, assessing, mitigating, and monitoring risks from AI systems, such as inaccurate outputs, bias, security threats, privacy violations, and regulatory exposure. Frameworks such as the NIST AI Risk Management Framework provide common guidance.
AI TransformationEnterprise AI Strategy: The organization-wide process of redesigning operations, products, and decision-making around AI. It involves changes to strategy, data infrastructure, skills, processes, and culture, not just the deployment of individual AI tools.
AI WinterEnterprise AI Strategy: A period of reduced funding, interest, and progress in artificial intelligence research, typically following a cycle of inflated expectations. The best known AI winters occurred in the 1970s and late 1980s.
AirgappedKey termMLOps & Deployment: A deployment model in which a computer system or network is physically isolated from the public internet and other unsecured networks. Software, updates, and data must be transferred through controlled means, which reduces exposure to outside threats.
Anomaly DetectionKey termPredictive AI & ML: The practice of identifying data points, events, or patterns that differ significantly from expected behavior. Organizations use it to flag fraud, equipment faults, security intrusions, and data quality problems.
ArtifactsMLOps & Deployment: The files and outputs produced during a machine learning project, such as trained models, datasets, code, configuration files, metrics, and logs. Tracking artifacts lets teams reproduce results, compare experiments, and deploy the right version.
Artificial General IntelligenceKey termLLM & GenAI: A hypothetical form of AI that could understand, learn, and perform any intellectual task a human can, across domains, without being built for a specific job. No existing system is widely accepted as having reached this level.
Attention MechanismDeep Learning: A neural network technique that lets a model weight the parts of its input by relevance when producing each output, instead of compressing the whole input into one fixed vector. It first improved machine translation and is the basis of transformers.
AUC-ROCPredictive AI & ML: The area under the receiver operating characteristic (ROC) curve, which plots a classifier's true positive rate against its false positive rate at every decision threshold. A score of 1.0 means every positive example is ranked above every negative one, and 0.5 is no better than chance.Also: ROC AUC, AUROC
Automated Machine LearningKey termPredictive AI & ML: The use of software to automate repetitive steps in building machine learning models, such as data preparation, feature engineering, algorithm selection, and hyperparameter tuning. It helps teams produce accurate models faster with less manual effort.Also: AutoML

B

5 terms

Back PropagationDeep Learning: An algorithm for training neural networks that calculates how much each weight contributed to the prediction error. It passes the error backward through the network's layers so the weights can be adjusted to reduce future errors.
BERTLLM & GenAI: A language model introduced by Google in 2018, short for Bidirectional Encoder Representations from Transformers. It reads text in both directions at once to understand context and is widely used for classification, search, and question answering.
Bias-Variance TradeoffPredictive AI & ML: The balance between a model too simple to capture the real pattern (high bias, underfitting) and one so flexible it fits noise in the training data (high variance, overfitting). The goal is the complexity that gives the lowest error on new data.
Big DataEnterprise AI Strategy: Datasets too large, fast-moving, or varied to be handled with traditional database tools, often described by volume, velocity, and variety. Big data technologies store, process, and analyze this information to uncover patterns and insights.
Binary ClassificationPredictive AI & ML: A classification task with exactly two possible outcomes, such as fraud or not fraud. Models usually output a probability, and a threshold turns it into the final yes or no.

C

16 terms

Chain-of-Thought PromptingLLM & GenAI: Chain-of-thought (CoT) prompting is a prompt-design pattern that asks a large language model to write out intermediate reasoning steps before giving a final answer.
Citizen Data ScientistEnterprise AI Strategy: A business user without formal data science training who builds analytical or predictive models using accessible tools such as automated machine learning and low-code platforms. The role helps organizations extend analytics beyond specialist teams.
ClassificationKey termPredictive AI & ML: A supervised machine learning task that assigns each input to one of a set of predefined categories. Common examples include flagging emails as spam or not spam and predicting whether a customer will churn.
Classification Token ([CLS])LLM & GenAI: A special token placed at the start of every BERT input. Its final hidden state is used as the representation of the whole sequence for classification tasks, typically after fine-tuning.Also: CLS Token, Classify Token
CloudKey termMLOps & Deployment: A model for delivering computing resources such as servers, storage, databases, and software over the internet on demand. Organizations typically pay for what they use rather than owning and operating the underlying hardware.
ClusteringKey termPredictive AI & ML: An unsupervised machine learning technique that groups data points so that items in the same group are more similar to each other than to items in other groups. It is often used for customer segmentation and pattern discovery.
Coding AgentKey termAgentic AI: An AI agent that writes, edits, tests, and debugs software with a degree of autonomy. It can read a codebase, run commands, and make changes across files based on natural language instructions from a developer.
Computational LinguisticsLLM & GenAI: The study of human language with computational methods, combining linguistics and computer science. It overlaps heavily with natural language processing, which focuses more on building applications such as translation, parsing, and speech recognition.
Computer VisionDeep Learning: A field of artificial intelligence that enables computers to interpret and understand images and video. Common tasks include image classification, object detection, facial recognition, and reading text from documents.
Confusion MatrixPredictive AI & ML: A table that summarizes the performance of a classification model by comparing predicted labels with actual labels. It shows counts of true positives, true negatives, false positives, and false negatives.
ContainersMLOps & Deployment: Lightweight, portable software packages that bundle an application with its code, libraries, and settings so it runs the same way in any environment. Containers share the host operating system, making them faster to start than virtual machines.
Context EngineeringLLM & GenAI: The practice of deciding what information a language model receives with each request, such as instructions, retrieved documents, conversation history, and tool results. Good context engineering improves accuracy, relevance, and cost by giving the model exactly what it needs.
Context WindowLLM & GenAI: The maximum amount of text, measured in tokens, that a language model can consider at one time, including both the input and its response. Content beyond this limit is cut off or must be summarized or retrieved selectively.
Conversational AIKey termLLM & GenAI: Technology that lets software hold natural dialogue with people through text or voice, as in chatbots and virtual assistants. It combines language understanding, dialogue management, and language generation to interpret requests and respond appropriately.
Convolutional Neural NetworksDeep Learning: A type of deep learning model designed to process grid-like data such as images. It uses filters that scan across the input to detect features like edges, shapes, and textures, building up to recognize whole objects.
Cross-ValidationPredictive AI & ML: A technique for estimating how well a model will perform on new data by repeatedly splitting the dataset into training and testing portions. Averaging results across the splits gives a more reliable performance estimate than a single split.

