Jacob Nielsen

Jacob.jpg

Odense, Denmark

I am a computer scientist from the University of Southern Denmark (SDU), specializing in artificial intelligence and machine learning. Currently, I am employed as an industrial PhD student at Ordbogen A/S and SDU, supervised by Peter Schneider-Kamp, Lukas Galke and Melih Kandemir. I primarily work with efficient deep learning in all aspects from training to inference. I am interested in many aspects and keep getting inspired, but examples are extreme quantization aware training and inference, distributed training optimization, optimizers and understanding and using latent spaces in the optimization process.

news

May 15, 2025 🚀 Continual Quantization-Aware Pre-Training: When to transition from 16-bit to 1.58-bit pre-training for BitNet language models?, has been accepted to the ACL 2025 Findings!
Feb 01, 2025 I officially started as an industrial PhD at Ordbogen A/S and University of Southern Denmark.
Jan 24, 2024 I successfully defended my masters thesis titled: “Object Detection with Transformers from A Drone Perspective” graded 12 on the Danish grading scale (ECTS mark: A)

selected publications

  1. CVPR
    Multiview Aerial Visual Recognition (MAVREC): Can Multi-view Improve Aerial Visual Perception?
    Aritra Dutta, Srijan Das, Jacob Nielsen, and 2 more authors
    In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024
  2. DeLTA
    BitNet b1. 58 Reloaded: State-of-the-art Performance Also on Smaller Networks
    Jacob Nielsen, and Peter Schneider-Kamp
    In International Conference on Deep Learning Theory and Applications, 2024
  3. ICAART
    When are 1.58 bits enough? A Bottom-up Exploration of BitNet Quantization
    Jacob Nielsen, Lukas Galke, and Peter Schneider-Kamp
    In , 2025