Felix Draxler

I’m a Postdoctoral Scholar at UC Irvine. I work on bringing uncertainty to every AI prediction. I achieve this by building flexible, accurate, and fast generative models based on information theoretic principles.

Lately, I co-developed Free-Form Flows, a one-step generative modeling framework compatible with symmetric distributions, distributions on manifolds, and joint autoencoder training. I showed that normalizing flows based on coupling layers are distributional universal approximators and volume preserving generative models are not. Before that, I analyzed the energy landscape and spectral bias of neural networks.

My work is consistently published at top machine learning conferences such as ICLR, ICML, NeurIPS and AISTATS. I was awarded a best paper honorable mention at GCPR 2020. My PhD thesis was selected as one of the best PhD thesis of 2024 by my alma mater, Heidelberg University, Germany.