AI & Machine Learning terms

Study this subject with thinQcards™

• AccuracyThe fraction of all predictions that are correct: (TP + TN) / total.Foundational
• Activation FunctionA nonlinear function applied to a neuron's output that lets networks learn complex patterns.Foundational
• AdamAdaptive Moment Estimation: an optimizer combining momentum with per-parameter adaptive learning rates.Intermediate
• Adversarial ExampleAn input deliberately perturbed, often imperceptibly, to cause a model to make a wrong prediction.Intermediate
• Agent Development Kit (ADK)Google's open-source framework for building, evaluating and deploying multi-agent systems.Advanced
• Agent LoopThe repeated cycle in which an agent calls the model, executes requested tools and feeds results back until the task completes.Intermediate
• Agent MemoryMechanisms that let an agent retain information within a task (short-term) and across sessions (long-term).Intermediate
• Agent SkillsPackaged folders of instructions, scripts and resources that Claude loads on demand to perform specialized tasks.Advanced
• Agent2Agent Protocol (A2A)An open protocol, introduced by Google and now under the Linux Foundation, for communication and task delegation between AI agents.Advanced
• AgentforceSalesforce's platform for building and deploying autonomous AI agents across Salesforce CRM data and workflows.Intermediate
• Agentic AIAI systems that act with autonomy to plan and execute multi-step tasks toward goals, rather than only responding to single prompts.Foundational
• Agentic RAGRAG in which an agent decides when, where and how to retrieve, possibly over multiple iterations and sources.Advanced
• AI AgentAn AI system that uses a model to pursue a goal autonomously by planning, calling tools and acting on results in a loop.Foundational
• AI AlignmentThe effort to ensure AI systems pursue goals and behave in accordance with human intentions and values.Intermediate
• AI GovernanceThe policies, roles, processes and controls that direct and oversee how an organization develops and uses AI.Foundational
• AI Red TeamingAdversarially testing AI systems to discover harmful behaviors, security flaws and safety failures before attackers do.Intermediate
• AI SafetyThe field focused on preventing AI systems from causing accidental or deliberate harm.Foundational
• AI Supply Chain SecurityProtecting the integrity of models, datasets, libraries and services used to build AI systems.Advanced
• Algorithmic BiasSystematic, unfair errors in AI outputs that disadvantage certain groups, often inherited from data.Foundational
• All-ReduceA collective operation that combines values (such as gradients) from all processes and distributes the result back to all.Advanced
• Amazon BedrockAWS's fully managed service for building generative AI applications with foundation models through a single API.Foundational
• Amazon Bedrock AgentCoreAWS services for deploying and operating AI agents securely at scale with any framework or model.Advanced
• Amazon Bedrock GuardrailsConfigurable safeguards for Bedrock and other model applications: content filters, denied topics, PII redaction, grounding checks and more.Intermediate
• Amazon Bedrock Knowledge BasesA managed Bedrock feature that implements RAG: ingestion, chunking, embedding, vector storage and retrieval.Intermediate
• Amazon QAWS's generative AI assistant family: Amazon Q Business for enterprise knowledge and Amazon Q Developer for software development.Foundational
• Amazon SageMaker AIAWS's fully managed service to build, train, tune, deploy and monitor machine learning models.Foundational
• Amazon SageMaker CanvasA no-code SageMaker AI interface for preparing data, building ML models and using generative AI.Intermediate
• Amazon SageMaker ClarifyA SageMaker AI capability for detecting bias in data and models and explaining predictions.Intermediate
• Amazon SageMaker Ground TruthAWS service for creating labeled training datasets using human workforces and automated labeling.Intermediate
• Amazon SageMaker JumpStartA SageMaker AI hub of pretrained and foundation models and solutions that can be deployed and fine-tuned in your account.Intermediate
• Anomaly DetectionIdentifying data points, events or patterns that deviate significantly from expected behavior.Intermediate
• AnthropicAn AI safety and research company that builds the Claude models and created Constitutional AI and the Model Context Protocol.Foundational
• Approximate Nearest Neighbor (ANN)Similarity search that returns near-closest vectors quickly by accepting a small loss in accuracy.Intermediate
• Area Under the Curve (AUC)The area under the ROC curve, summarizing a classifier's ability to rank positives above negatives (0.5 = random, 1.0 = perfect).Intermediate
• Artificial General Intelligence (AGI)Hypothetical AI with human-level ability to understand, learn and apply knowledge across virtually any intellectual task.Foundational
• Artificial Intelligence (AI)The field of building computer systems that perform tasks normally requiring human intelligence, such as reasoning, perception and language.Foundational
• Attention MechanismA technique that lets a model weight the relevance of different input elements when producing each output.Intermediate
• AutoencoderA neural network trained to reconstruct its input through a compressed bottleneck representation.Intermediate
• AutoMLAutomated machine learning: tools that automate algorithm selection, feature preprocessing and hyperparameter tuning.Foundational
• Autoregressive GenerationProducing output one token at a time, with each new token conditioned on all previously generated tokens.Intermediate
• AWS AI ServicesPretrained, task-specific AWS APIs such as Rekognition, Comprehend, Textract, Transcribe, Polly, Translate, Lex, Kendra and Personalize.Foundational
• Azure AI Content SafetyAn Azure service that detects harmful content, prompt attacks and ungrounded output in text and images.Intermediate
• Azure AI SearchAzure's search service supporting keyword, vector, hybrid and semantic ranking, commonly used as the retrieval layer for RAG.Intermediate
• Azure AI ServicesAzure's prebuilt AI APIs for vision, language, speech, translation and document intelligence.Foundational
• Azure Machine LearningAzure's platform for training, deploying and managing ML models with MLOps capabilities.Intermediate
• Azure OpenAIMicrosoft's service providing OpenAI models such as GPT and embedding models with Azure security, networking and compliance.Foundational
• BackpropagationThe algorithm that computes gradients of the loss with respect to every weight by applying the chain rule backward through the network.Intermediate
• BaggingBootstrap aggregating: training models in parallel on random samples with replacement and averaging their outputs.Intermediate
• Base Command Manager (BCM)NVIDIA software for provisioning, managing and monitoring AI clusters.Advanced
• Base ModelA pretrained model before instruction tuning or alignment, which continues text rather than following instructions.Intermediate
