Glossary

Core MAITE Terms

AI problem

Defined by choosing concrete types for the three primitives (input type, target type, and metadata type) and specifying behavioral expectations; examples include image classification, object detection, and multi-object tracking; see MAITE Layered Architecture for more details

Augmentation

A MAITE component that takes a batch of data as input and returns a potentially modified batch of data as output; see the Augmentation API for more details

component

Implementer of a MAITE-defined Python protocol class (DataLoader, Dataset, Augmentation, Model, or Metric) that follows prescribed semantics; see Vision for Interoperability in AI Test and Evaluation for more details

DataLoader

A MAITE component that provides batch-level data access via an iterator; see the DataLoader API for more details

Dataset

A MAITE component that provides datum-level data access via index-based lookup; see the Dataset API for more details

datum

An individual data item that’s a tuple of input, target (output), and metadata

Metric

A MAITE component that computes some measure of “agreement” between model predictions and ground-truth labels; see the Metric API for more details

Model

A MAITE component that takes a batch of inputs and produces a batch of outputs, with types appropriate to the particular AI problem; see the Model API for more details

primitive

Object with class and semantics of a member variable type, argument type, or return type of a MAITE-defined Python protocol class; see Vision for Interoperability in AI Test and Evaluation for more details

task

A Python callable that accepts only arguments typed as MAITE components or MAITE primitives, and returns MAITE components, MAITE primitives, and/or Python objects of built-in/broadly-accepted types with well-documented semantics; see Vision for Interoperability in AI Test and Evaluation for more details

wrapper

A Python class that implements a MAITE component protocol by translating to and from a native component

Typing Concepts

ArrayLike

A protocol type representing objects that can be coerced to numpy arrays; the foundational inherent type for most MAITE primitives including images, bounding boxes, and model outputs; see ArrayLike for details

batch

A collection of multiple data items (datums) processed together; DataLoader MAITE-defined protocol classes yield batches, while Dataset MAITE-defined protocol classes provide individual datums that are collected into batches; batches are the fundamental unit for Model, Augmentation, and Metric MAITE-defined protocol classes

inherent type

The actual Python type (e.g., ArrayLike, TypedDict) before domain-specific aliasing; the runtime type that best fits from the Python language; the starting point in MAITE’s three-layer type alias system (inherent type → semantic alias → role alias); see MAITE Layered Architecture for more details

role alias

Type alias specifying which semantic type occupies which position in generic protocols (e.g., InputType, TargetType, DatumMetadataType); the final layer in MAITE’s type alias system (inherent type → semantic alias → role alias); makes protocol signatures both generic and self-documenting; see MAITE Layered Architecture for more details

semantic alias

Type alias that captures domain meaning (e.g., Image, BoundingBox) with behavioral expectations documented in its docstring; the middle layer in MAITE’s type alias system (inherent type → semantic alias → role alias); separates what a type technically is from what it means in the domain; see MAITE Layered Architecture for more details

static type checker

A tool (e.g., Pyright, mypy) that analyzes code for type compatibility at development time without executing it; MAITE recommends Pyright for verifying protocol compliance

structural subtyping

Type compatibility determined by matching attributes, methods, and type signatures rather than nominal inheritance; Python Protocol classes and TypedDict classes use structural subtyping to enable plug-and-play component substitution without requiring explicit inheritance; see the Python documentation on protocols for more information

Ecosystem and General Terms

interoperability

The ability of components from different libraries to work together seamlessly; MAITE’s primary design objective is enabling broad interoperability across the JATIC ecosystem through standardized interfaces

JATIC

Joint AI Test Infrastructure Capability; the broader ecosystem of AI test and evaluation Python libraries that MAITE serves by providing common interfaces and standards

protocol

A Python structural type used to provide strict, consistent, and machine-readable definition of MAITE components that specify minimum expected attribute names, attribute types, method names, and method type signatures; see the Python documentation on protocols for more information

test and evaluation (T&E)

The process of evaluating the performance of an AI model under various conditions (that hopefully match/mimic the deployment environment as closely as possible)