Sustainable AutoML for Engineering

Automated Machine Learning automates important design decisions in the machine learning workflow, such as preprocessing, model selection, hyperparameter optimization, and evaluation. Sustainable AutoML extends this perspective by taking resource consumption and practical deployment constraints into account.

Instead of optimizing only for predictive performance, Sustainable AutoML considers multiple objectives, such as:

• model accuracy,
• computational cost,
• energy consumption,
• training time,
• data efficiency,
• robustness,
• interpretability,
• and practical usability in engineering workflows.

This allows us to develop machine learning solutions that are not only accurate, but also efficient, transparent, and suitable for real-world engineering applications.

Engineering problems often differ from standard benchmark machine learning tasks. Data can be noisy, scarce, expensive to generate, or collected from complex physical processes. In many cases, engineers also need to understand why a model makes a certain prediction, because decisions may affect safety, product quality, energy usage, or material costs.

Sustainable AutoML helps engineers by reducing the need for manual trial-and-error and by making the model development process more systematic. It can support faster experimentation, better use of limited resources, and more reproducible decision-making.

For example, in process engineering or materials science, AutoML can help identify which process parameters influence the final product quality. In energy systems, it can support optimization under efficiency and sustainability constraints. In manufacturing, it can help detect quality issues or optimize production processes while minimizing waste.

Our vision is to develop AutoML methods that support engineers in building better models with fewer resources. By combining automated optimization, explainability, domain knowledge, and sustainability-aware objectives, we aim to make machine learning more practical, transparent, and responsible for engineering applications.

Sustainable AutoML for Engineering is therefore not only about automating machine learning. It is about creating machine learning systems that are efficient, understandable, and aligned with real-world engineering constraints.