Abstract/Summary

This article presents our work in progress in supporting automated machine learning in the model-driven engineering process of AI-enabled software systems. We argue that the state of practice suffers from two key issues. First, data scientists often follow a trial-and-error process and use certain heuristics to practice machine learning engineering. Therefore, their results are typically far from optimized as we show through an example in this study. Second, software engineers without deep knowledge of machine learning are often required to collaborate with data scientists, integrate and maintain their code, or even take over their tasks due to a general shortage of data scientists worldwide. Hence, there is an urgent need for tools that can support these novice machine learning practitioners. To address the mentioned issues, we deploy the model-driven engineering paradigm and enable automated machine learning in an existing software development methodology and tool that supports this paradigm.