D

16 terms

Data GovernanceKey termAI Governance & Compliance: The policies, roles, standards, and processes that ensure an organization's data is accurate, secure, consistent, and used appropriately. It covers data ownership, quality, access control, privacy, lineage, and compliance.
Data PoisoningAI Governance & Compliance: An attack in which someone deliberately corrupts the data used to train or update an AI model, causing it to learn incorrect behavior, produce biased results, or contain hidden backdoors triggered by specific inputs.
Data PreparationEnterprise AI Strategy: The process of collecting, cleaning, transforming, and organizing raw data so it can be used for analysis or machine learning. Typical steps include handling missing values, fixing errors, joining sources, and formatting features.
Data ProfilingEnterprise AI Strategy: The process of examining a dataset to understand its structure, content, and quality, such as value distributions, missing values, data types, and relationships. Profiling helps teams spot issues before analysis or model training.
Data ScienceEnterprise AI Strategy: An interdisciplinary field that uses statistics, programming, and domain knowledge to extract insights and predictions from data. Data scientists collect and prepare data, build models, and communicate findings to support decisions.
Data Science Life CycleKey termEnterprise AI Strategy: The sequence of stages in a data science project, typically problem definition, data collection, preparation, exploration, modeling, evaluation, deployment, and monitoring. The cycle is iterative, with teams revisiting earlier steps as they learn.
Decision TreePredictive AI & ML: A model that makes predictions by following a series of if-then rules arranged in a branching, tree-like structure. Each internal node tests a feature, and each leaf holds a predicted class or value.
Deep LearningKey termDeep Learning: A subset of machine learning that uses neural networks with many layers to learn patterns directly from large amounts of data. It powers applications such as image recognition, speech recognition, and language models.Also: Neural Networking and Deep Learning
Deep Learning CloudDeep Learning: Cloud computing resources and services used to train and run deep learning models. They typically provide on-demand access to GPUs or other accelerators, storage, and frameworks, so teams can scale workloads without buying specialized hardware.
Deep Learning Use CasesDeep Learning: The practical business applications of deep learning, such as image and speech recognition, document processing, fraud detection, recommendation systems, medical imaging analysis, and natural language tasks like translation and chat.
Deep ResearchKey termAgentic AI: An AI capability in which an agent conducts multistep research on a question: it searches many sources, reads and compares them, and produces a synthesized report with citations. Several AI providers offer features under this name.
Differentiable ProgrammingDeep Learning: A programming approach in which programs are written so their outputs can be automatically differentiated with respect to their parameters. This lets gradient-based optimization tune whole programs, extending the techniques used to train neural networks.
Diffusion ModelDeep Learning: A type of generative AI model that learns to create data, such as images, by gradually removing noise from random input. It is trained by adding noise to real examples and learning to reverse that process.
Document AILLM & GenAI: A capability of Enterprise h2oGPTe that reads documents with large language models and returns the fields you ask for as schema-validated JSON, without templates, labeling, or training.
Document ExtractKey termLLM & GenAI: The step that turns an unstructured document into structured data; in Enterprise h2oGPTe an extractor applies a JSON schema and an LLM of your choice to each document and returns the same fields every time.Also: Extractor, Extractors, Document Extraction
Document UnderstandingLLM & GenAI: The ability of an AI system to read a document the way a person would, recognizing its text, tables and layout and the meaning of each value.Also: Document intelligence

E

4 terms

EmbeddingLLM & GenAI: A list of numbers that represents a piece of data, such as a word, sentence, or image, so that items with similar meaning sit close together. Embeddings power semantic search, recommendations, clustering, and retrieval for language models.
EvaluationKey termMLOps & Deployment: The process of measuring how well a machine learning model or AI system performs against defined criteria, such as accuracy, reliability, fairness, or safety. Evaluation guides decisions about whether a model is ready to deploy.
Explainable AIAI Governance & Compliance: Methods and techniques that make the outputs of AI models understandable to humans, showing which factors drove a prediction or decision. Explainable AI supports trust, debugging, fairness review, and regulatory requirements.
Exponential SmoothingPredictive AI & ML: A time series forecasting method that predicts the next value as a weighted average of past observations, with weights that shrink exponentially for older data. Extensions such as Holt-Winters add trend and seasonality.