• Baseline ModelA simple reference model used to judge whether a more complex model adds value.Foundational
• Batch InferenceGenerating predictions for a large dataset offline in a single job rather than per request.Foundational
• Batch NormalizationNormalizing each layer's inputs across the mini-batch to stabilize and speed up training.Intermediate
• Batch SizeThe number of training examples processed together before the model's parameters are updated.Foundational
• Bayesian OptimizationHyperparameter tuning that builds a probabilistic model of performance to choose the most promising next trial.Advanced
• BenchmarkA standardized dataset and metric used to compare model capabilities.Intermediate
• BERTBidirectional Encoder Representations from Transformers: Google's 2018 encoder-only model pretrained with masked language modeling.Intermediate
• BERTScoreA text-similarity metric that compares contextual embeddings of generated and reference text.Advanced
• BF16Brain Floating Point 16: a 16-bit format with FP32's exponent range but less mantissa precision.Intermediate
• Bias-Variance TradeoffThe balance between error from overly simple assumptions (bias) and error from sensitivity to training data (variance).Intermediate
• BigQuery ML (BQML)A Google Cloud feature that lets users create and run ML models in BigQuery using SQL.Intermediate
• Binary ClassificationClassification with exactly two possible output classes, such as yes/no or positive/negative.Foundational
• BLEUBilingual Evaluation Understudy: a metric scoring generated text by n-gram precision overlap with reference text.Intermediate
• BM25Best Matching 25: a standard lexical ranking function based on term frequency and inverse document frequency.Intermediate
• BoostingAn ensemble method that trains weak learners sequentially, weighting each toward the errors of its predecessors.Intermediate
• Byte-Pair Encoding (BPE)A subword tokenization algorithm that iteratively merges the most frequent adjacent symbol pairs into new tokens.Advanced
• Canary DeploymentReleasing a new model to a small fraction of traffic first, then increasing gradually if metrics are healthy.Intermediate
• Catastrophic ForgettingWhen fine-tuning on new data causes a model to lose previously learned capabilities.Advanced
• Chain-of-Thought Prompting (CoT)Prompting a model to reason step by step before giving a final answer.Intermediate
• ChunkingSplitting documents into smaller passages before embedding them for retrieval.Intermediate
• CitationsReferences in a model's answer that link claims to the specific source passages that support them.Intermediate
• Class ImbalanceA dataset where one class is far more frequent than others, which can bias models toward the majority class.Intermediate
• ClassificationA supervised learning task that predicts which discrete category or class an input belongs to.Foundational
• Classification ThresholdThe probability cutoff above which a classifier predicts the positive class.Intermediate
• ClaudeAnthropic's family of large language models, available in Opus, Sonnet and Haiku tiers.Foundational
• Claude Agent SDKAnthropic's SDK, built on the Claude Code agent harness, for building custom agents in Python or TypeScript.Advanced
• Claude CodeAnthropic's agentic coding tool that works in the terminal, IDEs, desktop and web to read, edit and run code.Intermediate
• Claude Messages APIAnthropic's primary API for sending a conversation of messages to Claude and receiving a response.Intermediate
• CLAUDE.mdA Markdown file of persistent project or user instructions that Claude Code loads into context automatically.Intermediate
• CLIPContrastive Language-Image Pre-training: OpenAI's model that embeds images and text into a shared vector space.Advanced
• ClusteringAn unsupervised task that groups similar data points together so items in a cluster are more alike than items in other clusters.Foundational
• Collaborative FilteringRecommending items based on the preferences of similar users or similarity between items in interaction data.Intermediate
• Computer UseA capability that lets a model operate a computer interface by viewing screenshots and issuing mouse and keyboard actions.Advanced
• Concept DriftA change in the relationship between inputs and the target variable over time.Intermediate
• Confusion MatrixA table comparing predicted classes with actual classes, showing true/false positives and negatives.Foundational
• Constitutional AI (CAI)Anthropic's method for training harmless, helpful models using a written set of principles and AI-generated feedback.Intermediate
• Content FilteringAutomatically detecting and blocking harmful categories of content in AI inputs and outputs.Foundational
• Context CompactionSummarizing or pruning earlier conversation content so a long-running agent stays within its context window.Advanced
• Context EngineeringDesigning and managing everything placed in a model's context window to maximize task performance.Advanced
• Context WindowThe maximum number of tokens (input plus output) a model can consider in a single request.Foundational
• Contextual RetrievalPrepending chunk-specific context, generated from the whole document, to each chunk before embedding and indexing.Advanced
• Continued PretrainingFurther self-supervised training of a foundation model on large unlabeled domain data.Advanced
• Continuous BatchingAn LLM serving technique that adds and removes requests from the running batch at each generation step.Advanced
• Continuous Training (CT)Automatically retraining and redeploying models on new data on a schedule or when drift is detected.Intermediate
• Convolutional Neural Network (CNN)A neural network that uses convolutional filters to detect spatial patterns, primarily for images.Intermediate
• Cosine SimilarityA similarity measure based on the angle between two vectors, ranging from -1 to 1.Intermediate
• Cross-EncoderA model that scores a query and document together in one pass, giving accurate relevance at higher cost.Advanced
• Cross-Entropy LossA loss that measures the difference between predicted class probabilities and the true labels; standard for classification and LLMs.Intermediate
• Cross-ValidationA resampling method that trains and evaluates a model on multiple different splits to estimate generalization more reliably.Intermediate
• CUDACompute Unified Device Architecture: NVIDIA's parallel computing platform and programming model for GPUs.Foundational
• cuDFA RAPIDS GPU DataFrame library with a pandas-like API for loading, joining, aggregating and filtering data.Intermediate
• Curse of DimensionalityThe set of problems that arise as feature count grows: data becomes sparse and distance measures lose meaning.Intermediate
• Data AugmentationExpanding a training set by creating modified copies of existing examples.Intermediate
• Data Center GPU Manager (DCGM)NVIDIA tools for monitoring, health-checking and diagnosing GPUs in clusters.Advanced
