Albert Kjøller Jacobsen

PhD Student in Geometric Approximate Bayesian Inference

me.png

Section for Cognitive Systems

Dept. of Applied Mathematics and Computer Science

Technical University of Denmark

I am a PhD student at the Technical University of Denmark (DTU), supervised by Georgios Arvanitidis from the Section for Cognitive Systems. My project seeks to advance approximate Bayesian inference methods by leveraging concepts from differential geometry.

I have a strong background in machine learning and hold a Master’s degree in Human-Centered Artificial Intelligence with the thesis focusing on leveraging concepts from differential geometry in optimization for improving generalization performance.

I’m particularly interested in the theoretical aspects of machine learning and deep learning. Currently, I’m focusing on gaining further knowledge on:

  • differental geometry in machine learning
  • approximate Bayesian inference
  • deep learning theory

news

Jun 25, 2025 The Danish Data Science Academy (DDSA) has awarded me a PhD Fellowship grant! This is truly amazing and I can’t wait to start my project on geometric approximate Bayesian inference 🔎
Apr 06, 2025 Just arrived in Hyderabad for ICASSP 2025 :airplane: I’ll present our poster Friday morning!
Mar 01, 2025 I joined Georgios Arvanitidis’ team as a research assistant :sunrise_over_mountains:
Feb 06, 2025 Now officially an engineer! I defended my MSc. Thesis on “Riemannian Sharpness-Aware Minimization for General Losses” :sparkles:

latest posts

selected publications

  1. Preprint
    mongesam.png
    Monge SAM: Robust Reparameterization-Invariant Sharpness-Aware Minimization Based on Loss Geometry
    Albert Kjøller Jacobsen and Georgios Arvanitidis
    arXiv preprint arXiv:2502.08448, 2025
  2. ICASSP
    redundancy.png
    How Redundant Is the Transformer Stack in Speech Representation Models?
    Teresa Dorszewski, Albert Kjøller Jacobsen, Lenka Tětková, and 1 more author
    In ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2025