Skip to main content

Header

10 August, 2021

AI4EU Experiments – New Release Summer 2021

Discover three major updates to the AI4EU Experiments platform: A re-designed interface makes the onboarding of AI models more convenient. The shared folder concept allows to keep large datasets in one place and access them flexibly. And finally, developers can now use the AI4EU Experiments platform to train their models. These features offer the European AI community new possibilities for collaboratively creating an even broader range of AI solutions.

Source
AI4EU – by Martin Welss & Inga Daase

Body

Logo AI4EU Experiments

Update 1: Streamlined Onboarding

Get your AI models on the AI4EU Experiments platform more quickly and conveniently!

AI4EU developers streamlined the onboarding process by removing all unsupported options from the onboarding interface. In addition, the re-designed interface now features the button “Set values for Docker Hub” to make the onboarding of docker containers even more convenient.

On Boarding on AI4EU Experiments Platform
The updated interface of the AI4EU Experiments platform makes the onboarding of AI models, especially docker containers, more convenient.


Update 2: Shared Folders

Keep large data sets in one central storage facility and access them flexibly!

With the AI4EU experiments platform, developers can create AI solutions, also called pipelines, which usually contain several building blocks or nodes. Some of these nodes might use the same data sets. With the new concept of shared folders, it is now possible to connect several nodes to a common storage facility. To use shared folders, pipeline builders can enter the Design Studio and add the node “SharedFolderProvider” to the pipeline. Then, the shared folder is ready to be connected to the nodes that need to access it.

AI4EU Experiments Shared Folders
Insight into the Design Studio: The developer created a pipeline for classification of news articles. The shared folder (with the green outline) is connected to the node “NewsDatabroker1” and “NewsClassifer1”.


Update 3: Integrated Training

Use the AI4EU Experiments Platform to train your AI models!

The AI4EU experiments platform is now ready to allow training of your AI models. For this, you need firstly a model that supports a training method and secondly a special training data broker which feeds the training data to the model.

As an example, let’s have a look at the model for classifying news articles and discover how to train it. For the training, the pipeline designer chose the node “newstrainer1” and connected it to the model “newsclassifer1”. For diagnostic purposes, a tensorboard can be added to the pipeline. Once again, a shared folder is used to feed data to the news classifier model, the training node and the tensorboard. Then, the training pipeline for the news classifier model is complete and ready for deployment.

AI4EU Experiments Training Pipeline
Snapshot of the deployment process for the completed training pipeline.


We encourage and welcome the AI4EU developer community to discover and test the new features of the AI4EU Experiments platform’s summer 2021 release! We are looking forward to getting your feedback on the new additions. And of course, we are always open for ideas and suggestions for future updates and new developments on our platform.

For more technical insights and guidance on how to make the most of the new AI4EU Experiments platform’s features, please refer to the video tutorial made by Martin Welss.

To the AI4EU Experiments platform

Edited by
Inga Daase
Published on
10.08.2021
Source
AI4EU – by Martin Welss & Inga Daase