Laurits Fredsgaard Larsen

About me

I’m a PhD student in Applied Mathematics and Computer Science at the Technical University of Denmark, in the Cognitive Systems section. My research is part of the Bayesian neural networks for molecular discovery project, supervised by Mikkel N. Schmidt and Michael Riis Andersen.

What I enjoy most is developing solutions to complex problems. That might mean investigating how to properly train deep ensembles (TMLR 2025), or how to train them efficiently on distributed hardware (torch-distributed-ensemble). It might mean comprehensively benchmarking which uncertainty quantification methods work best for graph neural networks, so that molecular discovery pipelines can support reliable lab-in-a-loop decision making. These problems usually get interesting at scale, where methods that look clean on toy benchmarks often fall apart.

From October 2025 to March 2026, I was a visiting researcher at the National Institute of Informatics (NII) in Tokyo, hosted by Prof. Mahito Sugiyama, where I worked on calibration of autoregressive graph generators.

What’s next

I’ll hand in my PhD thesis in mid-September 2026 and am looking for hybrid research-applied ML roles in Copenhagen. I’m most interested in teams doing cutting-edge work where the methods have to deliver. If that sounds like your team, please get in touch.

Publications

• Semi-Supervised Learning for Molecular Graphs via Ensemble Consensus
  Rasmus H. Tirsgaard, Laurits Fredsgaard, Marisa Wodrich, Mikkel Jordahn, Mikkel N. Schmidt
  International Conference on Machine Learning (ICML), 2026.
  Earlier versions presented at:

  • NeurIPS 2025 Workshop: AI for Accelerated Materials Design (AI4Mat) (Poster). OpenReview | PDF
  • ELLIS 2025 Workshop: ML4Molecules (Oral & Poster).
  • EurIPS 2025 Workshop: SIMBIOCHEM (Poster).

• On Joint Regularization and Calibration in Deep Ensembles
  Laurits Fredsgaard, Mikkel N. Schmidt
  Transactions on Machine Learning Research (TMLR), 2025. arXiv | OpenReview | PDF

• Capsid-like particles decorated with the SARS-CoV-2 receptor-binding domain elicit strong virus neutralization activity
  C. Fougeroux, …, Laurits Fredsgaard et al.
  Nature Communications, 2021. Article

• Head-to-head comparison of modular vaccines developed using different capsid virus-like particle backbones and antigen conjugation systems
  Laurits Fredsgaard, Louise Goksøyr, Susan Thrane, Kara-Lee Aves, Thor G. Theander, Adam F. Sander
  MDPI Vaccines, 2021. Article

Workshops & Preprints

• Same Graph, Different Likelihoods: Calibration of Autoregressive Graph Generators via Permutation-Equivalent Encodings
  Laurits Fredsgaard, Aaron Thomas, Michael Riis Andersen, Mikkel N. Schmidt, Mahito Sugiyama.
  arXiv | Code
  Accepted at:
  • AISTATS 2026 Workshop: Calibration for Modern AI (Poster). Workshop Website

Background

I came to ML from a life-science background, holding degrees in both Molecular Biomedicine and Machine Learning & Data Science from the University of Copenhagen.

My earlier research was in virology and vaccine design. I studied the Hepatitis C Virus at Osaka University’s RIMD under Yoshiharu Matsuura (2017) and for my bachelor’s project at Hvidovre Hospital with Jens Bukh (2018). My master’s thesis (2018-2020) in Adam Sander’s lab supported preclinical development of the ABNCoV2 COVID-19 vaccine, which has since proven highly successful in clinical trials.

I shifted toward computation as a student bioinformatician at LEO Pharma, where I built data analysis pipelines for omics data. For my final project in Machine Learning (2022-2023), I worked with Wouter Boomsma on deep learning models for protein sequences and structures.