• Data DriftA change in the statistical distribution of input data in production compared with training data.Intermediate
• Data LabelingThe process of annotating raw data with labels so it can be used for supervised training or evaluation.Foundational
• Data LeakageWhen information unavailable at prediction time leaks into training, producing unrealistically good evaluation results.Intermediate
• Data LineageThe record of where data originated and how it was transformed on its way to a model or report.Intermediate
• Data Masking for AIRedacting or replacing sensitive data such as PII before it reaches an AI model or appears in outputs.Intermediate
• Data ParallelismDistributed training where each GPU holds a full model copy and processes a different slice of each batch.Intermediate
• Data PoisoningAn attack that corrupts training, fine-tuning or retrieval data to manipulate model behavior or implant backdoors.Intermediate
• Data Processing Unit (DPU)A programmable processor that offloads networking, storage and security tasks from the host CPU.Intermediate
• Databricks Mosaic AIDatabricks capabilities for building, evaluating, serving and governing ML and generative AI on the lakehouse.Intermediate
• Decision TreeA model that predicts by learning a hierarchy of if-then splits on feature values.Foundational
• Decoder-Only ArchitectureA transformer that uses only the decoder stack with causal (masked) attention to generate text one token at a time.Intermediate
• Decoding StrategyThe method used to choose each next token from the model's probability distribution.Intermediate
• Deep Learning (DL)A subset of machine learning that uses multi-layer neural networks to learn hierarchical representations directly from raw data.Foundational
• Differential PrivacyA mathematical privacy guarantee that limits how much any individual's data can affect a model's or query's output.Advanced
• Diffusion ModelA generative model that learns to create data by reversing a gradual noising process.Intermediate
• Dimensionality ReductionTechniques that reduce the number of input features while preserving as much useful information as possible.Intermediate
• Direct Preference Optimization (DPO)A preference-tuning method that optimizes a model directly on preferred vs. rejected responses without a separate reward model.Advanced
• Discriminative ModelA model that learns the boundary between classes, estimating P(label | input).Intermediate
• Dot ProductThe sum of element-wise products of two vectors, used as a similarity score in retrieval and attention.Intermediate
• DropoutA regularization technique that randomly deactivates a fraction of neurons during each training step.Intermediate
• Early StoppingHalting training when validation performance stops improving to prevent overfitting.Intermediate
• Edge AIRunning AI inference on devices near the data source, such as cameras, robots or phones, rather than in the cloud.Intermediate
• Einstein Trust LayerSalesforce's security architecture that protects data and responses in generative AI features.Intermediate
• EmbeddingA dense numeric vector representing the meaning of text, images or other data so that similar items are close together.Foundational
• Embedding ModelA model specialized in converting inputs into embedding vectors for search and similarity tasks.Intermediate
• Encoder-Decoder ArchitectureA transformer with an encoder that reads the input and a decoder that generates output while attending to the encoder.Intermediate
• Encoder-Only ArchitectureA transformer that uses only the encoder stack with bidirectional attention, suited to understanding tasks.Intermediate
• Ensemble LearningCombining predictions from multiple models to achieve better performance than any single model.Intermediate
• EpochOne complete pass of the training algorithm through the entire training dataset.Foundational
• EU AI ActThe European Union's risk-based AI regulation classifying systems as prohibited, high-risk, limited-risk or minimal-risk.Intermediate
• Evaluation HarnessA framework that runs a model or application over test cases and computes metrics automatically and repeatably.Advanced
• Excessive AgencyAn LLM risk where an agent has more functionality, permissions or autonomy than needed, enabling damaging actions.Intermediate
• Experiment TrackingRecording the code, data, parameters, metrics and artifacts of each training run for comparison and reproducibility.Intermediate
• ExplainabilityThe ability to describe why an AI model produced a particular output in terms humans can understand.Foundational
• Exploding GradientA training problem where gradients grow uncontrollably large, causing unstable updates and NaN losses.Intermediate
• Extended ThinkingA Claude feature that lets the model produce visible step-by-step reasoning before its final answer, within a token budget.Intermediate
• F1 ScoreThe harmonic mean of precision and recall: 2PR / (P + R).Foundational
• FairnessThe principle that AI systems should not produce unjustly different outcomes for individuals or groups.Foundational
• FAISSFacebook AI Similarity Search: Meta's open-source library for efficient vector similarity search, with GPU support.Advanced
• FeatureAn individual measurable input variable that a model uses to make predictions.Foundational
• Feature EngineeringCreating, transforming and selecting input variables to improve model performance.Foundational
• Feature ScalingTransforming numeric features to a common range or distribution so no feature dominates due to its units.Foundational
• Feature SelectionChoosing a subset of the most relevant features and discarding redundant or irrelevant ones.Intermediate
• Feature StoreA centralized repository for storing, sharing and serving ML features consistently for training and inference.Intermediate
• Federated LearningTraining a shared model across decentralized devices or sites without moving raw data to a central server.Intermediate
• Few-Shot PromptingIncluding several examples of inputs and desired outputs in the prompt to guide the model.Foundational
• Fine-TuningFurther training a pretrained model on a smaller task- or domain-specific dataset to adapt its behavior.Intermediate
• FlashAttentionAn IO-aware exact attention algorithm that reduces GPU memory traffic to speed up attention.Advanced
• Foundation Model (FM)A large model pretrained on broad data that can be adapted to many downstream tasks.Foundational
• FP1616-bit half-precision floating point, halving memory versus FP32 with a narrower numeric range.Intermediate
• FP3232-bit single-precision floating point, the traditional default for neural network math.Intermediate
• FP4A 4-bit floating-point format accelerated by NVIDIA Blackwell Tensor Cores for high-throughput inference.Advanced
• FP8An 8-bit floating-point format supported by NVIDIA Hopper and later Tensor Cores for faster training and inference.Advanced
• Frontier ModelA state-of-the-art, highly capable general-purpose AI model at the leading edge of capability.Intermediate
• Full Fine-TuningUpdating all of a model's weights during fine-tuning.Intermediate
