Knowledge snack: Efficient AI training for quality monitoring using active learning

Active learning (AL) is a machine learning approach that significantly reduces the amount of training data required while maintaining or even improving the performance of the AI model. During the learning process, an AI model selects the data points that have the highest information content. These are then labelled by a human. The targeted labelling approach significantly reduces the manual effort and therefore also the costs for creating a model.

In this knowledge snack, we will give you a compact overview of the core principles of active learning. You will also learn about the latest AL strategies for image and time series data sets. Using two practical examples - wood defect detection and tool wear detection on milling machines - we will show how active learning can rationalise data annotation and improve model accuracy at the same time. Finally, we present a demonstrator that iteratively trains AI models from scratch using active learning, illustrating the efficiency and impact of this workflow.

Speaker: Akshaya Bindu Gowri, Data Analysis Systems at Fraunhofer IIS in Dresden

More information