SEMLA: Software Engineering for Machine Learning – integrated approach

We have witnessed the rapid development of AI, in particular of machine learning (ML), as the most promising technology of the last decade.

However, building AI applications requires more than just building ML algorithms, which are only a fraction of the total software code. AI functions are components of a broader system.

Developing such AI-enabled systems requires coping with a number of challenges, as the development process of ML components is different from development of traditional software components, and as the ML functions depend on data, which is significantly more challenging to manage than code. Software engineering practices exist for complex systems development, but not for AI-enabled systems.

This calls for the development of "Software Engineering for AI", to provide that support.

This project addresses this problem, by a) providing a new development process that integrates the ML and software development processes, b) building a new component model for AI components that exposes the information specific to AI components, such as the datasets used or the accuracy, and c), by analyzing system consistency using component and system properties and triggering information exchange between the software and ML development process, and providing a seamless continuous evolution of the software system.

The project starts from the needs from the practice, and extends the currant sate of the art. It will evaluate the results in a real-world industrial context.