2023.4.0

Release date: 26 April 2023

AI Center 2023.4 LTS Release

We are excited to release AI Center 2023.4 LTS!

What's New

AI Center API

You can now manage your AI Center™ projects, datasets, ML packages, ML skills, and pipelines from an external application with the use of AI Center API endpoints.

Check the related pages in the new AI Center API section of the guide:

• Overview
• API List

Improved training pipelines

Training pipelines now also provide model performance scores.

ML Package version display

This release brings several improvements in how ML package versions are displayed:

• If you create a new version of the model, your new version will be displayed at the top of the drop-down list.
• Each minor version of a package is now displayed in the same category with the major version.

New OCR Engine

UiPath Chinese-Japanese-Korean OCR is added to Document Understanding models.

New default language package for ML Packages

The new default language when creating a new ML Package is Python 3.9.

Notification Service integration

We are happy to announce the new notifications feature within Automation Cloud™ portal for AI Center events. You can now receive built-in, in-app notifications to notify you with the most relevant events:

• Dataset
  • Download readiness
• ML Package validation
• ML Pipelines status change
• ML Skill status change

New Open Source ML package versions

This release brings new versions for the Open Source ML packages:

• ObjectDetection
  • Version: 22.12.0
  • Python Version: Python 3.8 OpenCV
• NamedEntityRecognition
  • Version: 22.12.0
  • Python Version: Python 3.8
• ImageModeration
  • Version: 22.12.0
  • PythonVersion: Python 3.8
• TPOTAutoMLRegression
  • Version: 22.12.0
  • PythonVersion: Python 3.8
• LanguageDetection
  • Version: 22.12.0
  • PythonVersion: Python 3.8
• MultilingualTranslator
  • Version: 22.11.2
  • PythonVersion: Python 3.8
• SentimentAnalysis
  • Version: 22.12.0
  • PythonVersion: Python 3.8

Improvements

• This release brings improvements to cluster and infrastructure stability.
• This release brings general security and accessibility improvements.
• For improved usability, a proper exception is used instead of a 500 error when deleting a Project that was already soft deleted.
• For improved usability, the Action Center data labeling task titles are renamed to DL-sessionName-filename.
• Packages created from the Document Understanding center no longer have NULL in the Modified by field of the package details.
• For improved usability, the Waiting for licenses error message due to licenses not available is updated to a more clear message. A failed reason for the pipeline is now added in the table. The failed reason is displayed in the interface if the pipeline failed.
• For improved usability, the Downloading message when a dataset is ready for download is updated to a more clear message. The updated message is now Packaging.
• If a Full Pipeline fails during the Evaluation step, the trained model is now kept.
• Minimum pipeline frequency has been raised to one hour in order to prevent scheduling too many pipelines, which can lead to infra issues.. Previously, recurring pipelines with wrong cron jobs could have been started every minute, leading to a high amount of resource consumption and waiting times.
• The Time based interface from the Create new pipeline run page is improved.
• If you are using the AI Units licensing model, AI Robots are no longer displayed in the Licensing page.
• Required fields from the Data Labeling Session / Create New form are now marked as Required with an asterisk.
  
  

  

  
  

Bug Fixes

• Fixed an issue where in certain situations a Data Labeling session was not created.
• Fixed an issue where skill deletion was not retried in case of kubernetes operation failure.
• Fixed an issue where in some cases users are added by defaults to projects without having any permissions assigned.
• Previously, if a skill was deleted while a deployment is in progress, an alert was raised and the deployment failed. This is now fixed and no alert is raised in this case.