# Multimodal Data Glossary

Dive into our comprehensive multimodal data glossary. Quickly find terms and concepts using the search bar or browse alphabetically for a thorough exploration.

All#ABCDEFGHIJKLMNOPQRSTUVWXYZ

## A

[

### ACID - Atomicity, Consistency, Isolation, Durability

](/glossary/acid)[

### API - Application Programming Interface

](/glossary/api)[

### Agentic Workflows - LLM-driven processes that plan, call tools and loop until a task is done

](/glossary/agentic-workflows)[

### Approximate Nearest Neighbor (ANN) - Fast similarity search trading exactness for speed

](/glossary/approximate-nearest-neighbor)[

### Attention Mechanism - Dynamic weighting of input elements for contextual processing

](/glossary/attention-mechanism)[

### Audio Embedding - Dense vector representations of audio content

](/glossary/audio-embedding)[

### Audio Classification - Categorizing audio clips by content type or event

](/glossary/audio-classification)[

### Acoustic Fingerprinting - Creating compact identifiers for audio content recognition

](/glossary/acoustic-fingerprinting)[

### Automatic Speech Recognition (ASR) - Converting spoken language into written text

](/glossary/automatic-speech-recognition)[

### Audio Denoising - Removing unwanted noise from audio recordings

](/glossary/audio-denoising)[

### Autocomplete - Suggesting completions as users type search queries

](/glossary/autocomplete)[

### Active Learning - Strategically selecting data for human labeling

](/glossary/active-learning)[

### Agentic AI - AI systems that autonomously plan and execute multi-step tasks

](/glossary/agentic-ai)[

### Agentic Retrieval - AI agents that autonomously plan and execute multi-step retrieval strategies

](/glossary/agentic-retrieval)[

### Agentic RAG - RAG systems with autonomous agents that plan and execute multi-step retrieval

](/glossary/agentic-rag)[

### Audio Fingerprinting - Identifying audio content through spectral embeddings

](/glossary/audio-fingerprinting)

## B

[

### Blob Storage - Binary Large Object Storage

](/glossary/blob-storage)[

### BYOD - Bring Your Own Data

](/glossary/byod)[

### Bounding Box - The rectangle that marks where an object sits in an image or page

](/glossary/bounding-box)[

### Boolean Retrieval - Exact match search

](/glossary/boolean-retrieval)[

### BM25 - Best Matching 25

](/glossary/bm25)[

### BERT - Bidirectional Encoder Representations from Transformers

](/glossary/bert)[

### Bi-Encoder - Dual-tower model encoding queries and documents independently

](/glossary/bi-encoder)[

### Brand Safety - Ensuring advertisements and brand content appear alongside appropriate material

](/glossary/brand-safety)[

### Brand Safety Scanning - Automated detection of brand logos and trademarks in video/image content

](/glossary/brand-safety-scanning)[

### Brand Safety Scanning - Automated detection of brand risks in media content

](/glossary/brand-safety-scanning)[

### Bucket - A schema-backed container for the raw objects you upload, before any processing

](/glossary/bucket)

## C

[

### CLIP - Contrastive Language-Image Pretraining

](/glossary/clip)[

### Content-Based Retrieval - Feature-based search

](/glossary/content-based-retrieval)[

### Cosine Similarity - Similarity measure

](/glossary/cosine-similarity)[

### Convolutional Neural Network (CNN) - A neural network that learns image features with sliding filters

](/glossary/convolutional-neural-network)[

### Contrastive Learning - Learning representations by comparing similar and dissimilar pairs

](/glossary/contrastive-learning)[

### Cross-Modal Retrieval - Searching across different data types with unified queries

](/glossary/cross-modal-retrieval)[

### Cross-Attention - Attention between two different input sequences or modalities

](/glossary/cross-attention)[

### Chain-of-Thought - Prompting models to show step-by-step reasoning

](/glossary/chain-of-thought)[

### Content-Based Image Retrieval - Retrieving images by analyzing their visual content

](/glossary/content-based-image-retrieval)[

### ColBERT - Late-interaction retrieval model using contextualized token embeddings

](/glossary/colbert)[

### Cross-Encoder - Joint query-document encoding model for precise relevance scoring

](/glossary/cross-encoder)[

### Content Moderation - Automated filtering and classification of user-generated content for safety

](/glossary/content-moderation)[

### Computer Vision - The field of AI focused on enabling machines to interpret and understand visual data

](/glossary/computer-vision)[

### ColPali - Late-interaction retrieval model for visually rich document pages

](/glossary/colpali)[

### Cross-Modal Retrieval - Searching across different data types using any modality as the query

](/glossary/cross-modal-retrieval)[

### Clinical NLP - Natural language processing for medical text

](/glossary/clinical-nlp)[

### Celebrity Likeness Detection - AI-powered identification of recognizable faces in media content

](/glossary/celebrity-likeness-detection)[

### Celebrity Likeness Detection - AI-powered identification of recognizable faces in media

](/glossary/celebrity-likeness-detection)[

### Context Engineering - The discipline of designing and building the information environment that an AI system needs to perform a task correctly