F

11 terms

Feature EngineeringKey termPredictive AI & ML: The process of creating, transforming, or combining raw data variables into inputs that help a machine learning model learn more effectively. Examples include extracting the day of week from a date or computing ratios between fields.
Feature SelectionPredictive AI & ML: The process of choosing the most relevant input variables for a machine learning model and discarding the rest. It can improve accuracy, reduce training time, and make models easier to interpret.
Federated LearningMLOps & Deployment: A machine learning approach in which a model is trained across many devices or servers that each keep their data locally. Only model updates are shared and combined, which helps protect data privacy.
FedRAMP HighKey termAI Governance & Compliance: The highest baseline of the US Federal Risk and Authorization Management Program, which sets security requirements for cloud services used by federal agencies. It applies to systems handling data where a breach could have severe or catastrophic impact.
Few-Shot PromptingLLM & GenAI: A prompting technique in which a language model is given a small number of worked examples inside the prompt before the actual task. The examples show the desired format and reasoning, helping the model respond correctly without additional training.
Fine Tuning EngineKey termLLM & GenAI: A software component or service that manages the process of fine-tuning pretrained models on an organization's own data. It typically handles data preparation, training runs, hyperparameter settings, evaluation, and deployment of the customized model.
Fine-TuningLLM & GenAI: The process of further training a pretrained model on a smaller, task-specific or domain-specific dataset. Fine-tuning adapts general capabilities to a particular use case, style, or vocabulary without building a model from scratch.
ForecastingKey termPredictive AI & ML: The practice of predicting future values based on historical data and identified patterns, usually over time. Organizations use it to estimate demand, sales, inventory needs, and resource requirements.
Forward PropagationDeep Learning: The process by which input data moves through a neural network, layer by layer, to produce an output or prediction. Each layer applies weights, biases, and activation functions to the values it receives.
Foundation ModelLLM & GenAI: A large AI model trained on broad, diverse data that can be adapted to many downstream tasks, such as writing, summarization, coding, or image analysis. Large language models are the most common example.
Function CallingAgentic AI: A capability that lets a language model produce structured requests to call external functions or APIs, including the function name and arguments. The application executes the call and returns the result to the model.

G

11 terms

GDPR / EU AI Act / ISO 27001Key termAI Governance & Compliance: Three key frameworks for data and AI: the EU General Data Protection Regulation governs personal data privacy, the EU AI Act regulates AI systems by risk level, and ISO/IEC 27001 is an international standard for information security management.
Generalized Linear ModelsPredictive AI & ML: A family of statistical models that extends linear regression to handle target variables with different distributions, such as binary outcomes or counts. Logistic regression and Poisson regression are common examples.
Generative Adversarial NetworkDeep Learning: A type of generative model made of two neural networks trained against each other: a generator that creates synthetic data and a discriminator that tries to tell it apart from real data. The competition improves realism over time.
Generative AIKey termLLM & GenAI: A category of artificial intelligence that creates new content, such as text, images, audio, code, or video, based on patterns learned from training data. It responds to prompts rather than only classifying or predicting from existing data.
Generative Pre-trained TransformerKey termLLM & GenAI: A family of large language models, known as GPT, that are pretrained on large amounts of text to predict the next word and then generate new text. The name describes the approach: generative output, pretraining, and the transformer architecture.Also: GPT
GLUE BenchmarkLLM & GenAI: The General Language Understanding Evaluation benchmark: nine English tasks, such as sentiment, paraphrase, and inference, used to compare natural language understanding models. Its harder successor is SuperGLUE.Also: General Language Understanding Evaluation
GPUKey termDeep Learning: A graphics processing unit: a processor built to perform many calculations in parallel. Originally designed for rendering graphics, GPUs are now widely used to train and run machine learning and deep learning models.
Gradient DescentDeep Learning: An optimization algorithm that trains machine learning models by repeatedly adjusting parameters in the direction that most reduces error. Each step follows the slope, or gradient, of the error function toward a lower value.
Grid SearchPredictive AI & ML: A hyperparameter tuning method that trains and evaluates a model for every combination of values in a predefined set of candidate settings. The combination with the best validation performance is then selected.
Grounded GenerationKey termLLM & GenAI: A method of producing AI responses that are tied to specific, verifiable source material, such as documents or databases, rather than relying only on what the model learned in training. It reduces unsupported answers and often includes citations.
GuardrailsAI Governance & Compliance: Controls placed around an AI system to keep its inputs and outputs within safe, accurate, and approved limits. Examples include filtering harmful content, blocking sensitive data, restricting topics, and validating responses before they reach users.

H

3 terms

HallucinationsKey termLLM & GenAI: An AI hallucination is when a language model generates content that sounds plausible but is factually wrong, made up, or not supported by the source data provided to it.Also: AI Hallucination
HIPAAKey termAI Governance & Compliance: The Health Insurance Portability and Accountability Act, a US law that sets standards for protecting the privacy and security of individually identifiable health information. It applies to healthcare providers, health plans, clearinghouses, and their business associates.
Hyperparameter OptimizationPredictive AI & ML: The process of finding the best configuration settings for a machine learning algorithm, such as learning rate or tree depth, which are set before training rather than learned from data. Common methods include grid search, random search, and Bayesian optimization.

I

3 terms

Information RetrievalKey termLLM & GenAI: The field and practice of finding relevant documents or data from a large collection in response to a query. Search engines are the best-known example, and retrieval is a core step in systems that feed information to language models.
Intelligent Character Recognition (ICR)LLM & GenAI: Reading with understanding: LLMs and vision-language models recognize the characters, handwriting and layout on a page and understand what each value means, so differently labeled fields land in the same place.
Intelligent Document Processing (IDP)LLM & GenAI: The category of software that turns documents such as forms, claims, invoices and contracts into structured data a business system can use.

