AI for Assistive Health Technologies
The research department AI for Assistive Health Technologies (AGT) develops AI and signal processing methods for the personalization and individualization of health-related measures in prevention, diagnostics, therapy, rehabilitation, and care.
The most important application scenarios are personalized nutrition, sensor- and video-based movement analysis, active assisted living, affective computing, and acoustic event recognition. Beyond the health context, the research work also extends to application domains such as smart homes, smart cities, sales and consumption forecasting, production monitoring, predictive maintenance, and process optimization.
The focus is on the development of AI methods that learn complex relationships in a data-driven way and thus overcome the limitations of classical mathematical models.
The work addresses the following research problems in AI and signal processing:
- combining classical signal processing approaches with learning-based AI methods
- automated methods for the artificial generation of training data (data augmentation)
- minimizing bias effects when training AI models
- transfer of pre-trained AI models to related application scenarios (transfer learning)
- resolving unwanted interdependencies between features (disentanglement)
- explainability and interpretability of complex AI algorithms
- adaptive adjustment of AI models to mitigate the effects of data distribution shifts
- optimization of AI approaches for applications in resource-limited environments (edge AI)
The research department AI for Assistive Health Technologies operates in close collaboration with industrial companies and healthcare institutions. Its aim is to assist these entities in improving their processes, services, and products through AI-based analyses and predictions. Under the guidance of Prof. Grzegorzek, the team adheres to the MLOps principles (Machine Learning Operations). They combine aspects from the fields of sensor technology, signal processing, data engineering, machine learning, and software engineering. This comprehensive approach ensures that the solutions they develop are efficiently and sustainably transferred into commercial applications.