](/glossary/context-engineering)[

### Collection - A processed, searchable set of documents derived from one or more sources

](/glossary/collection)[

### Cluster - A group the system discovered in your data, rather than a label you declared

](/glossary/cluster)

## D

[

### Data Lakehouse - Hybrid data architecture

](/glossary/data-lakehouse)[

### Data Modality - Type or form of data

](/glossary/data-modality)[

### Data Tokenization - Data segmentation

](/glossary/data-tokenization)[

### Dimensionality Reduction - Data simplification

](/glossary/dimensionality-reduction)[

### Dense Retrieval - Retrieval using learned dense vector representations

](/glossary/dense-retrieval)[

### Depth Estimation - Predicting distance of scene points from camera

](/glossary/depth-estimation)[

### Data Pipeline - Automated workflow for moving and transforming data

](/glossary/data-pipeline)[

### Data Lake - Centralized repository storing raw data at any scale

](/glossary/data-lake)[

### Data Versioning - Tracking changes to datasets over time

](/glossary/data-versioning)[

### Data Lineage - Tracking data origin, movement, and transformations

](/glossary/data-lineage)[

### Data Catalog - Organized inventory of available data assets

](/glossary/data-catalog)[

### Data Mesh - Decentralized domain-oriented data architecture

](/glossary/data-mesh)[

### Diffusion Model - Generative model that creates data by denoising random noise

](/glossary/diffusion-model)[

### Document Understanding - AI-powered extraction of structured information from complex document layouts

](/glossary/document-understanding)[

### Dense Passage Retrieval - Embedding-based passage retrieval for open-domain QA

](/glossary/dense-passage-retrieval)[

### Document Intelligence - AI-powered extraction and understanding of structured and unstructured documents

](/glossary/document-intelligence)

## E

[

### Embedding - Vector representation

](/glossary/embedding)[

### Entity Linking - Connecting mentions

](/glossary/entity-linking)[

### ETL (Extract, Transform, Load) - Process of extracting, transforming, and loading data

](/glossary/etl)[

### Ensemble Methods - Combining multiple models for improved predictions

](/glossary/ensemble-methods)[

### Embedding Portability - The ability to move, share, or reuse vector embeddings across different systems, models, or organizations without loss of meaning

](/glossary/embedding-portability)[

### Embedding Versioning - Strategies for upgrading embedding models without breaking retrieval quality or causing downtime in production vector systems

](/glossary/embedding-versioning)

## F

[

### FAISS - Facebook AI Similarity Search

](/glossary/faiss)[

### Feature Extraction - Data representation

](/glossary/feature-extraction)[

### F1 Score - The harmonic mean of precision and recall

](/glossary/f1-score)[

### Fuzzy Search - Approximate match search

](/glossary/fuzzy-search)[

### Fuzzy Search (Levenshtein Distance) - Approximate match search

](/glossary/fuzzy-search-levenshtein)[

### Face Detection - Locating human faces in images and video

](/glossary/face-detection)[

### Feature Store - Centralized repository for machine learning features

](/glossary/feature-store)[

### Faceted Search - Filtering search results by structured attribute categories

](/glossary/faceted-search)[

### Fine-Tuning - Adapting pretrained models on task-specific data

](/glossary/fine-tuning)[

### Few-Shot Learning - Learning from very few labeled examples per class

](/glossary/few-shot-learning)[

### Federated Learning - Training models across decentralized data without sharing it

](/glossary/federated-learning)[

### Face Embedding - A vector that encodes who a face belongs to, so the same person's photos match

](/glossary/face-embedding)[

### Feature URI - A universal address for any feature emitted by a Mixpeek extractor, enabling query-time model compatibility across the warehouse.

