Research topics

Our portfolio for the complete pipeline for AI supported data analysis

Researcher Theresa Neubauer points to a computer screen showing her research work on artificial intelligence. A man listens attentively.
VRVis researcher Theresa Neubauer develops AI-based image processing methods for materials science and industry.

Data preparation

The quality of training data in the context of each application is decisive for the quality of the result. Our experience and the use of elementary data science methods, as well as tools that simplify data annotation by experts, are the basis for preparing a sufficient volume of big data and compiling adequate training sets.

An image collage depicting the data pipeline of how training data for artificial intelligence is produced, using the example of meteroid shatter cones placed on the surface of Mars.
Good training data for artificial intelligence algorithms is often in short supply - especially when it comes to extraterrestrial data. This figure shows the pipeline of how such data is generated.

Generation of training data for AI systems

AI solutions are only as good as the data with which they are trained. But good training data is often scarce, sometimes not even available at all or difficult/expensive to obtain. We support our company partners by artificially simulating the necessary data to train neural networks. Synthetic data is secure, anonymized and unbureaucratic. The artificially generated training data sets are characterized by very good data quality and thus help to accelerate innovation processes, for example by greatly reducing time-to-market due to their rapid availability.

Image collage showing the individual steps of how sidewalks can be "extracted" and recognized from the data of a point cloud with the help of segmentation and 3D reconstruction.
The data pipeline shows how sidewalks and sidewalk edges are detected from large point clouds using segmentation and 3D reconstruction.

Development of tailor-made AI solutions

We develop deep learning and machine learning technologies in close cooperation with our partners and customers in order to design targeted solutions for processing, analyzing and visualizing large volumes of data, multi-terabyte laser scan data and for analyzing images.

Ein Diagramm, dass den Ablauf einer Künstliche Intelligenz-Pipeline anzeigt, die vertrauenswürdig ist.
VRVis develops methods that make the basis for the decisions of a neural network visible and thus comprehensible (explainable artificial intelligence).

Explainable Artificial Intelligence (XAI)

Decisions that are made by machines must remain intelligible, reliable and transparent for humans, particularly in sensitive areas such as medicine or in self-driving vehicles ("ethical AI"). We develop methods that make machine behaviour explainable and consequently provide transparency for complex decisions that are made with the aid of artificial intelligence. Our methods support the evaluation of security and stability of trained networks.

Areas of application

Our research centre has already produced several AI-based applications and patents, as well as high-impact publications in the fields of image analysis and data science, and the areas of application of digital radiology, neuroscience, life science,material science and manufacturing, and digital agriculture.

Publications

Projects