AI4EU project developer Martin Welss is working on a technical AI platform for hosting AI solutions. He presented the results of his work in his talk “A Perfect Match: AI4EU and Acumos for Europe” at the “Open Source Summit Europe” (OSS Europe). The annual conference was held from October 28 to 30, 2019 in Lyon, France. In his presentation, Mr. Welss highlighted the link between Linux Foundation’s Acumos project and the AI4EU platform.

The conference talk aimed at introducing the audience to the EU-funded AI4EU project, presenting the first project results and gathering feedback from the community. Within the framework of the AI4EU project, Mr. Welss and a team of developers are working on designing the AI4EU Experiments Platform. This platform is being implemented as a fine-tuned Acumos instance, allowing users to onboard algorithms, machine learning models as well as datasets. Moreover, the AI4EU Experiments Platform will enable users to train models and build their own solutions in the form of AI pipelines.

To make the creation of these pipelines intuitive and easy, the platform offers a design studio with a graphical user interface where data and models can be combined. The design studio contains a pool of AI resources: pipeline developers can select suitable models, data sources as well as data transformation tools, such as splitters or mappers. For models to be included into an AI pipeline, they have to be onboarded in the form of dockerized containers with protobuf specification. Each model is then deployed in its own docker container as a microservice.

In his presentation, Mr. Welss gave several examples for problem solving with the help of AI pipelines. The identification of individual speakers who appear in video material is an example for such a task. Developers obtain the best results when they separate visual and audio data by a splitter, analyze both data formats individually and then combine the results by a storage collator.

The AI4EU technical platform will interconnect with existing platforms such as Beat, which provides a means of comparing scientific data related to AI, Mundi, which gives access to satellite data, the European Language Grid, the International Data Spaces and the Human Brain Project. Through this connection, platform users will get access to diverse and high-quality datasets and models which they can use as building blocks for their own AI solutions.

Mr. Welss, who is an experienced application developer working as a researcher at Fraunhofer IAIS, is responsible for designing the interface between Acumos and the AI4EU platform. The collaboration already led to significant additions to the Acumos project, such as the onboarding of C++ models. On the other hand, the AI4EU platform benefits from the Acumos project as it provides the technical foundation, including a state-of-the-art graphical user interface and services to manage AI solutions.