Patrik Reizinger

The main motivation for my research is to advance our understanding of how and why deep learning models work. My research toolkit currently focuses around identifiable causal and self-supervised representation learning and out-of-distribution (OOD) generalization, with a focus on compositionality in language models. During my Ph.D., I realized that current machine learning theory is insufficient to explain especially the interesting and useful properties of deep neural networks. I aim to help close this gap, by focusing on: 1) extending machine learning theory to understand the role of inductive biases (e.g., model architecture or optimization algorithm); 2) grounding machine learning in the physical world via (causal) principles and humanity’s prior knowledge; 3) extending our understanding of out-of-distribution and compositional generalization; 4) uncovering overarching patterns across different fields in machine learning. I have done both my M.Sc. and B.Sc. at the Budapest University of Technology in electrical engineering and specialized in control engineering and intelligent systems. In my free time, I enjoy being outdoors and often bring my camera with me.