• Fully Sharded Data Parallel (FSDP)A PyTorch training strategy that shards parameters, gradients and optimizer states across data-parallel GPUs.Advanced
• Function CallingAn API feature where a model outputs a structured request to call a developer-defined function with arguments.Foundational
• GB200 NVL72A liquid-cooled rack-scale NVIDIA system linking 36 Grace CPUs and 72 Blackwell GPUs in a single NVLink domain.Advanced
• GeminiGoogle's family of natively multimodal foundation models, available in the Gemini app, Gemini API and Vertex AI.Foundational
• Generative Adversarial Network (GAN)A generative model in which a generator and a discriminator are trained against each other.Intermediate
• Generative AI (GenAI)AI that creates new content such as text, images, code, audio or video based on patterns learned from training data.Foundational
• Generative Pre-trained Transformer (GPT)OpenAI's family of decoder-only transformer models pretrained on large text corpora for generation.Foundational
• Golden DatasetA curated set of inputs with verified expected outputs used as the reference for evaluation and regression testing.Intermediate
• GPUDirect RDMANVIDIA technology that lets network adapters read and write GPU memory directly, bypassing host memory and CPU.Advanced
• Gradient AccumulationSumming gradients over several mini-batches before updating weights to simulate a larger batch.Advanced
• Gradient BoostingAn ensemble technique that builds models sequentially, each one correcting the residual errors of the previous ones.Intermediate
• Gradient DescentAn optimization algorithm that iteratively adjusts parameters in the direction that most reduces the loss.Intermediate
• Graphics Processing Unit (GPU)A massively parallel processor, originally for graphics, that accelerates the matrix math behind AI training and inference.Foundational
• GraphRAGRAG that uses a knowledge graph of entities and relationships to retrieve connected information.Advanced
• Grid SearchHyperparameter tuning that exhaustively tries every combination from predefined value lists.Intermediate
• Ground TruthThe verified correct answer or label used to train and evaluate models.Foundational
• GroundednessAn evaluation measure of whether a response's claims are supported by the provided context.Intermediate
• Grounding (Generative AI)Anchoring a model's responses to verifiable source data supplied at runtime.Foundational
• GuardrailsControls that constrain AI inputs and outputs to keep behavior safe, on-topic and compliant.Foundational
• HallucinationWhen a generative model produces fluent output that is false, fabricated or unsupported by its sources.Foundational
• High Bandwidth Memory (HBM)Stacked DRAM packaged alongside the GPU that provides very high memory bandwidth.Intermediate
• HNSWHierarchical Navigable Small World: a graph-based ANN index offering fast, high-recall vector search.Advanced
• HooksUser-defined commands that run automatically at points in an agent's lifecycle, such as before or after a tool call.Advanced
• Hugging FaceA platform and open-source community hosting models, datasets and libraries such as Transformers.Foundational
• Human EvaluationAssessment of model outputs by human raters, often treated as the gold standard for quality.Foundational
• Human OversightEnsuring humans can monitor, understand, intervene in and override AI system behavior.Foundational
• Human-in-the-Loop (HITL)Design in which humans review, approve or correct AI outputs or actions at defined points.Foundational
• HumanEvalA code-generation benchmark of programming problems scored by whether generated code passes unit tests.Intermediate
• Hybrid SearchCombining keyword (lexical) and vector (semantic) search results for better retrieval.Intermediate
• HyperparameterA configuration value set before training that controls the learning process, rather than being learned from data.Foundational
• Hyperparameter TuningSearching for the hyperparameter values that give the best validation performance.Intermediate
• Image ClassificationA computer vision task that assigns a label to an entire image.Foundational
• Image SegmentationA computer vision task that labels each pixel of an image with a class or object instance.Intermediate
• ImputationFilling in missing values in a dataset with estimated substitutes.Foundational
• In-Context Learning (ICL)A model's ability to learn a task from instructions and examples in the prompt without weight updates.Intermediate
• Indirect Prompt InjectionPrompt injection delivered through external content the model processes, such as a web page, email, file or tool output.Intermediate
• InferenceUsing a trained model to make predictions or generate outputs on new data.Foundational
• Inference ParametersSettings supplied at request time that shape model output, such as temperature, top-p, max tokens and stop sequences.Foundational
• Inference ThroughputThe amount of work an inference system completes per unit time, such as tokens or requests per second.Intermediate
• InfiniBandA high-throughput, low-latency network fabric with native RDMA, widely used to connect GPU clusters.Intermediate
• Instruction TuningSupervised fine-tuning on many instruction-response examples so a model follows natural-language instructions.Intermediate
• ISO/IEC 42001The international standard for an AI management system (AIMS), certifiable like ISO/IEC 27001.Intermediate
• JailbreakA technique that manipulates a model into bypassing its safety training to produce prohibited content.Intermediate
• JSON SchemaA standard vocabulary for describing the structure and validation rules of JSON data.Intermediate
• K-Means ClusteringAn unsupervised algorithm that partitions data into k clusters by assigning points to the nearest centroid and iteratively updating centroids.Foundational
• K-Nearest Neighbors (k-NN)A non-parametric algorithm that predicts by majority vote or average of the k closest training examples.Intermediate
• KerasA high-level deep learning API for defining and training neural networks; Keras 3 runs on TensorFlow, JAX or PyTorch.Intermediate
• Keyword SearchLexical search that matches documents containing the query's exact terms, typically ranked with BM25.Foundational
• Knowledge CutoffThe date after which a model has no information from its training data.Foundational
• Knowledge DistillationTraining a smaller student model to reproduce the outputs or behavior of a larger teacher model.Intermediate
• Knowledge GraphA structured representation of entities and the relationships between them.Intermediate
• Kubernetes GPU OperatorAn NVIDIA operator that automates deployment of drivers, container toolkit, device plugin and monitoring for GPUs in Kubernetes.Advanced
• KV CacheStored attention keys and values for previous tokens, reused so each new token does not recompute them.Advanced
• L1 RegularizationAdds a penalty proportional to the absolute value of weights, pushing some weights to exactly zero (Lasso).Intermediate
