Automated Machine Learning (AutoML) - Efficient search for optimal ML pipelines and models.

Finding the optimal machine learning pipeline, suitable model or fine-tuning hyperparameters can be time-consuming and often requires many attempts - and may even involve errors. This is where automated machine learning (AutoML) comes in, automating this process and saving ML experts valuable resources.

AutoML uses intelligent search budgets such as successive bisection or Bayesian optimisation to find optimal parameter combinations. It automates the search for the best model architecture, the optimal hyperparameters and the most efficient ML pipeline for a given task. A variety of techniques and algorithms are used to effectively search the search space and obtain the best results.

AutoML strategy: Intelligent allocation of search budgets

Compare your approach in our game "Beat the ML" (random search) with the strategy "Successive halving". If you play for an hour, you could try many combinations. AutoML would instead start thousands of attempts and then finish the worst 50% every 15 minutes. In the end, this strategy finds a solution that is close to the optimum.

One of the challenges in AutoML research is to find the right balance between automation and the involvement of domain or ML experts. This human-in-the-loop approach ensures that expert knowledge is integrated into the automation process. It allows experts to contribute their expertise in defining the objective function, selecting relevant features or interpreting the results.

The advantages of AutoML are multiple

resource savings: AutoML significantly reduces the effort required to manually search for optimal ML pipelines or models. The automated search saves ML experts valuable time and resources that can instead be used for other important tasks.

efficient search: By using intelligent search budgets and optimisation algorithms, AutoML enables an efficient search in the vast space of possible models and hyperparameter combinations. As a result, promising solutions are identified more quickly.

improved performance: AutoML can help improve model performance by identifying more optimal configurations. Through systematic search, models are found with higher accuracy and better fit to the task.

Inclusion of expert knowledge: The human-in-the-loop approach allows domain and ML experts to bring their expertise into the automation process. This allows specific requirements, constraints or expertise to be taken into account to achieve optimal results.



AutoML at ADA Lovelace Center

Automatic and adaptive learning (AutoML) deals with the automation of the AI process and of particularly labour-intensive, manual tasks that are usually performed by experts. AutoML covers a large area, starting with the automation of feature detection and selection for given data sets as well as model search and optimisation, continuing with their automated evaluation and ending with the adaptive adjustment of models through training data and system feedback.


AutoML (only german)

Automatisiertes Maschinelles Lernen (AutoML) genießt derzeit viel Aufmerksamkeit, da es verspricht die Entwicklung und Konfiguration von KI Prozessen zu automatisieren. Gemeinsam mit unserem Kunden aus dem Bereich der industriellen Fertigung haben wir deshalb untersucht, welche spezifischen Anpassungen für den Einsatz von AutoML Systemen in der praktischen Anwendung im Unternehmen sinnvoll sind.


ML für mehr Effizienz in der industriellen Qualitätssicherung (only german)

Im industriellen Kontext wird Maschinelles Lernen (ML) immer wichtiger – insbesondere in der Qualitätssicherung. Diese ist für Unternehmen oft mit viel Aufwand verbunden, vor allem wenn strenge Fehlertoleranzvorgaben eingehalten werden müssen.