Smart Data & Knowledge Services

In the field of pattern recognition, we examine data of different nature for similarities, repetitions, and regularities to make them practically usable. A characteristics of pattern recognition is the addressed wide field of application with different data formats and objectives, which is covered by a large spectrum of methods and approaches.

The objectives of pattern recognition are in the automatic transfer of recognition results into standardized formats, the prediction of properties, states and temporal developments, the detection of anomalies/problems as well as the optimization of control systems and medical treatments.

The application fields we consider include business documents, technical drawings, medical protocols and data (e.g., gene sequences and image material), measured values in complex technical systems (e.g., in vehicles, production machines, and energy supply), and cyber communities.

Our specific competencies are:

1. Hybrid AI: In many application fields of pattern recognition, there is already established expert knowledge. This is used by us to improve Deep Learning based analysis methods. By combining expert knowledge and Deep Learning, we improve the recognition accuracy of Deep Learning algorithms, especially for smaller training sets.
2. Explainable AI: We extend Deep Learning with explanatory components - an indispensability for the acceptance of the results of a machine algorithm.
3. Privacy Preserving Machine Learning: We aim to exploit the potential of big data for learning without compromising the privacy rights of contributing individuals.

Our vision is to harness the potential of Deep Learning for intelligent systems that incorporate expert human knowledge, make decisions that are comprehensible to humans, and protect the privacy of each individual.

Current work in our team involves:

• Hybrid Artificial Intelligence for Automated Driving.
• Predictive maintenance and forecasting
• Anomaly detection
• Analysis of medical images (genome analysis, cell analysis)
• Explanation components for time series analysis and medical applications
• Monitoring of sensor networks for analysis and control
• Privacy Preserving Machine Learning
• Ethics and fairness in artificial intelligence
• Federated Learning
• Few Shots Learning
• Document analysis in the context of business processes
• Table recognition and analysis
• Analysis of technical drawings/schedules
• Automated analysis, monitoring and profiling in scientific cyber communities