](/glossary/feature-uri)[

### Feature Extractor - The unit of processing that turns raw media into searchable signals and vectors

](/glossary/feature-extractor)

## G

[

### GPU Acceleration - Performance boost

](/glossary/gpu-acceleration)[

### Generative AI - AI systems that create new content across modalities

](/glossary/generative-ai)[

### Grounding - Connecting AI outputs to verifiable source data

](/glossary/grounding)[

### GARM (Global Alliance for Responsible Media) - A World Federation of Advertisers initiative that set brand safety and suitability definitions, discontinued in August 2024

](/glossary/garm)[

### Grounding DINO - Open-set object detection model guided by text prompts

](/glossary/grounding-dino)

## H

[

### Hybrid Search - Combined search approach

](/glossary/hybrid-search)[

### HNSW (Hierarchical Navigable Small World) - Graph-based approximate nearest neighbor search algorithm

](/glossary/hnsw)[

### Hallucination - AI generating plausible but incorrect or fabricated information

](/glossary/hallucination)

## I

[

### Indexing (Multimodal) - Efficient retrieval

](/glossary/indexing-multimodal)[

### Image Segmentation - Partitioning images into meaningful regions or pixel masks

](/glossary/image-segmentation)[

### Image Classification - Assigning category labels to entire images

](/glossary/image-classification)[

### Image Captioning - Generating natural language descriptions of images

](/glossary/image-captioning)[

### Inverted Index - Data structure mapping terms to their document locations

](/glossary/inverted-index)[

### Index Optimization - Tuning search indices for performance and accuracy

](/glossary/index-optimization)[

### Image Similarity Search - Finding visually similar images using embedding-based vector comparison

](/glossary/image-similarity-search)[

### Image Embedding - Dense vector representations of image content

](/glossary/image-embedding)[

### IP Clearance - Pre-publication verification that content doesn't infringe intellectual property

](/glossary/ip-clearance)[

### IP Clearance - Intellectual property clearance before publication

](/glossary/ip-clearance)

## J

[

### JPEG, JPG, JSON, JSONL - Common data formats

](/glossary/jpeg-jpg-json-jsonl)

## K

[

### Knowledge Graph - Structured knowledge

](/glossary/knowledge-graph)[

### Keypoint Detection - Locating specific points on an object, such as joints or facial landmarks

](/glossary/keypoint-detection)

## L

[

### LLM - Large Language Model

](/glossary/llm)[

### Latent Space - Abstract vector space

](/glossary/latent-space)[

### Latent Semantic Indexing (LSI) - Concept-based retrieval

](/glossary/latent-semantic-indexing)

## M

[

### Multimodal Fusion - Cross-modal integration

](/glossary/multimodal-fusion)[

### Multimodal Retrieval - Cross-modal search

](/glossary/multimodal-retrieval)[

### Metadata - Descriptive data

](/glossary/metadata)[

### Media Asset Management (MAM) - Software for storing, finding and reusing video and audio

](/glossary/media-asset-management)[

### Mean Reciprocal Rank (MRR) - How high the first correct result ranks, averaged over queries

](/glossary/mean-reciprocal-rank)[

### Mean Average Precision (MAP) - Precision averaged over every relevant result, then over queries

](/glossary/mean-average-precision)[

### Machine Translation - Automatically translating text between languages

](/glossary/machine-translation)[

### Music Information Retrieval - Extracting structured information from music audio

](/glossary/music-information-retrieval)[

### Mel Spectrogram - Frequency-time representation aligned with human hearing

](/glossary/mel-spectrogram)[

### Model Distillation - Compressing large models into smaller efficient ones

](/glossary/model-distillation)[

### Multimodal Alignment - Learning shared representations across different data types

](/glossary/multimodal-alignment)[

### Multimodal Search - Search across multiple data types like text, images, video, and audio in a single query

](/glossary/multimodal-search)[

### Multimodal RAG - Retrieval-augmented generation across multiple content types

](/glossary/multimodal-rag)[

### Multimodal Learning - Machine learning across multiple data modalities simultaneously

](/glossary/multimodal-learning)[

### Multimodal Foundation Model - Large pretrained models that process multiple data modalities

](/glossary/multimodal-foundation-model)[

### Multimodal RAG - Retrieval-Augmented Generation across text, images, video, and audio

](/glossary/multimodal-rag)[

### Multimodal AI - AI systems capable of processing and reasoning across multiple data types simultaneously

](/glossary/multimodal-ai)[

### Multimodal Data Warehouse - An integrated system that decomposes unstructured objects into queryable features, stores them across cost tiers, and reassembles them through multi-stage retrieval pipelines

](/glossary/multimodal-data-warehouse)[

### Multi-Stage Retrieval Pipeline - A composable chain of filter, sort, reduce, enrich, and apply stages that progressively refine search results over unstructured data.

