Prof. Dr. Orsolya Csiszár
Orsolya Csiszár received her Master's degree in mathematics from Eötvös Loránd University and her PhD in applied mathematics and computer science from Óbuda University, Budapest in 2016. She is currently Professor in Mathematics and Applied Artificial Intelligence at Aalen University.
Her main research interest lies in developing explainable artificial intelligence methods for applications in complex systems. Her research group focuses on combining machine learning with existing knowledge, such as traditional rule-based approaches (neurosymbolic AI) and mathematical physics models (physics-informed computational methods).
AI techniques, especially deep learning models, are revolutionizing the business and technology world. One of today's greatest challenges in deep learning is the increasing need to address the problem of interpretability and to improve model transparency, performance, and safety (XAI: eXplainable Artificial Intelligence). Complexity is by its nature accompanied by a lack of transparency. Without understanding why an AI model delivered a specific result, one can misinterpret its results and their reliability, attribute cause-effect relationships incorrectly, and ultimately reach wrong conclusions. Combining neural networks with continuous logic and multi-criteria decision-making tools can contribute to better interpretability, transparency, and safety. This approach, together with other evolving methods belongs to neuro-symbolic hybrid artificial intelligence; a novel area of AI research that combines traditional rule-based approaches with modern deep learning techniques. Neuro-symbolic models have been shown to obtain high accuracy with significantly less training data than traditional models. Neural networks and symbolic systems can complement each other’s strengths and weaknesses, enabling systems that are accurate, sample efficient, and interpretable.
In machine learning (ML), an artificial system learns from patterns and relationships in data. The training of ML models usually requires large amounts of data. These are, however, often not available. One way of solving this problem is to combine machine learning with existing knowledge. A physics-informed machine learning approach integrates data and mathematical physics models, even in partially understood, uncertain, and high-dimensional settings. This prior knowledge can narrow down the search space considerably, making training ML models more efficient. As a result, more accurate and robust predictions can be made with less data.
Modeling, predicting, and controlling complex systems is being revolutionized by datadriven discovery. Empirical models or derivations from first principles are not sufficient for solving the most challenging scientific and engineering problems. In a wide range of complex systems, researchers are increasingly relying on data-driven approaches. These systems are typically high-dimensional, characterized by dominant underlying patterns that need to be modeled for eventual prediction and control. We can now tackle previously insurmountable challenges thanks to modern mathematical methods.