K

3 terms

Kaggle GrandmasterKey termEnterprise AI Strategy: The highest performance tier on Kaggle, an online data science competition platform. The title is earned by achieving top results across multiple competitions and is widely recognized as a mark of elite machine learning skill.
Knowledge GraphKey termAgentic AI: A structured representation of information as entities, such as people, products, or places, connected by labeled relationships. Knowledge graphs help AI systems reason over connected facts and ground answers in reliable data.
KubernetesKey termMLOps & Deployment: An open source system for automating the deployment, scaling, and management of containerized applications across clusters of machines. It schedules workloads, restarts failed containers, and balances traffic.

L

10 terms

Language ModelingLLM & GenAI: The task of predicting the next token, or a missing token, in a sequence of text. Large language models learn grammar, facts, and style this way during pretraining, before any fine-tuning.
Large Language ModelsLLM & GenAI: AI models trained on vast amounts of text to understand and generate human language. Built on the transformer architecture with billions of parameters, they can answer questions, summarize, translate, write, and reason across many tasks.
Layer NormalizationDeep Learning: A technique that normalizes each example's values across a layer's features to zero mean and unit variance, then applies a learned scale and shift. It helps deep networks such as transformers train faster and more reliably.Also: LayerNorm
LGPD / Argentina PDPA / Colombia PDPL / Mexico LFPDPPPKey termAI Governance & Compliance: Data protection laws in Brazil (LGPD), Argentina, Colombia, and Mexico that govern how organizations collect, use, store, and share personal information. They generally require a lawful basis for processing, security safeguards, and rights for individuals to access and correct their data.
Linear RegressionPredictive AI & ML: A modeling technique that predicts a numeric outcome as a weighted sum of input variables, fitting a straight line or plane through the data. It is one of the simplest and most interpretable regression methods.
LLM EvaluationKey termMLOps & Deployment: The process of measuring how well a large language model performs on tasks such as answering questions, summarizing, or following instructions. It assesses qualities like accuracy, relevance, safety, and consistency using benchmarks, human review, or automated scoring.
LLMOpsKey termMLOps & Deployment: The set of practices and tools for deploying, monitoring, and maintaining large language model applications in production. It covers prompt management, evaluation, fine-tuning, cost control, and safety, extending MLOps to the needs of language models.
Logistic RegressionPredictive AI & ML: A modeling technique used for classification that estimates the probability of an outcome, such as yes or no, from input variables. It applies a logistic function to a weighted sum of inputs to produce values between 0 and 1.
LoRA (Low-Rank Adaptation)LLM & GenAI: A technique for fine-tuning large models efficiently by training a small set of added parameters instead of updating all of the original weights. It cuts the memory and compute needed to customize a model while keeping quality close to full fine-tuning.
Low CodeKey termEnterprise AI Strategy: An approach to building software using visual interfaces, drag-and-drop components, and configuration instead of extensive hand coding. Low-code platforms let business users and developers create applications and workflows faster.

M

22 terms

Machine LearningKey termPredictive AI & ML: A branch of artificial intelligence in which computer systems learn patterns from data to make predictions or decisions without being explicitly programmed for each task. Performance typically improves as the system is exposed to more relevant data.
Machine Learning AlgorithmsPredictive AI & ML: The mathematical procedures a computer uses to learn patterns from data and produce a model. Examples include linear regression, decision trees, gradient boosting, k-means clustering, and neural networks.
Machine Learning ExperimentPredictive AI & ML: A structured attempt to build and evaluate a machine learning model using a specific dataset, algorithm, and set of parameters. Teams run and compare many experiments to find the model that performs best.
Machine Learning LifecycleMLOps & Deployment: The end-to-end process of developing and maintaining a machine learning model. Stages typically include defining the problem, collecting and preparing data, training, evaluating, deploying, monitoring, and retraining the model as conditions change.
Machine Learning ModelsPredictive AI & ML: The output of training a machine learning algorithm on data: a learned set of patterns and parameters that can generate predictions or decisions on new inputs. Models are evaluated, deployed, and monitored over time.
Machine Learning TrainingPredictive AI & ML: The process of feeding data to a machine learning algorithm so it can adjust its internal parameters to learn patterns. The goal is a model that makes accurate predictions on new, unseen data.
Mask Token ([MASK])LLM & GenAI: A special token that replaces a random share of input tokens during masked language model pretraining, as in BERT. The model learns by predicting the original token at each masked position from the surrounding context.
MCP RunnerKey termAgentic AI: A component that hosts and runs Model Context Protocol servers, which expose tools and data to AI applications through a standard interface. It manages starting, connecting, and securing those servers so agents can use them.
Mixture of Experts (MoE)LLM & GenAI: A model architecture that splits a network into many specialized subnetworks, called experts, and uses a routing mechanism to activate only a few of them for each input. This lets models grow very large while keeping computation per request manageable.
ML InterpretabilityKey termAI Governance & Compliance: The degree to which humans can understand how a machine learning model reaches its predictions. Interpretability techniques reveal which inputs matter and how, helping teams validate models, detect bias, and explain decisions.
MLOpsKey termMLOps & Deployment: MLOps (Machine Learning Operations) is the set of engineering practices for deploying, monitoring, retraining, and governing machine learning models in production reliably and at scale.Also: Machine Learning Operations
Model Context Protocol (MCP)Key termAgentic AI: The Model Context Protocol (MCP) is an open standard that defines how large language models and AI agents connect to external tools, data sources, and other agents.
Model DistillationLLM & GenAI: A technique for training a smaller model, called the student, to reproduce the behavior of a larger model, called the teacher. The result is a faster, cheaper model that keeps much of the original's capability for a given task.
Model DriftKey termMLOps & Deployment: A decline in a model's predictive performance over time as real-world data or relationships change from what the model learned during training. Common forms include data drift and concept drift.
Model FittingPredictive AI & ML: The process of adjusting a model's parameters so that it best matches the patterns in training data. A well fitted model captures real relationships without memorizing noise, avoiding both underfitting and overfitting.
Model HubKey termMLOps & Deployment: A central repository where teams store, share, discover, and manage machine learning models, often with versions, documentation, and metadata. It makes it easier to reuse existing models and deploy them consistently.
Model MonitoringKey termMLOps & Deployment: The ongoing tracking of a deployed machine learning model's performance and behavior in production. It watches metrics such as accuracy, drift, latency, and data quality so teams can catch problems and retrain when needed.
Model Scoring / InferenceKey termMLOps & Deployment: The process of using a trained machine learning model to make predictions on new data. Scoring can happen in real time for individual requests or in batches for large datasets.
MonitoringKey termMLOps & Deployment: The continuous collection and review of metrics, logs, and alerts to track the health and performance of systems and applications. Monitoring helps teams detect failures, slowdowns, and unusual behavior quickly.
Multiclass ClassificationPredictive AI & ML: A classification task in which each input is assigned to one of three or more possible categories. Examples include identifying the type of product in an image or routing a support ticket to one of several teams.
Multilayer PerceptronDeep Learning: A basic type of feedforward neural network made of an input layer, one or more hidden layers, and an output layer. Each neuron connects to every neuron in the next layer, letting the network learn nonlinear patterns.
Multimodal AILLM & GenAI: AI systems that can process and combine more than one type of data, such as text, images, audio, and video. A multimodal model might answer a question about a chart or describe what appears in a photo.