](/glossary/multi-stage-retrieval-pipeline)[

### Multi-Stage Retrieval - A pipeline that chains discrete search operations to express complex information needs.

](/glossary/multi-stage-retrieval)[

### Model Context Protocol (MCP) - An open standard for connecting AI agents to external tools and data sources

](/glossary/model-context-protocol)[

### Multimodal Embeddings - Vector representations that encode different data types (text, images, video, audio) into a shared mathematical space for cross-modal search and comparison

](/glossary/multimodal-embeddings)

## N

[

### Neural Search - Deep learning-based search

](/glossary/neural-search)[

### NER - Named Entity Recognition

](/glossary/ner)[

### NDCG (Normalized Discounted Cumulative Gain) - Ranking quality with graded relevance, scored from 0 to 1

](/glossary/ndcg)[

### Namespace - An isolated vector space with its own schema, indexes and access boundary

](/glossary/namespace)

## O

[

### Optical Character Recognition (OCR) - Turning text in images, scans and video frames into machine-readable text

](/glossary/ocr)[

### Object Detection - Locating and classifying objects within images or video

](/glossary/object-detection)[

### Optical Flow - Estimating pixel-level motion between video frames

](/glossary/optical-flow)[

### Object Decomposition - The process of breaking complex unstructured files into their semantic components (features) for independent indexing and retrieval.

](/glossary/object-decomposition)

## P

[

### Pretrained Models - Pretrained models

](/glossary/pretrained-models)[

### Prompt Engineering - Query design

](/glossary/prompt-engineering)[

### Precision and Recall - Retrieval metrics

](/glossary/precision-recall)[

### Product Quantization - Vector compression by subspace decomposition and codebook encoding

](/glossary/product-quantization)[

### Pose Estimation - Detecting human body joint positions in images

](/glossary/pose-estimation)[

### Pruning - Removing redundant parameters from neural networks

](/glossary/pruning)[

### Pre-Publication Screening - Content review workflow before distribution

](/glossary/pre-publication-screening)[

### Pre-Publication Screening - Content review before distribution

](/glossary/pre-publication-screening)

## Q

[

### Query Expansion - Enhanced search

](/glossary/query-expansion)[

### Question Answering - Automatically finding answers to natural language questions

](/glossary/question-answering)[

### Query Understanding - Interpreting user search intent and meaning

](/glossary/query-understanding)[

### Quantization - Reducing model precision for efficient inference

](/glossary/quantization)

## R

[

### RAG - Retrieval-Augmented Generation

](/glossary/rag)[

### Relevance Ranking - Result ordering

](/glossary/relevance-ranking)[

### Relevance Feedback - User feedback refinement

](/glossary/relevance-feedback)[

### RAG Evaluation (RAGAS) - Measuring whether a RAG system retrieves the right context and answers from it

](/glossary/rag-evaluation)[

### ROC Curve and AUC - How a classifier trades true positives against false positives across thresholds

](/glossary/roc-auc)[

### Representation Learning - Learning useful features from raw data automatically

](/glossary/representation-learning)[

### Re-ranking - Refining search result order after initial retrieval

](/glossary/reranking)[

### Retrieval-Augmented Generation - Grounding language model outputs with retrieved information

](/glossary/retrieval-augmented-generation)[

### Reverse Video Search - Finding matching or similar videos by querying with a video, clip, or frame

](/glossary/reverse-video-search)[

### Retriever - A saved, versioned search pipeline: the ordered stages that turn a query into ranked results

](/glossary/retriever)

## S

[

### Schema-on-Read - Flexible data modeling

](/glossary/schema-on-read)[

### Speech-to-Text (STT) - Audio transcription

](/glossary/speech-to-text)[

### Synonyms - Alternative words

](/glossary/synonyms)[

### Sentence Transformers - Models producing semantically meaningful sentence embeddings

](/glossary/sentence-transformers)[

### Sparse Retrieval - Retrieval using high-dimensional sparse term-based vectors

](/glossary/sparse-retrieval)[

### Siamese Network - Twin networks sharing weights for similarity comparison

](/glossary/siamese-network)[

### Scene Recognition - Classifying the environment or setting in images

](/glossary/scene-recognition)[

### Sentiment Analysis - Detecting emotional tone and opinion in text

](/glossary/sentiment-analysis)[

### Speaker Diarization - Identifying who spoke when in audio recordings

](/glossary/speaker-diarization)[

### Schema Evolution - Managing changes to data structure over time

](/glossary/schema-evolution)[

### Streaming Data - Continuous real-time data processing as it arrives

](/glossary/streaming-data)[

### Semantic Search - Search based on meaning rather than exact keywords

](/glossary/semantic-search)[

### Sharding - Distributing data across multiple storage nodes

](/glossary/sharding)[

### Search Relevance - Measuring how well search results match user needs

](/glossary/search-relevance)[

### Self-Supervised Learning - Learning representations from unlabeled data using pretext tasks

](/glossary/self-supervised-learning)[

### Semantic Chunking - Splitting documents into meaningful segments based on content boundaries rather than fixed sizes

](/glossary/semantic-chunking)[

### SPLADE - Learned sparse retrieval model using term expansion

](/glossary/splade)[

### SigLIP - Sigmoid loss for image-language pretraining, an improved CLIP variant

](/glossary/siglip)[

### Storage Tiering - Automatic lifecycle management that moves vector data between hot, warm, and cold storage tiers based on query frequency and cost targets.

](/glossary/storage-tiering)[

### Semantic Join - A cross-collection enrichment operation that attaches context from one collection to results from another, using semantic similarity as the join key.

](/glossary/semantic-join)[

### Storage Tiering - Placing data on different storage backends based on access frequency and cost requirements.

](/glossary/storage-tiering)

## T

[

### Text-to-Image / Image-to-Text - Cross-modal tasks

](/glossary/text-to-image-image-to-text)[

### Tensor - Data representation

](/glossary/tensor)[

### TF-IDF - Term importance measure

](/glossary/tf-idf)[

### Tokenization - Splitting text into discrete units for model processing

](/glossary/tokenization)[

### Transformer Architecture - Self-attention-based neural network architecture

](/glossary/transformer-architecture)[

### Text Classification - Assigning predefined categories to text documents

](/glossary/text-classification)[

### Topic Modeling - Discovering abstract themes across document collections

](/glossary/topic-modeling)[

### Text Summarization - Condensing documents into shorter representative text

](/glossary/text-summarization)[

### Text-to-Speech (TTS) - Synthesizing natural-sounding speech from text

](/glossary/text-to-speech)[

### Transfer Learning - Reusing knowledge from pretrained models for new tasks

](/glossary/transfer-learning)[

### Taxonomy - A declared label hierarchy that content is classified against, flat or nested

](/glossary/taxonomy)

## U

[

### Unstructured Data - Non-tabular data

](/glossary/unstructured-data)

## V

[

### Vector Database - Embedding storage and retrieval

](/glossary/vector-database)[

### Vision-Language Model (VLM) - Multimodal understanding

](/glossary/vision-language-model)[

### Video Analysis AI - AI that watches video and returns what is in it, with timestamps

](/glossary/video-analysis-ai)[

### Vector Quantization - Compressing vectors by mapping to codebook entries

](/glossary/vector-quantization)[

### Visual Grounding - Linking natural language to specific image regions

](/glossary/visual-grounding)[

### Voice Activity Detection - Detecting presence of human speech in audio

](/glossary/voice-activity-detection)[

### Vision Transformer (ViT) - Transformer architecture applied to image understanding

](/glossary/vision-transformer)[

### Video Intelligence - AI-powered analysis and understanding of video content at scale

](/glossary/video-intelligence)[

### Visual Search - Finding content using images as search queries

](/glossary/visual-search)[

### Video Embedding - Dense vector representations of video content

](/glossary/video-embedding)[

### Video Understanding - AI comprehension of visual and temporal video content

](/glossary/video-understanding)[

### Video Scene Detection - Automatically segmenting videos into distinct scenes or shots

](/glossary/video-scene-detection)[

### Visual Search - Image-based search technology for finding visually similar content

](/glossary/visual-search)[

### Video Understanding - AI comprehension of video content including scenes, actions, speech, and objects

](/glossary/video-understanding)[

### Vector Search - Semantic retrieval using vector embeddings

](/glossary/vector-search)[

### Visual Fingerprinting - Identifying copyrighted visual content through perceptual hashing

](/glossary/visual-fingerprinting)[

### Visual Fingerprinting - Techniques for identifying visual content through embeddings

](/glossary/visual-fingerprinting)

## W

[

### Waveform - Audio representation

](/glossary/waveform)[

### Word2Vec - Neural word embedding model using shallow networks

](/glossary/word2vec)[

### Warehouse Pricing - A pricing model that charges for ingestion and storage tiers rather than per query.

](/glossary/warehouse-pricing)[

### World Foundation Models - Large generative models that learn physics, geometry, and cause-effect from video to simulate and predict real-world environments

](/glossary/world-foundation-models)[

### Write-Ahead Log (WAL) - An append-only record of every change, made durable before the change is applied

](/glossary/write-ahead-log)

## X

[

### XAI - Explainable AI

](/glossary/xai)

## Z

[

### Zero-shot Learning - Generalization capability

](/glossary/zero-shot-learning)[

### Zero-Shot Classification - Classifying data into categories without task-specific training examples

](/glossary/zero-shot-classification)