Artificial Intelligence Methods
In artificial intelligence, we design software systems that exhibit some form of "intelligent behavior" -- ranging from perceiving the system's environment to making effective decisions in a rational and autonomous manner without human intervention. Traditionally, capabilities associated with intelligence were examined such as language understanding, (logical) reasoning, learning, and decision-making. The ultimate goal is both to incorporate existing knowledge ("world models") into these software agents as well as to equip them with the ability to learn from data they perceive and thus continuously improve. Our group "Artificial Intelligence Methods" particularly focuses on techniques that allow us to address commercial and industrial use cases of artificial intelligence such as recommender systems for mechanical design, learning parameter settings for CFRP (carbon fiber reinforced polymer) applications from simulations and observations, and optimization according to preferences, e.g. in smart energy systems. Achieving these goals requires a blend of machine learning and constraint optimization techniques that we actively research as our two central pillars. We furthermore investigate how to make foundational techniques in AI accessible by means of developing "engineering practices" and domain-specific languages.
Machine Learning for CFRP
Carbon fiber reinforced polymer production (CFRP) is faced with a high inherent variety in the input materials. We apply machine learning to provide quality assurance and process control for, e.g., resin transfer moulding processes.
Due to the lack of a large volume of data in industrial applications of machine learning, we combine simulated and real data by means of transfer learning.
Intelligent Process Automation
Many recurring tasks during manufacturing can be automated and supported using data-driven methods. We investigate novel approaches to analyzing artifacts from construction.
Learning for Industry 4.0
We investigate time-series based approaches to classification and regression that are crucial for Industry 4.0 use cases such as condition monitoring, predictive maintenance, and predictive control.
Discrete optimization problems can often be formalized and modeled on a high level before sending them to concrete solvers. The group maintains a modeling language for soft constraints.
Leveraging machine learning for the efficient high-volume production of carbon-fiber reinforced lightweight plastic components
We use machine learning to predict errors during cutting and machining in industrial contexts.
Dr. Nando de Freitas, Lead Scientist at Google DeepMind, Former Professor at the University of Oxford
- Design, development, and optimization of machine learning systems
- Joint research projects in novel application areas of AI
- Introductory lectures on artificial intelligence and machine learning
- Constraint-based optimization including soft constraints
Institute for Software & Systems Engineering
The Institute for Software & Systems Engineering (ISSE), directed by Prof. Dr. Wolfgang Reif, is a scientific institution within the Faculty of Applied Computer Science of the University of Augsburg. In research, the institute supports both fundamental and application-oriented research in all areas of software and systems engineering. In teaching, the institute facilitates the further development of the faculty's and university's relevant course offerings.