N

10 terms

Naive BayesPredictive AI & ML: A family of simple probabilistic classification algorithms based on Bayes' theorem that assume input features are independent of each other. They are fast to train and often used for text classification tasks like spam filtering.
Named Entity RecognitionKey termLLM & GenAI: A natural language processing task that finds and labels specific items in text, such as names of people, organizations, locations, dates, and amounts. It is widely used to extract structured information from documents, emails, and records.
Natural Language GenerationKey termLLM & GenAI: The branch of AI that produces human-readable text from data or other inputs. Examples include writing report summaries from figures, drafting emails, and generating chatbot replies.Also: Text Generation
Natural Language ProcessingKey termLLM & GenAI: The field of AI concerned with enabling computers to read, interpret, and produce human language. It covers tasks such as translation, classification, summarization, sentiment analysis, and question answering.
Natural Language UnderstandingKey termLLM & GenAI: The part of natural language processing focused on interpreting the meaning and intent behind text or speech. It helps systems work out what a user is asking, including context, entities, and sentiment, so they can respond correctly.
Neural NetworkKey termDeep Learning: A machine learning model loosely inspired by the brain, made of layers of connected nodes called neurons. Each connection has a weight that is adjusted during training so the network learns to map inputs to outputs.
Neural Network ArchitectureDeep Learning: The overall design of a neural network, including the number and type of layers, how they connect, and how data flows through them. Common architectures include convolutional, recurrent, and transformer networks.
NLP PredictionKey termLLM & GenAI: The use of natural language processing models to make predictions from text, such as assigning a category, detecting sentiment, extracting entities, or estimating an outcome. It turns unstructured language into labels or scores that support decisions.
NTreesPredictive AI & ML: A model parameter that sets the number of trees built in a tree ensemble, such as a random forest or gradient boosting model. More trees can improve accuracy but increase training time and may lead to overfitting in boosting.
NumPyDeep Learning: An open source Python library for numerical computing. It provides fast multidimensional arrays and mathematical functions and serves as a foundation for many data science and machine learning tools.

O

8 terms

ObservabilityKey termMLOps & Deployment: The ability to understand a system's internal state from the data it produces, such as logs, metrics, and traces. Observability helps teams investigate why problems occur, not just detect that something is wrong.
On-premKey termMLOps & Deployment: Short for on-premises: a deployment model in which software and hardware run in an organization's own data centers or facilities rather than in a public cloud. The organization manages the infrastructure directly.
OntologyKey termAgentic AI: A formal definition of the concepts in a domain, their properties, and the relationships between them. Ontologies give data and AI systems a shared vocabulary, making information consistent and easier to connect and reason over.
Open Source LLMsKey termLLM & GenAI: Large language models whose weights, and sometimes training code and data, are publicly released for anyone to download, run, and modify under a license. Organizations can host them on their own infrastructure and customize them freely.
Operationalizing AIMLOps & Deployment: The process of moving AI models from experiments into reliable, everyday use in business operations. It includes deployment, integration with existing systems, monitoring, governance, and ongoing maintenance.
Optical Character RecognitionLLM & GenAI: Technology that converts images of printed or handwritten text, such as scanned documents, photos, or PDFs, into machine-readable text. It is a common first step in automating document processing.
OrchestrationKey termAgentic AI: The automated coordination of multiple tasks, services, or systems so they run in the right order and work together as a single workflow. In AI, it includes managing models, data pipelines, tools, and agents.
OverfittingPredictive AI & ML: A modeling problem in which a model learns the training data too closely, including its noise and random quirks. The model performs well on training data but poorly on new, unseen data.

