This section includes definitions of common words and phrases associated with aiWARE.
Access-control entry (ACE) When managing object level permissions, ACE is the combination of a group, an object, and the permission set that a member of the group can use on an object.
Access-control list (ACL) When managing object level permissions, an ACL is a collection of ACEs.
Adapter A Docker container that ingests data from an external source and provides the extract, transform, load (ETL) logic to import that data into aiWARE. Data may be in the form of a file or stream.
Aggregator engine A Docker container that collects related data to organize it. Aggregator engines process the output from one or more other engines.
aiWARE The brand name of Veritone's AI operating system.
AI Data Veritone AI Data is the digital asset management (DAM) system used to manage, share and monetize content. AI Data ingests and manages digital media and text, provides long-term storage of job results, and provides hosted applications.
AI Processing AI Processing is the cognitive processing subsystem of aiWARE. Its primary purpose is to ingest data and to cognitively analyze this data in a predictable manner that meets business objectives.
AION Veritone's output standard.
API Application Programming Interface. An API specifies how software components can be interacted with programmatically. Veritone's primary application programming interface (API) is a GraphQL interface. It is located at https://api.veritone.com/v3/graphql and can be accessed like other HTTP APIs (e.g. REST) with POST requests.
Application In aiWARE, a bundle of software that can be installed into an organization and provisioned to users. An application usually has a URL, may call aiWARE's API, may process ingest data using adapters or process data using engines, and may define particular data structures through schemas.
Build An uploaded Docker container containing the code for a particular version of an engine.
Capabilities Abilities of AI with analogs to human senses and cognition (e.g., natural language processing, object detection, face recognition).
Classification In data science, the act of determining to what group a certain observation belongs. For example, determining if an image is of an object, animal or person.
Cognition The mental act or process by which knowledge is acquired, including perception, intuition, and reasoning.
Cognitive engines AI engines that process data in a way that mimics the functioning of the human brain to extract value or insight from unstructured data.
Container A container is a standard unit of software that packages up code and all its dependencies so the application runs quickly and reliably from one computing environment to another. In aiWARE this is almost always a Docker container.
Controller A controller, in a computing context, is a hardware device or a software program that manages or directs the flow of data between two entities. In aiWARE, the controller is the central service that registers hosts and engine instances, manages all work, and communicates to the database layer. Within a multi-node instance, a single controller is promoted to be a primary controller and serves as the AI Processing supervisor, responsible for specific critical functions.
Data science A study which deals with identification, representation and extraction of meaningful information from data sources to be used for business purposes.
Dataset In data science, a collection of data used to train an engine.
Default organization The organization where the sysadmin is added. It is meant to be the default or where you add users that do not need elevated permissions like instance administrator.
Detection The process of determining where an object is in unstructured data. For example, finding the faces in an image. This is distinct from recognition, which determines which thing is in the unstructured data.
Docker A computer program that performs operating-system-level virtualization, also known as "containerization".
Engine An algorithm that takes some data in and outputs some insights or calculations regarding that data. Veritone has three main types of engines: cognition, aggregator, and ingestion.
Engine processing mode The way in which an engine consumes data: either by segment or by stream.
Entity A record of a defined concept, such as a person, company, organization, advertising campaign, type of object, etc. in a library. Each entity is given an ID, which can be referenced by engines and other parts of aiWARE. Each entity can have one or more identifiers (i.e. picture of a face, audio clip) that can be used to train engines to find that entity.
Functional permission When managing object level permissions functional permission is a single permission, for example, read.
GraphQL A query language for APIs and a runtime for fulfilling those queries with your existing data.
Group When managing object level permissions, a group is an object that contains user accounts as members of a particular group.
Host A physical computer or a virtual machine (VM).
Ingestion Refers to bringing external data into aiWARE where it can be processed. Adapters are Docker containers that ingest data into aiWARE. Data may be in the form of a file or stream.
Instance An aiWARE instance refers to either a single node running aiWARE, or, in multi-node installations, to the entire set of nodes running any parts of an aiWARE stack.
Job A list of tasks to run against a piece of data.
Library A collection of named entities, along with files that act as identifiers for those entities. Libraries allow engines to tag their results with entity IDs that point to a specific person, place, thing, etc. rather than just using textual descriptions. To make an engine aware of a library, train it against that library.