• L2 RegularizationAdds a penalty proportional to the squared weights, shrinking all weights toward zero (Ridge, weight decay).Intermediate
• LabelThe known target value attached to a training example that a supervised model learns to predict.Foundational
• Large Language Model (LLM)A very large transformer model trained on massive text data to understand and generate natural language and code.Foundational
• Latent SpaceA compressed, learned representation space where similar inputs map to nearby points.Intermediate
• Layer NormalizationNormalizing activations across the features of each individual example rather than across a batch.Advanced
• Learning RateThe hyperparameter that sets how large a step the optimizer takes when updating weights.Foundational
• Linear RegressionA regression algorithm that models the target as a weighted sum of input features plus an intercept.Foundational
• LLM-as-a-JudgeUsing a language model to grade or compare outputs of another model against a rubric.Intermediate
• LLMOpsOperational practices for deploying and managing large language model applications in production.Intermediate
• Logistic RegressionA classification algorithm that applies the sigmoid function to a linear combination of features to output a probability.Foundational
• Long Short-Term Memory (LSTM)An RNN variant with gates that control what information to keep, forget and output, enabling long-range memory.Intermediate
• Loss FunctionA function that measures how far a model's predictions are from the true values; training minimizes it.Intermediate
• Lost in the MiddleThe tendency of some LLMs to use information at the start and end of a long context better than information in the middle.Advanced
• Low-Rank Adaptation (LoRA)A PEFT method that freezes base weights and trains small low-rank matrices added to selected layers.Intermediate
• Machine Learning (ML)A subset of AI in which systems learn patterns from data to make predictions or decisions without being explicitly programmed with rules.Foundational
• Machine Learning Operations (MLOps)Practices and tooling that apply DevOps principles to build, deploy, monitor and maintain ML models reliably in production.Foundational
• Machine Translation (MT)Automatically translating text or speech from one language to another.Foundational
• Masked Language Modeling (MLM)A pretraining objective in which random tokens are hidden and the model predicts them from surrounding context.Advanced
• Max TokensAn inference parameter limiting the number of tokens the model may generate in a response.Foundational
• MCP ClientThe component inside an AI host application that maintains a connection to an MCP server.Intermediate
• MCP Resources and PromptsMCP primitives besides tools: resources expose readable data, prompts expose reusable templates.Advanced
• MCP ServerA program that exposes tools, resources and prompts to AI applications via the Model Context Protocol.Intermediate
• Mean Absolute Error (MAE)The average of absolute differences between predicted and actual values.Foundational
• Mean Squared Error (MSE)The average of squared differences between predicted and actual values.Foundational
• Message Batches APIAn asynchronous API for submitting large volumes of Claude requests at a discount, processed within 24 hours.Intermediate
• Metadata FilteringRestricting vector search to documents whose attributes match conditions such as date, source or access level.Intermediate
• Microsoft FoundryMicrosoft's unified Azure platform for building AI apps and agents with models, tools and governance (formerly Azure AI Foundry).Foundational
• MITRE ATLASAdversarial Threat Landscape for Artificial-Intelligence Systems: a knowledge base of adversary tactics and techniques against AI.Intermediate
• Mixed Precision TrainingTraining with lower-precision number formats (FP16/BF16/FP8) for most operations while keeping critical values in FP32.Intermediate
• Mixture of Experts (MoE)An architecture where a router activates only a few expert sub-networks per token, increasing capacity without proportional compute.Advanced
• ML PipelineAn automated, repeatable sequence of steps such as data preparation, training, evaluation and deployment.Intermediate
• MLflowAn open-source platform for experiment tracking, model packaging, model registry and evaluation.Intermediate
• Model CardA document describing a model's intended use, performance, limitations, training data and ethical considerations.Foundational
• Model Context Protocol (MCP)An open protocol, introduced by Anthropic, that standardizes how AI applications connect to tools, data sources and prompts.Intermediate
• Model EvaluationMeasuring a trained model's performance, robustness and fitness for purpose before and after deployment.Foundational
• Model Inversion and ExtractionAttacks that reconstruct sensitive training data from a model or steal a model's functionality via queries.Advanced
• Model MonitoringContinuously tracking a deployed model's inputs, outputs, performance and fairness to detect degradation.Intermediate
• Model ParametersThe internal values, such as weights and biases, that a model learns from training data.Foundational
• Model RegistryA central catalog that stores model versions with metadata, lineage and approval status for deployment.Intermediate
• Model ServingHosting a trained model behind an interface so applications can request predictions.Foundational
• Model TrainingThe process of fitting a model's parameters to data by optimizing an objective function.Foundational
• Multi-Agent System (MAS)An architecture in which multiple specialized agents collaborate, often coordinated by an orchestrator.Intermediate
• Multi-Head Attention (MHA)Running several attention operations in parallel so the model can attend to different relationships at once.Advanced
• Multi-Instance GPU (MIG)An NVIDIA feature that partitions one GPU into up to seven isolated instances with dedicated compute and memory.Intermediate
• Multiclass ClassificationClassification where each input is assigned to exactly one of three or more classes.Intermediate
• Multilayer Perceptron (MLP)A feedforward neural network with one or more hidden layers of fully connected neurons.Intermediate
• Multimodal EmbeddingAn embedding that places different modalities, such as images and text, in the same vector space.Intermediate
• Multimodal ModelA model that can process or generate more than one data type, such as text, images, audio and video.Foundational
• Naive BayesA probabilistic classifier based on Bayes' theorem that assumes features are independent given the class.Intermediate
• Named Entity Recognition (NER)An NLP task that identifies and classifies entities such as people, organizations, locations and dates in text.Foundational
• Narrow AI (ANI)AI designed to perform a specific task or narrow set of tasks, as opposed to general human-level intelligence.Foundational
• Natural Language Processing (NLP)The field of AI that enables computers to understand, interpret and generate human language.Foundational