P

12 terms

PDPA / PIPL / Privacy Act / DPDP Act / MAS GuidelinesKey termAI Governance & Compliance: Data protection laws and regulatory guidance across Asia Pacific, including privacy laws in Singapore, China, Australia, and India and guidelines from the Monetary Authority of Singapore. They govern how organizations collect, use, transfer, and secure personal data.
PII Detection and RedactionAI Governance & Compliance: Finding personally identifiable information such as names, account numbers and addresses in text or documents and masking it before a model or a person sees it.Also: PII redaction, Personally identifiable information
Precision and RecallPredictive AI & ML: Two measures of a classifier's positive predictions. Precision is the share of predicted positives that are correct, and recall is the share of actual positives the model finds; raising one usually lowers the other.
PredictionPredictive AI & ML: The output a trained model produces when given new input data, such as a forecast value, a category, or a probability. Predictions are used to support decisions like approving a loan or scheduling maintenance.
Predictive AIKey termPredictive AI & ML: The use of machine learning and statistical techniques to analyze historical data and forecast future outcomes or behaviors. Common applications include demand forecasting, fraud detection, churn prediction, and risk scoring.
Probing ClassifierLLM & GenAI: A simple classifier trained on a model's internal representations to test what information they encode, such as part of speech or syntax. Probes are a common way to study what language models have learned.Also: Probing Classifiers
Prompt CatalogKey termLLM & GenAI: A shared library of tested, reusable prompts that teams can browse, reuse, and maintain. It helps organizations standardize how they instruct language models, share what works, and keep prompt versions consistent across applications.
Prompt EngineeringKey termLLM & GenAI: The practice of designing and refining the instructions given to a language model to get accurate, useful, and consistent results. It involves choosing wording, examples, structure, and constraints that guide the model's output.
Prompt InjectionAI Governance & Compliance: An attack in which malicious instructions are hidden in input text, documents, or web content to manipulate a language model into ignoring its original instructions, leaking data, or taking unintended actions.
Proprietary LLMsKey termLLM & GenAI: Large language models owned and controlled by a company that does not publicly release the model weights. Access is usually provided through a paid API or product, and users cannot inspect or modify the underlying model.
Python AutoMLPredictive AI & ML: The use of automated machine learning tools through Python, the most widely used programming language for data science. Python libraries and interfaces let practitioners automate model building, tuning, and evaluation within their existing code and notebooks.
PytorchDeep Learning: An open source machine learning framework for Python used to build and train neural networks. It is popular in research and production for its flexible approach to defining and debugging models.

Q

2 terms

QuantizationLLM & GenAI: A technique that reduces the numerical precision of a model's weights, for example from 16-bit to 4-bit numbers. It shrinks memory use and speeds up inference, allowing large models to run on smaller or cheaper hardware with modest accuracy loss.
Query Data (Text2SQL)Key termLLM & GenAI: A capability that converts a question written in everyday language into a SQL query, runs it against a database, and returns the result. It lets business users get answers from structured data without writing code.

R

14 terms

RAG EvaluationKey termMLOps & Deployment: The process of measuring how well a retrieval-augmented generation system performs. It assesses both retrieval, whether the right source documents are found, and generation, whether answers are accurate, relevant, and grounded in those sources.
Random ForestPredictive AI & ML: An ensemble machine learning method that builds many decision trees on random subsets of the data and features, then combines their outputs. Averaging or voting across trees improves accuracy and reduces overfitting compared with a single tree.
Recommendation SystemEnterprise AI Strategy: A system that predicts and suggests items a user is likely to want, such as products, content, or next actions, based on past behavior, similar users, and item attributes. Common techniques include collaborative filtering and machine learning.
Recurrent NetworkDeep Learning: A type of neural network designed for sequential data such as text, speech, or time series. It keeps an internal memory of earlier inputs, so each output depends on both the current input and what came before.
RegressionKey termPredictive AI & ML: A supervised machine learning task that predicts a continuous numeric value, such as price, revenue, or temperature, from input variables. It also describes the family of methods used to model relationships between variables.
Regression TreesPredictive AI & ML: Decision trees designed to predict continuous numeric values rather than categories. The data is split into branches based on feature values, and each leaf predicts a number, typically the average of the training examples it contains.
Regularization Parameter (Lambda)Predictive AI & ML: A setting that controls how strongly a model is penalized for complexity during training, typically by shrinking coefficient sizes. Higher values produce simpler models that are less likely to overfit, while lower values allow closer fitting to the training data.
Reinforcement LearningDeep Learning: A type of machine learning in which an agent learns to make decisions by taking actions in an environment and receiving rewards or penalties. Over time it learns strategies that maximize total reward.
Reinforcement Learning from Human Feedback (RLHF)LLM & GenAI: A training method in which people rank or rate a model's responses, and those judgments are used to train a reward signal that guides further learning. It helps align language models with human preferences for helpfulness and safety.
Responsible AIKey termAI Governance & Compliance: An approach to designing, building, and deploying AI that is fair, transparent, accountable, safe, and respectful of privacy. It turns ethical principles into concrete practices, reviews, and controls across the AI life cycle.
Retrieval PipelineLLM & GenAI: The sequence of steps that finds and prepares relevant information for an AI system, typically ingesting documents, splitting them into chunks, creating embeddings, searching an index, and ranking results. It supplies the context used in retrieval-augmented generation.
Retrieval-Augmented Generation (RAG)LLM & GenAI: Retrieval-Augmented Generation (RAG) is an AI architecture that improves large language model outputs by retrieving relevant passages from an external knowledge source and adding them to the prompt before generation.Also: Graph RAG, Agentic RAG
Risk Governance FrameworkAI Governance & Compliance: A structured set of policies, roles, processes, and controls an organization uses to identify, assess, manage, and oversee risk. It defines accountability, risk appetite, and reporting so decisions about risk are consistent and transparent.
RuleFitPredictive AI & ML: An interpretable machine learning method that extracts decision rules from an ensemble of trees and combines them with the original features in a sparse linear model. The result is a set of readable rules with weights.

