Andreas Bergmeister

prof_pic.jpg

I am a PhD candidate in AI at TU Munich, in the group of Prof. Dr. Stefanie Jegelka and affiliated with the Munich Center for Machine Learning. I received my Bachelor’s and Master’s degrees in Computer Science from ETH Zurich and interned at the Swiss Data Science Center.

My research is motivated by a long-standing interest in bringing aspects of Kahneman’s System 2 — slow, deliberate, and logical thinking — into AI systems. I aim to make such “slow thinking” more efficient by exploring latent-space reasoning with diffusion and flow-based models, accelerating their training, and aligning them with reward signals to improve reasoning.

Earlier, I worked on geometric deep learning, making graph generation scalable. I also worked on the fundamentals of neural network training, developing PJAX as an alternative to autodiff frameworks for training neural networks without gradients.

selected publications

  1. pjax.png
    A projection-based framework for gradient-free and parallel learning
    Andreas Bergmeister, Manish Krishan Lal, Stefanie Jegelka, and 1 more author
    2025
  2. waveform.png
    High Resolution Seismic Waveform Generation using Denoising Diffusion
    Andreas Bergmeister, Kadek Hendrawan Palgunadi, Andrea Bosisio, and 5 more authors
    2024
  3. graph-generation.png
    Efficient and Scalable Graph Generation through Iterative Local Expansion
    Andreas Bergmeister, Karolis Martinkus, Nathanaël Perraudin, and 1 more author
    ICLR, 2024