• NCCLNVIDIA Collective Communications Library: optimized multi-GPU and multi-node collectives such as all-reduce.Advanced
• NeMo GuardrailsNVIDIA's open-source toolkit for adding programmable safety, topic and security rails to LLM applications.Advanced
• Neural Network (NN)A model made of layers of interconnected artificial neurons that learn weighted transformations of input data.Foundational
• Next-Token PredictionThe training objective of predicting the next token in a sequence given the preceding tokens.Intermediate
• NIST AI Risk Management Framework (AI RMF)A voluntary U.S. NIST framework for managing AI risks, organized into Govern, Map, Measure and Manage functions.Intermediate
• NormalizationMin-max scaling that rescales a feature to a fixed range, usually 0 to 1.Foundational
• NVIDIA AI Enterprise (NVAIE)NVIDIA's licensed, supported software suite for developing and deploying production AI, including NIM, NeMo and Triton.Intermediate
• NVIDIA BlackwellNVIDIA's GPU architecture following Hopper, used in B200, B300 and GB200/GB300 systems, adding FP4 and fifth-generation NVLink.Intermediate
• NVIDIA CosmosNVIDIA's platform of world foundation models for generating and understanding physical-world video and data.Advanced
• NVIDIA DGXNVIDIA's integrated AI systems combining GPUs, NVLink, networking and an optimized software stack.Intermediate
• NVIDIA DGX SuperPODNVIDIA's reference architecture for scalable AI supercomputers built from DGX systems, InfiniBand or Ethernet fabrics and storage.Advanced
• NVIDIA DynamoNVIDIA's open-source distributed inference framework for serving reasoning and generative models across many GPUs.Advanced
• NVIDIA HopperNVIDIA's GPU architecture behind the H100 and H200, introducing FP8 Transformer Engine and fourth-generation NVLink.Intermediate
• NVIDIA NeMoNVIDIA's framework and microservices for building, customizing, evaluating and guarding generative AI models and agents.Intermediate
• NVIDIA NGCNVIDIA's catalog of GPU-optimized containers, pretrained models, Helm charts and SDKs.Intermediate
• NVIDIA NIMNVIDIA Inference Microservices: prebuilt, optimized containers that serve AI models through industry-standard APIs.Intermediate
• NVIDIA OmniverseNVIDIA's platform for building and operating physically accurate 3D simulations and digital twins based on OpenUSD.Advanced
• nvidia-smiNVIDIA System Management Interface: a command-line tool to query and manage GPU state.Intermediate
• NVLinkNVIDIA's high-bandwidth, low-latency GPU-to-GPU interconnect.Intermediate
• NVSwitchAn NVIDIA switch chip that connects multiple GPUs over NVLink for all-to-all full-bandwidth communication.Advanced
• Object DetectionA computer vision task that identifies and locates multiple objects in an image with bounding boxes.Intermediate
• Observability for LLM ApplicationsTracing, logging and measuring prompts, responses, tool calls, tokens, latency and quality in LLM systems.Advanced
• One-Hot EncodingRepresenting a categorical variable as multiple binary columns, one per category.Foundational
• ONNXOpen Neural Network Exchange: an open format for representing models so they can move between frameworks and runtimes.Intermediate
• Open-Weight ModelA model whose trained weights are publicly downloadable for self-hosting and customization.Intermediate
• OpenUSDUniversal Scene Description: an open, extensible framework for describing, composing and collaborating on 3D scenes.Advanced
• Optical Character Recognition (OCR)Technology that extracts machine-readable text from images or scanned documents.Foundational
• OptimizerThe algorithm that updates model weights using computed gradients, such as SGD, Adam or AdamW.Intermediate
• Orchestrator-Worker PatternAn agent pattern where a lead model breaks a task into subtasks, delegates them to workers and synthesizes results.Advanced
• OverfittingWhen a model learns training data too closely, including noise, and performs poorly on new data.Foundational
• OWASP Top 10 for LLM ApplicationsOWASP's list of the most critical security risks for applications built on large language models.Intermediate
• PagedAttentionA KV-cache memory technique, introduced by vLLM, that stores cache in fixed-size blocks like virtual memory pages.Advanced
• Parameter-Efficient Fine-Tuning (PEFT)Fine-tuning methods that update only a small set of added or selected parameters while freezing the base model.Intermediate
• PerplexityA language-model metric measuring how well a model predicts text; lower perplexity means better prediction.Intermediate
• Pipeline ParallelismSplitting a model's layers into sequential stages placed on different GPUs, with micro-batches flowing through.Advanced
• PolicyIn reinforcement learning, the strategy that maps states to actions.Intermediate
• Positional EncodingInformation added to token embeddings so a transformer knows the order of tokens.Advanced
• PrecisionOf all positive predictions, the fraction that are actually positive: TP / (TP + FP).Foundational
• Prefill and DecodeThe two phases of LLM inference: processing the whole prompt in parallel (prefill), then generating tokens sequentially (decode).Advanced
• PretrainingThe initial, compute-intensive training of a model on vast unlabeled data to learn general representations.Intermediate
• Principal Component Analysis (PCA)A linear technique that projects data onto orthogonal directions of maximum variance to reduce dimensionality.Intermediate
• PromptThe input text (and optionally images or files) given to a generative model to direct its output.Foundational
• Prompt CachingReusing the processed state of a repeated prompt prefix across requests to cut latency and cost.Intermediate
• Prompt ChainingBreaking a complex task into a sequence of prompts where each output feeds the next.Intermediate
• Prompt EngineeringDesigning and refining model inputs to reliably elicit accurate, useful and safe outputs.Foundational
• Prompt InjectionAn attack in which crafted input causes an LLM to ignore its instructions or perform unintended actions.Foundational
• Prompt TemplateA reusable prompt with placeholders that are filled with variables at runtime.Foundational
• Prompt TuningA PEFT method that learns continuous 'soft prompt' embeddings prepended to inputs while the model stays frozen.Advanced
• Provisioned ThroughputReserved model inference capacity purchased for consistent performance, rather than paying per token on demand.Intermediate
• PruningRemoving less important weights, neurons or layers from a model to make it smaller and faster.Advanced
• PyTorchAn open-source deep learning framework, originally from Meta and now governed by the PyTorch Foundation, known for dynamic computation graphs.Foundational
• QLoRALoRA fine-tuning on top of a base model quantized to 4-bit precision, sharply reducing memory.Advanced
• QuantizationReducing the numeric precision of model weights (and sometimes activations) to shrink memory and speed inference.Intermediate
• Query RewritingReformulating a user's query, often with an LLM, to improve retrieval.Advanced