S

22 terms

SaaSKey termMLOps & Deployment: Software as a service: a delivery model in which software is hosted by a provider and accessed by users over the internet, usually through a subscription. The provider handles infrastructure, maintenance, and updates.
Self-AttentionDeep Learning: An attention mechanism in which each element of a sequence, such as each word in a sentence, is compared with the elements of that same sequence to decide how much each one should shape its new representation. In text generation models, each element looks only at earlier ones. It is the core operation of transformer models.
Semantic SearchLLM & GenAI: A search method that finds results based on meaning and intent rather than exact keyword matches. It typically compares embeddings of the query and documents, so a search for "car repair" can also find "auto maintenance."
Sentiment AnalysisLLM & GenAI: A natural language processing technique that determines the emotional tone of text, such as positive, negative, or neutral. It is commonly used to analyze customer reviews, support tickets, survey responses, and social media posts.
Separator Token ([SEP])LLM & GenAI: A special token that BERT and similar models place at the end of each input segment, marking the boundary between two segments, such as a question and the passage that answers it.Also: SEP Token, Separate Token
Shapley ValuesAI Governance & Compliance: A method from cooperative game theory used to explain machine learning predictions by assigning each input feature a fair share of its contribution to the result. SHAP is a popular approach built on Shapley values.
SLM DevelopmentKey termLLM & GenAI: The process of building, training, or adapting small language models for specific tasks or environments. It often involves distillation, fine-tuning on domain data, and optimization so the model runs efficiently on limited hardware.
Small Language Model (SLM)Key termLLM & GenAI: A language model with far fewer parameters than a large language model, designed to be cheaper and faster to run. Small language models are often tuned for specific tasks and can run on local devices or modest servers.
SOC 2Key termAI Governance & Compliance: An auditing framework developed by the American Institute of Certified Public Accountants that evaluates a service organization's controls for security, availability, processing integrity, confidentiality, and privacy. A SOC 2 report is issued by an independent auditor.
Sovereign AIKey termMLOps & Deployment: AI that an organization owns and controls: its agents, models, data, and results stay under its control, and it can run them in any cloud, on premises, or air-gapped.
Speech-to-TextLLM & GenAI: Technology that converts spoken audio into written text, also called automatic speech recognition. It is used for transcription, voice assistants, call center analytics, and captioning.
SQuADLLM & GenAI: The Stanford Question Answering Dataset, a reading comprehension benchmark of questions about Wikipedia articles where each answer is a span of the passage. Version 2.0 adds questions the passage cannot answer.Also: Stanford Question Answering Dataset
Stack EnsemblePredictive AI & ML: A modeling technique that combines predictions from several different base models by training a second model, called a metalearner, to learn how best to weight them. It often achieves higher accuracy than any single base model.
Stochastic Gradient DescentPredictive AI & ML: A version of gradient descent that updates a model's parameters from one example or a small batch at a time instead of the full dataset. Each update is far cheaper to compute, which is why it and its variants, such as Adam, train most neural networks.Also: SGD
Structured OutputLLM & GenAI: Model output that follows a defined, machine-readable format, such as JSON matching a schema, rather than free-form text. It makes language model responses reliable to parse and easy to pass into other software.
Structured vs Unstructured DataEnterprise AI Strategy: The distinction between data organized in a fixed format, such as rows and columns in a database, and data without a predefined model, such as text, images, audio, and video.
Supervised Machine LearningKey termPredictive AI & ML: An approach to machine learning in which models learn from labeled examples, where each input is paired with the correct output. Classification and regression are the two main types of supervised learning tasks.Also: Supervised Learning
Support Vector MachinePredictive AI & ML: A supervised learning algorithm that classifies data by finding the boundary that separates categories with the widest possible margin. Using kernel functions, it can also handle complex, nonlinear relationships and regression tasks.
System of ContentKey termEnterprise AI Strategy: A platform that stores, manages, and governs an organization's documents and other unstructured content, making it searchable and reusable. The term is used by analogy to a system of record, which manages structured transactional data.
System of IntelligenceKey termEnterprise AI Strategy: A layer of software that sits on top of systems of record and applies analytics and AI to their combined data to generate insights, predictions, and recommended actions. It connects data across applications to support decisions.
System of RecordKey termEnterprise AI Strategy: The authoritative source for a specific type of business data, such as customer records in a CRM or financial transactions in an ERP. Other systems rely on it as the single trusted version of that data.
System PromptKey termLLM & GenAI: A set of instructions given to a language model, separate from user messages, that defines its role, tone, rules, and boundaries for a conversation. It shapes how the model behaves across every response in a session.