Manifest A JSON-formatted text file that describes the characteristics of an engine build.
Multi-node An installation model for aiWARE in which multiple computers (or virtual computers) run one or more of the run modes of aiWARE. Together, these computing resources form an instance of aiWARE.
Natural Language Processing (NLP) A branch of AI that understands, interprets, and manipulates human language.
Node In aiWARE, a node is any computer resource (CPU or VCPU) that is running one or more of aiWARE's run modes as part of an aiWARE instance.
Object When managing object level permissions, refers to an object that is going to be controlled (secured).
Object Level Permission (OLP) An access control model that allows users to control who can access which objects, how, and when.
Optical Character Recognition (OCR) The ability of a program to identify characters and words in unstructured data.
Orchestration The automated arrangement, coordination, and management of computer systems, middleware, and services. In aiWARE, this predominantly refers to the coordination of multiple engines to accomplish a common task.
Organization An entity that centrally controls access to different resources and assets. An aiWARE organization controls access to resources for AI processing, such as media files or other data assets, and processing tools, including engines, applications, schemas, and flows. A Hub organization controls Hub user access to administrative functions for aiWARE instances. When managing object level permissions, an organization is an object that contains a set of groups.
Package A bundle of one or more aiWARE resources. Packages are used to create custom AI solutions. By combining interrelated resources (such as an engine and its required schema, or a group of translation engines for multiple languages) into a single package, they also make it easier to distribute resources for use by other organizations.
Package grants Permissions granting a package to be shared across organizations.
Permission set When managing object level permissions, a permission set is a named set of functional permissions.
Predictive modeling Exploiting patterns found in historical and transactional data to identify risks and opportunities in future data.
Real-time processing Processing data and returning results without significant delay.
Recognition The process of determining which object is in some unstructured data. For example, whose face is in the image? This is distinct from detection which determines where an object is.
Resource Any object used to build solutions in aiWARE. Resources enhance and orchestrate the processing capabilities of organizations. Examples include engines, schemas, flows, and applications.
Root admin The organization where the CLI and hub will provision packages. It is also where the base applications will exist that are provided through flyway. A user exists in this organization that is used for provisioning.
Schema A definition of the structure a piece of structured data must conform to. In aiWARE, schemas can be registered in Veritone Developer and are written using JSON.
Segment engine mode An engine processing mode that accepts a discrete "chunk" of data (i.e. file, video frame, audio segment).
Sentiment analysis Determines the attitude, contextual polarity or emotional reaction of a speaker or writer with respect to some topic, document, interaction or event.
Single node An installation model for aiWARE in which a single computer runs an instance of aiWARE.
Stream engine mode An engine processing mode that accepts the raw byte stream of incoming data. Because it accepts the full byte stream, it can save some state and output results based on past behavior.
Structured data Refers to data that resides in a recognizable structure (e.g. relational databases and spreadsheets). In aiWARE, structured data often conforms to a known schema. AI attempts to mine meaning in unstructured data and produce structured data results.
System control When managing object level permissions, system control is the feature that assigns permission sets to objects.
Task A request for a single engine to be run against a piece of data. A task also includes all the parameters, libraries, and other configuration needed to run the cognition properly. The task is the smallest unit of work in aiWARE. A task specifies an engine that will be run against a TDO.
Temporal Data Object (TDO) An all-purpose container object, aggregating information about jobs, assets, and temporal data, among other things.
Training The technique of giving an engine a dataset or library that will "teach" the engine to produce better or more specific results.
Training model Data generated during an engine's training step. The model is (in some cases) provided to an engine when the engine is run. A model can optionally contain a file representing the model data. For example, a face recognition engine trained on a library of people entities might generate a model as a file containing all the maps of face hashes to entity IDs. When a picture of a face is passed to the engine - along with the library of people specified in the task's payload - the training model (and its associated asset) would be used by the engine to return the correct entity ID.
Unstructured data Any data that does not have a parsable structure. It is unorganized and raw and can be non-textual or textual. Video, images, and audio recordings, text documents, and PDFs are all examples of unstructured data.
User When managing object level permissions, a user refers to a user account.
