Using AI Center

This page lists the core concepts used within AI Center.

Projects

A Project is an isolated group of resources (datasets, pipelines, packages, skills and logs) you may use to enable building a specific ML solution for different business automations.

ML Package

An ML Package is a group of package versions of the same package type. Think of it as a folder for holding package versions of the same type. A Package Version is a trained model you may deploy to a skill in order to integrate it into an RPA workflow.

Datasets

A Dataset is a folder of storage containing arbitrary files and sub-folders. A model is trained on a dataset.

Pipelines

Pipelines represent the various actions you may perform on packages or package versions.

It represents a description of an ML workflow, including all of the functions in the workflow and their order of execution. The pipeline includes the definition of the inputs required to run it and outputs to get from it.

A Pipeline Run is an execution of a pipeline based on code provided by the user. This code is where the functions called in the pipeline are actually implemented.

There are three types of pipelines:

• Training Pipeline - takes as input a package and a dataset, and produces a new package version.
• Evaluation Pipeline - takes as input a package version and a dataset, and produces a set of metrics and logs.
• Full Pipeline - runs a training pipeline and immediately after an evaluation pipeline.

ML Skills

An ML Skill is a live deployment of a package version, it can be used in an RPA workflow simply by dragging and dropping an ML skill activity in UiPath Studio.

ML Logs

ML Logs are a consolidated view of all events related to a project.

Examples

User flow

A user creates a project, uploads a trained package (or selects one of the provided packages), and deploys it as a skill.

An RPA developer can now drag and drop an activity to use the model in production.

User flow for continuous model retraining

A user creates a project and uploads a folder with data into a dataset. Then the user uploads a package that has yet-to-be trained, executes a training pipeline that outputs a trained model, and lastly deploys the trained model as a skill.

An RPA developer can then drag and drop an activity to use the model in production. In addition, the RPA developer can now send new labeled data back to the created dataset for the model to be continuously retrained.