T

15 terms

Tabular Foundation ModelKey termPredictive AI & ML: A large pretrained model designed to work with structured data organized in rows and columns, such as spreadsheets and database tables. It can make predictions on new tabular datasets with little or no task-specific training.
Target LeakagePredictive AI & ML: A modeling error that occurs when training data includes information that would not be available at prediction time, often because it is derived from the outcome itself. It makes models look accurate in testing but fail in production.
Target VariablePredictive AI & ML: The outcome a supervised machine learning model is trained to predict, such as whether a customer churns or the price of a house. It is also called the label, response, or dependent variable.
Telemetry & OpsKey termMLOps & Deployment: The collection of operational data, such as metrics, logs, events, and usage, from running systems, combined with the practices used to operate them. Teams use telemetry to monitor health, troubleshoot issues, and manage performance.
Text ClassificationLLM & GenAI: Assigning a label to a piece of text, such as its sentiment, topic, intent, or whether it is spam. It is one of the most common natural language processing tasks and can use classic machine learning or language models.
Time Series DataPredictive AI & ML: Data points collected or recorded in time order, usually at regular intervals, such as daily sales, hourly sensor readings, or monthly website traffic. Its order and time dependencies must be preserved during analysis.
Time Series ModelsKey termPredictive AI & ML: Models designed to analyze data ordered over time and forecast future values based on past observations, trends, and seasonal patterns. Examples include ARIMA, exponential smoothing, and machine learning approaches that use lagged features.
TokenKey termLLM & GenAI: A unit of text that a language model processes, often a word, part of a word, or a punctuation mark. Models measure input limits, output length, and usage costs in tokens.
TokenizationLLM & GenAI: The process of breaking text into smaller units, called tokens, that a language model can process. Each token is mapped to a number, and the method used affects how efficiently a model handles different languages and vocabulary.
TokenomicsKey termLLM & GenAI: In generative AI, the economics of token usage: how the number of input and output tokens drives cost, speed, and capacity when running language models. Managing it involves choosing models, prompt length, and caching to control spend.
Tool BuilderKey termAgentic AI: A capability for creating custom tools that AI agents can call, typically by wrapping an API, database query, or script with a defined name, description, and inputs. It lets teams extend agents with organization-specific actions.
Tracing MatrixKey termMLOps & Deployment: A structured view that maps traces, the recorded steps of individual requests or AI interactions, against dimensions such as components, models, or evaluation criteria. It helps teams locate where errors, delays, or quality issues occur.
Training SetsPredictive AI & ML: The portion of a dataset used to teach a machine learning model by letting it learn patterns and adjust its parameters. It is kept separate from validation and test data used to evaluate performance.
Transfer LearningDeep Learning: A machine learning technique in which a model trained on one task is reused as the starting point for a related task. It reduces the data and computing power needed to build accurate models.
Transformer ArchitectureKey termDeep Learning: A neural network design that uses a mechanism called attention to weigh the relationships between all parts of an input at once. Transformers are the foundation of most modern large language models.Also: Deep Learning Transformer

U

2 terms

UnderfittingPredictive AI & ML: A modeling problem in which a model is too simple to capture the underlying patterns in the data. It performs poorly on both training data and new data, often because of too few features or too little complexity.
Unsupervised Machine LearningKey termPredictive AI & ML: An approach to machine learning in which models find patterns, groupings, or structure in data that has no labeled outcomes. Common tasks include clustering, anomaly detection, and dimensionality reduction.Also: Unsupervised Learning

V

7 terms

Validation SetsPredictive AI & ML: A portion of data held out from training and used to evaluate a model during development, tune hyperparameters, and compare candidate models. It helps detect overfitting before final testing on unseen data.
VectorLLM & GenAI: An ordered list of numbers that represents data in a mathematical space. In AI, vectors such as embeddings capture the features or meaning of text, images, or other data so they can be compared by distance.
Vector DatabaseLLM & GenAI: A database designed to store and search vectors, such as embeddings, by similarity rather than exact match. It lets applications quickly find the most related documents, images, or records and is a core component of retrieval-augmented generation.
Vertical AgentsKey termAgentic AI: AI agents designed for a specific industry or business function, such as healthcare, banking, or insurance claims. They combine domain knowledge, specialized data, and relevant workflows to perform tasks more accurately than general-purpose agents.
Vision ModelingKey termDeep Learning: The process of building machine learning models that interpret images or video. These models perform tasks such as classifying images, detecting and locating objects, segmenting scenes, and extracting information from visual documents.
Vision-Language Model (VLM)LLM & GenAI: A model that takes images and text together as input, so it can read a scanned page, a chart or a photo and answer or extract information from it.
VPCKey termMLOps & Deployment: A virtual private cloud: a logically isolated section of a public cloud where an organization runs resources in its own private network. It controls IP addressing, subnets, routing, and access rules.

W

3 terms

Weights and BiasesDeep Learning: The adjustable parameters of a neural network. Weights set the strength of connections between neurons, and biases shift each neuron's output; both are tuned during training so the network makes accurate predictions.
Word2VecLLM & GenAI: A technique introduced by Google researchers in 2013 that learns vector representations of words from large text collections. Words used in similar contexts end up with similar vectors, capturing relationships such as meaning and analogy.
WordPieceLLM & GenAI: A subword tokenization algorithm, used by BERT, that splits words into frequent pieces, such as "unaffable" into "un", "##aff", and "##able". It keeps the vocabulary to a fixed size while still representing rare or unseen words.

X

1 term

XGBoostPredictive AI & ML: A widely used open source library that implements gradient boosted decision trees, building trees one after another so each corrects the errors of the previous ones. It is known for speed and strong performance on tabular data.

Z

2 terms

Zero-Shot LearningDeep Learning: A machine learning approach in which a model performs a task or recognizes categories it was not explicitly trained on. It relies on general knowledge or descriptions learned during training rather than labeled examples.
Zero-Shot PromptingLLM & GenAI: A prompting technique in which a language model is asked to perform a task with only instructions and no examples. It relies on knowledge the model gained during training to understand and complete the request.

See what H2O AI can do on your data

Request a demo