Concepts

This page defines the concepts used throughout DQM-ML documentation.

Note: These definitions are under review and will be updated to be consistent with definitions released by ETAIA. They come from the Confiance.ai program without modification, with one addition: Embedding.

Formal Concepts

These are the theoretical concepts that form the foundation of the data quality methodology.

Sample

A Sample is a collection of Features corresponding to one observation or experience that will be inferred by one AI Model. May contain data fields (image, signal, text, numerical values) and metadata fields.

Feature

A Feature refers to information extracted from data associated with each Sample. May be raw (acquired image, signal, token) or derived from other features.

Sample Selection

A Sample Selection is a set of samples chosen for processing or analysis. May represent full datasets, filtered subsets, or programmatically defined selections.

See Data Selection for how sample selections are configured.

Metric

A Metric is a quantitative measure computed on a group of samples from a Sample Selection. May be global, feature-level, absolute (single selection), or relative (comparing two selections).

See Metrics Guide for available metrics.

Batch

A Batch is a memory-fit subset of a Sample Selection used for efficient incremental processing.

Batch Metric

A Batch Metric is an intermediate metric computed on a Batch, designed to be aggregated into a final Metric without loading all samples at once.

See Metrics for how batch metrics work.

Domain Gap

A Domain Gap is a relative Metric comparing two Sample Selections (typically training vs. deployment), measuring distributional divergence or mismatch.

See Domain Gap for configuration details.

Embedding

An Embedding is a representation of a high-dimensional Sample into a lower-dimensional Feature of numerical vectors.

See Embedding Features for how embeddings are computed.

Core Concepts

These are DQM-ML-specific concepts that implement the formal framework above.

Data Selection

A Data Selection is a named subset of your data that can be used to compute Metrics, Features, or Domain Gap. It defines specific portions of your dataset for targeted quality assessments and comparisons.

See Data Selection for configuration details.

Processor

A Processor is a set of tools that process Batches or Samples to compute Features, Metrics, or Domain Gaps.

See Configuration Guide for how processors are configured.