Sebastian Raubitzek, researcher at SBA research within CORE research group, will give a lecture on “Fraktale Aktivierungsfunktionen in neuronalen Netzen”.
© Niklas Schnaubelt
Abstract
This lecture addresses the question of how ideas from fractal geometry can be used in the design of neural networks. At its core is a class of activation functions inspired by simple fractal constructions that introduce structure across many scales into otherwise classical models. The aim is not to make networks larger or more complicated, but to deliberately expand how they respond to inputs.
Through illustrative examples and intuitive visualizations, the talk shows how non-smooth, self-similar activation functions differ from common standard functions and why these differences may be relevant for learning processes. Experiments on established benchmark datasets demonstrate that this approach can achieve competitive results while also revealing new geometric properties of neural representations.
Overall, the lecture presents a perspective on how concepts from mathematics and physics can open alternative design approaches in machine learning and why fractal structures could become a meaningful building block for future neural models.
Speaker
Sebastian Raubitzek is a researcher at SBA Research in Vienna, Austria, and a member of the research group “Complexity and Resilience.” He holds a Master of Science in Theoretical Physics from the University of Graz and earned his PhD in Computer Science at TU Wien, focusing on the integration of chaos theory and complexity research with artificial intelligence.
His research areas include complexity science, artificial intelligence, theoretical and quantum physics, machine learning, and chaos. Raubitzek has published work on topics such as neural networks, machine learning applications, time series analysis, quantum-inspired methods, and data obfuscation techniques related to privacy. He is the author of numerous peer-reviewed articles and preprints in these fields.
About the Event
The Social Artificial Intelligence Night (SAINT) of the Department of Computer Science & Security is a highlight and an important meeting point for researchers, industry partners, and enthusiasts to exchange ideas about the latest breakthroughs and forward-looking projects in the field of artificial intelligence. Look forward to engaging presentations, inspiring posters, and a lively partner exhibition.
Further information
CORE – Complexity and Resilience Group
Program
Registration for SAINT 2026