Humans can accomplish a variety of tasks because they have the ability to understand problems, abstract them appropriately, and then combine their knowledge with their experience to develop appropriate solutions. Humans are able to learn from problem solving experience and thus continuously improve themselves, be it by accomplishing tasks better, with fewer mistakes, or faster.
We investigate experience-based learning systems in which we try to transfer comparable mechanisms of problem solving and learning to computer systems. Therefore we combine different AI technologies:
- Semantic technologies as a basis for knowledge representation and for representing the meaning of data and information at different levels of abstraction,
- Machine learning (incl. deep learning) as a methodology for automatic learning from interpreted information,
- Planning and constraint-based methods for automatic reasoning, and
- Case-based reasoning, as the central methodology of experience-based problem solving by analogy.
We are currently investigating experience-based learning systems in three interconnected areas.
The increasing business agility forces many companies and organizations to continuously adapt or improve their processes. The increasing digitalization as well as today's requirements regarding the flexibility of production and service provision require new intelligent approaches to process management based on AI methods. We investigate experience-based learning systems that allow for
- the redesign, adaptation, and optimization of processes,
- the flexible execution of processes by AI-based workflow systems, and
- the analysis of process data with regard to the diagnosis of resources and processes.
Current application fields:
- Industry 4.0: flexible production and predictive maintenance
- Construction: Work Processes in Defect Management in the Civil Engineering
- Agriculture: Application procedure for subsidies
- Medicine: Analysis of clinical processes
- Culinary: Generation of new cooking recipes
- E-Science: Scientific workflows for the analysis of scientific data
In the modern, digital knowledge society, knowledge and experience must increasingly be connected at the global level and made available, assessable, and usable. Knowledge and experience are often available in the form of extensive but unstructured data or document collections. From these, the relevant knowledge items must first be (automatically) extracted in order to enable further processing with AI methods. We investigate how experience-based learning systems can understand and analyze knowledge and experience in existing forms, how based on this, efficient search and decision support can be ensured and new relevant knowledge be synthesized.
Current application fields:
- Political Science: Analysis and synthesis of argumentations
- Law: Decision support in data protection law
- Business documents: Information extraction and process integration
- Service: Analysis of service reports for search and decision support
We demonstrate and evaluate our concepts and theories through the development of prototypes and demonstrators, which are used for experimental studies and practical trials. For this purpose, the ProCAKE Framework (Process-Oriented Case-Based Knowledge Engine) is being be developed and continuously improved. ProCAKE is a generic system for the realization of experience-based learning systems for process and knowledge management.
For our research we have an excellent environment in which we can work efficiently and professionally. This includes
- Our research is embedded in the research area Smart Data & Knowledge Services of the German Research Center for Artificial Intelligence (DFKI).
- Center for Informatics Research (CIRT): Organizational environment for interdisciplinary cooperation between computer scientists and scientists from other departments of the University of Trier.
- IOT Lab for the practical research in the context of Industry 4.0 and for the demonstration of our solutions
- Up-to-date hardware equipment with powerful workstations, a productive server infrastructure, and a dedicated high-performance GPU compute server with 8 NVidia Tesla V100 GPU graphic cards for efficient machine learning.
Here you find our list of publications.