Skip to content

Algorithms in DIGITbrain

Introduction

In DIGITbrain an Algorithm is handled as a preconfigured black box software (SW) component, providing only a description of its capabilities. The DIGITbrain approach enables users with low experience to easily make use of complex algorithms by simply providing the required inputs (i.e., data, model).

An Algorithm bundles one or more Microservice(s), that (together) enable the evaluation of a Model on Data. Within DIGITbrain, Algorithm shall be agnostic to IP instances (e.g. their application field or the context in which a model is being used). This context will be established later, when pairing the algorithm with a adequat Model as an MA-Pair, and further when providing Data to form the DMA-Tuple. The different Microservices within an Algorithm may be deployed on different resources (e.g. edge, cloud, HPC) depending on the individual mechanism of the Algorithm, and the needs of the Model.

Pre-requisites

To integrate an Algorithm within the DIGITbrain Solution, the following steps need to be performed:

  • Authoring: preparation and pre-configuration of Microservices using Authoring Tools.
  • Packaging: the authored Microservice is packaged following the Asset guidelines of the DIGITbrain Solution (i.e., Docker containers).
  • Publishing: definition and provision of metadata description for Microservices together with their publication in the DIGITbrain Asset Metadata Registry. Additionally, the different Microservices need to be linked to an Algorithm, defining its metadata, and publishing it as a new Algorithm Asset to the DIGITbrain Asset Metadata Registry using the respective Publishing Interface.

Describing Algorithms

For a closer look at the fields available for Algorithms, use the following resources: