Reuse and Automation for Machine Learning in the Context of MLOps

May 01, 2021 to Apr. 30, 2024

Artificial intelligence (AI) and machine learning (ML) are increasingly becoming the basis for intelligent services (smart services), such as predictive maintenance, or intelligent software (smart software), such as for route planning in road traffic. For such new applications to be developed, AI- and ML-based software must be able to be produced, operated and maintained with similar efficiency and quality as "classic" software. Suitable methods, tools and processes are needed for this.

Analogous to "classic" software, AI-based software must be implemented and validated according to the end user's requirements. It must integrate with other software and communication technologies and meet all the established quality characteristics of "classic" software (e.g., functionality, security, maintainability, interoperability) as well as several new quality characteristics (e.g., integrability, intelligent behavior, ethics, etc.). Their use must be technologically, socially, and ethically acceptable and safe. These characteristics must be carefully planned, realized, validated, and maintained throughout the software lifecycle.

The IML4E project has therefore brought together companies from the main sectors of the European software industry to develop a European framework for the development, operation and maintenance of AI-based software to ensure the development of intelligent services and intelligent software on an industrial scale.

"With the IML4E project, we want to strengthen the European software industry to be able to efficiently apply advantages of Artificial Intelligence and Machine Learning without having to take unnecessary quality risks."

SQC, Bild, IML4E, 210708
The IML4E project at a glance Fraunhofer FOKUS

Fraunhofer FOKUS is leading the overall European project IML4E. In the project, the scientists develop techniques, tools and methods to systematically document and model the uncertainty and risks in machine learning due to data quality problems. For this purpose, quality features and risk-based testing approaches will be developed that are suitable for continuous quality monitoring throughout the lifecycle of AI-based software. In another work package, methods and techniques that can be used to establish transparency and comprehensibility of machine learning models will be developed. Fraunhofer FOKUS is developing a risk-based quality assurance approach with which machine learning software can be tested for properties such as security and robustness and can thus also be used in critical application areas. In addition, a teaching and experimentation platform, as well as suitable teaching and training material for industrial users, is being created.

In the ITEA3 project IML4E, 18 partners from Finland, Hungary and Germany are working together. The project is funded by the German Federal Ministry of Education and Research (BMBF).