• R-Squared (R²)The coefficient of determination: the proportion of variance in the target explained by the model.Intermediate
• RAG EvaluationMeasuring a RAG system's retrieval quality and answer quality, typically with faithfulness, relevance and context metrics.Advanced
• Rail-Optimized TopologyA GPU cluster network design where each GPU index in every server connects to its own dedicated leaf switch (rail).Advanced
• Random ForestAn ensemble of decision trees trained on random data samples and feature subsets, averaging their predictions.Intermediate
• RAPIDSNVIDIA's suite of open-source GPU-accelerated data science libraries with familiar Python APIs.Intermediate
• ReActReason + Act: a prompting pattern where a model interleaves reasoning steps with tool actions and observations.Intermediate
• Real-Time InferenceServing predictions synchronously with low latency in response to individual requests.Foundational
• Reasoning ModelAn LLM trained to generate intermediate reasoning before answering, improving performance on complex multi-step tasks.Intermediate
• RecallOf all actual positives, the fraction the model correctly identifies: TP / (TP + FN). Also called sensitivity or true positive rate.Foundational
• Recommendation SystemA system that predicts items a user is likely to want, based on behavior, item attributes or both.Intermediate
• Recurrent Neural Network (RNN)A neural network that processes sequences by passing a hidden state from one step to the next.Intermediate
• RegressionA supervised learning task that predicts a continuous numeric value.Foundational
• RegularizationTechniques that constrain model complexity to reduce overfitting.Intermediate
• Reinforcement Learning (RL)Learning in which an agent takes actions in an environment and improves its policy by maximizing cumulative reward.Foundational
• Reinforcement Learning from Human Feedback (RLHF)Aligning a model by training a reward model on human preference rankings and optimizing the model against it.Intermediate
• ReLURectified Linear Unit: an activation that outputs the input if positive and zero otherwise, f(x)=max(0,x).Foundational
• RerankingRe-scoring an initial set of retrieved results with a more accurate model to put the most relevant first.Intermediate
• Residual ConnectionA skip connection that adds a layer's input to its output, easing gradient flow in deep networks.Advanced
• Responsible AIDeveloping and using AI in ways that are fair, reliable, safe, private, secure, transparent, accountable and inclusive.Foundational
• Responsible Scaling Policy (RSP)Anthropic's framework that ties increasingly strict safety and security measures to AI Safety Levels (ASLs) based on model capability.Advanced
• Retrieval-Augmented Generation (RAG)Retrieving relevant information from external sources at query time and adding it to the prompt so the model answers from it.Foundational
• Reward FunctionThe signal that tells a reinforcement learning agent how good an action or outcome was.Intermediate
• Reward HackingWhen an AI system exploits flaws in its reward signal to score highly without doing the intended task.Advanced
• Reward ModelA model trained to score outputs according to human or AI preferences, used to guide RL fine-tuning.Advanced
• ROC CurveReceiver Operating Characteristic curve: a plot of true positive rate against false positive rate across thresholds.Intermediate
• Role PromptingAssigning the model a persona or expertise in the prompt to shape its tone and focus.Foundational
• Root Mean Squared Error (RMSE)The square root of MSE, expressing typical prediction error in the target's own units.Foundational
• ROUGERecall-Oriented Understudy for Gisting Evaluation: n-gram and sequence overlap metrics for summarization.Intermediate
• RoutingClassifying an input and directing it to a specialized prompt, tool or model.Intermediate
• Scaling LawsEmpirical relationships showing model loss improves predictably with more parameters, data and compute.Advanced
• Self-AttentionAttention in which each token in a sequence attends to every other token in the same sequence.Intermediate
• Self-Supervised Learning (SSL)Learning in which the training signal is generated automatically from the data itself, without human labels.Intermediate
• Semantic SearchSearch that retrieves results by meaning, using embeddings, rather than by exact keyword match.Foundational
• Semi-Supervised LearningLearning that combines a small labeled dataset with a large unlabeled dataset to improve model accuracy.Intermediate
• Sensitive Information DisclosureAn LLM application risk where the model reveals confidential data, PII or secrets in its outputs.Intermediate
• Sentiment AnalysisAn NLP task that determines whether text expresses positive, negative or neutral opinion.Foundational
• Shadow AIUse of AI tools by employees without the organization's approval or oversight.Foundational
• Shadow DeploymentRunning a new model alongside production on live traffic without returning its predictions to users.Intermediate
• SHAPSHapley Additive exPlanations: a method that attributes a prediction to each feature using game-theoretic Shapley values.Intermediate
• SHARPScalable Hierarchical Aggregation and Reduction Protocol: NVIDIA in-network computing that performs collectives inside switches.Advanced
• Sigmoid FunctionAn S-shaped activation that maps any real number to the range 0 to 1.Foundational
• SlurmAn open-source workload manager and job scheduler widely used for HPC and large AI training clusters.Intermediate
• Small Language Model (SLM)A language model with relatively few parameters (typically under ~15B) designed for efficiency, edge or cost-sensitive use.Intermediate
• Snowflake Cortex AISnowflake's built-in AI features for running LLM functions, search and agents on data inside Snowflake.Intermediate
• SoftmaxA function that converts a vector of raw scores (logits) into a probability distribution summing to 1.Intermediate
• Spectrum-XNVIDIA's Ethernet networking platform for AI, pairing Spectrum-4 switches with BlueField-3 SuperNICs.Advanced
• Speculative DecodingSpeeding up generation by having a small draft model propose tokens that the large model verifies in parallel.Advanced
• Speech-to-Text (STT)Converting spoken audio into written text; also called automatic speech recognition (ASR).Foundational
• Stable DiffusionA family of open-weight latent diffusion models from Stability AI for text-to-image generation.Intermediate
• StandardizationScaling a feature to have mean 0 and standard deviation 1 (z-score scaling).Foundational
• Stochastic Gradient Descent (SGD)Gradient descent that updates parameters using one example or a small mini-batch at a time.Intermediate
• Stop ReasonA field in an LLM API response indicating why generation stopped, such as end_turn, max_tokens or tool_use.Intermediate
• StreamingReturning a model's output incrementally as tokens are generated instead of waiting for the full response.Intermediate
• Structured OutputConstraining model responses to a defined format, typically valid JSON matching a schema.Intermediate
