Methods & Algorithms for AI

Research Focus

The Research Area “Methods and Algorithms for AI” researches and develops a wide range of foundations, methods, and algorithms for trustworthy AI. We combine key technical concepts, and methods such as Machine Learning, Natural Language Processing (NLP), Deep Reinforcement Learning (DLR), and Data Science (DS). Our long-term goal is to develop trustworthy algorithms that are highly advanced. So, that they can automatically derive hypotheses from data from a wide variety of sources while maintaining the highest privacy standards, and can validate these hypotheses with additional data.
Theory-Inspired Machine Learning

Theory-based machine learning achieves domain specialization by incorporating prior knowledge in the selection, parameterization, or training of an ML model. In this research focus, methods are developed to incorporate prior knowledge into ML models or to complement theory-driven models with a data-driven component (hybrid models).

Privacy-Preserving Representations

We explore the less obvious clues for identifying individuals, such as writing style, when determining authorship. This is followed by questions about what constitutes “privacy” in terms of the individual task or application domain, and what techniques are appropriate for protecting privacy. Our research also focuses on representative aspects of Deep Learning.

Domain-Informed Reinforcement Learning

Our research goal is to provide domain-informed methods that make DRL accessible to a wide range of applications, thereby improving the practicality of DRL to unlock its potential. We also investigate how domain knowledge can overcome problems associated with high-dimensional continuous action spaces and limited data sets.