• SubagentA separate agent instance, with its own context window, prompt and tools, to which a main agent delegates a task.Advanced
• SummarizationCondensing a document into a shorter version that preserves key information.Foundational
• Supervised Fine-Tuning (SFT)Fine-tuning a model on labeled input-output pairs, such as instruction-response examples.Intermediate
• Supervised LearningMachine learning that trains a model on labeled examples, where each input is paired with the correct output.Foundational
• Support Vector Machine (SVM)A supervised algorithm that finds the hyperplane separating classes with the maximum margin.Intermediate
• SWE-benchA benchmark that measures whether models can resolve real GitHub issues by producing patches that pass the project's tests.Advanced
• Synthetic DataArtificially generated data used to train, fine-tune or evaluate models.Intermediate
• System PromptInstructions supplied separately from user messages that set a model's role, rules, tone and constraints for a conversation.Foundational
• System Prompt LeakageThe exposure of a system prompt's contents, potentially revealing rules, logic or embedded secrets.Intermediate
• TemperatureAn inference parameter that scales randomness in token selection; low values are more deterministic, high values more creative.Foundational
• TensorA multi-dimensional array of numbers, the core data structure in deep learning frameworks.Foundational
• Tensor CoreSpecialized NVIDIA GPU units that accelerate matrix multiply-accumulate operations used in deep learning.Intermediate
• Tensor ParallelismSplitting individual layers' weight matrices across GPUs so each computes part of every operation.Advanced
• Tensor Processing Unit (TPU)Google's custom ASIC for accelerating neural network training and inference, available on Google Cloud.Intermediate
• TensorFlowAn open-source machine learning framework from Google, with Keras as its high-level API.Foundational
• TensorRTNVIDIA's SDK for optimizing trained models into high-performance inference engines.Intermediate
• TensorRT-LLMNVIDIA's open-source library for optimized LLM inference on NVIDIA GPUs.Advanced
• Test DatasetData kept completely separate until the end to measure a final model's generalization performance.Foundational
• Test-Time ComputeAdditional computation spent during inference, such as longer reasoning or multiple samples, to improve answer quality.Advanced
• Text-to-ImageGenerating images from natural-language descriptions.Foundational
• Text-to-Speech (TTS)Generating natural-sounding spoken audio from text.Foundational
• Time Series ForecastingPredicting future values of a variable from its historical, time-ordered observations.Intermediate
• Time to First Token (TTFT)The latency between sending a request and receiving the first generated token.Intermediate
• TokenThe basic unit of text an LLM processes, typically a word, subword, character or punctuation mark.Foundational
• Token PricingCharging for LLM usage per input and output token, often with separate rates for cached and batched tokens.Foundational
• Tokenization (NLP)Splitting text into tokens that a model can convert to numeric IDs.Foundational
• TokenizerThe component that converts text to token IDs and back, defined by its vocabulary and splitting rules.Intermediate
• Tool DefinitionThe name, description and JSON input schema that tell a model what a tool does and how to call it.Intermediate
• Tool PoisoningAn attack that hides malicious instructions in tool or MCP server descriptions or outputs to manipulate an agent.Advanced
• Tool UseA model's ability to request execution of external functions, APIs or programs and use their results.Foundational
• Top-k SamplingSampling the next token only from the k most probable candidates.Intermediate
• Top-p SamplingNucleus sampling: choose the next token from the smallest set whose cumulative probability exceeds p.Intermediate
• Train-Test SplitDividing a dataset into separate subsets for training and evaluation.Foundational
• Training DatasetThe portion of data used to fit a model's parameters.Foundational
• Transfer LearningReusing a model trained on one task as the starting point for a related task.Intermediate
• TransformerA neural network architecture based on self-attention that processes all tokens in a sequence in parallel; the basis of modern LLMs.Foundational
• Transformer EngineNVIDIA hardware and software that dynamically manages FP8/FP4 precision to accelerate transformer models.Advanced
• TransparencyOpenness about how an AI system works, its data, limitations and when people are interacting with AI.Foundational
• Triton Inference ServerNVIDIA's open-source inference server (now Dynamo-Triton) for serving models from many frameworks on GPUs and CPUs.Intermediate
• UnderfittingWhen a model is too simple to capture underlying patterns and performs poorly on both training and new data.Foundational
• Unified Fabric Manager (UFM)NVIDIA's platform for provisioning, monitoring and managing InfiniBand fabrics.Advanced
• Unsupervised LearningMachine learning that finds structure, groupings or patterns in data that has no labels.Foundational
• Validation DatasetHeld-out data used during development to tune hyperparameters and select models.Foundational
• Vanishing GradientA training problem where gradients shrink toward zero in early layers, so they learn very slowly or not at all.Intermediate
• Variational Autoencoder (VAE)A generative autoencoder that learns a probabilistic latent space from which new samples can be drawn.Advanced
• Vector DatabaseA database optimized for storing embeddings and performing fast similarity search over them.Foundational
• Vector IndexA data structure that organizes embeddings for efficient similarity search.Intermediate
• Vertex AIGoogle Cloud's unified AI platform for building, deploying and scaling ML models and generative AI applications and agents.Foundational
• Vision Transformer (ViT)A transformer that processes an image as a sequence of fixed-size patches embedded like tokens.Advanced
• Vision-Language Model (VLM)A model that jointly understands images and text, for tasks such as visual question answering and captioning.Intermediate
• vLLMA popular open-source, high-throughput LLM serving engine built around PagedAttention.Advanced
• WatermarkingEmbedding detectable signals in AI-generated content to identify it as machine-made.Intermediate
• WeightsLearned numeric parameters that scale the strength of connections between neurons.Foundational
• Workflow OrchestrationCoordinating LLM calls and tools through predefined code paths such as chaining, routing and parallelization.Intermediate
• XGBoostExtreme Gradient Boosting: a fast, regularized, widely used gradient-boosted decision tree library.Intermediate
• XML Tags in PromptsUsing XML-style tags to delimit instructions, documents and examples within a prompt.Intermediate
• Zero-Shot PromptingAsking a model to perform a task with instructions only and no examples.Foundational