SkafteNicki / README.md

Hi, my name is Nicki and I do machine learning research. I am an Associate Professor in the Section for Cognitive Systems at Technical University of Denmark (DTU). I consider myself a jack of all trades, master of none in machine learning and deep learning, having experience with computer vision, NLP, supervised models, generative models, and reinforcement learning. Current research interests are in the intersection of efficient machine learning, designing of machine learning systems, and machine learning operations (MLOps) in general. Furthermore, I am an active open-source developer, contributing to frameworks that help researchers scale their models and experiments. Feel free to contact me if you want to work with me.

Highlights

02476 Machine Learning Operations

course

A course introducing students to coding practices for organizing, scaling, monitoring, and deploying machine learning models in research and production settings.

TorchMetrics

Python

Measuring reproducibility in PyTorch — a library with 100+ machine learning metrics.

PyTorch Lightning

Python

The lightweight PyTorch wrapper for high-performance AI research. Scale your models, not the boilerplate.

libcpab

Python

Library for diffeomorphic transformations in numpy, TensorFlow, and PyTorch. Supports 1D, 2D, and 3D data.

News

2026-06
Looking for a Postdoc

Looking for a postdoc within AI systems engineering apply here in relation to the first moonshot project at CAISA which I co-lead. Part engineering, part research and a good chunk of breaking things and learning from them.

2026-06
Two papers accepted at GLOW

Two papers accepted at the 4th International Workshop on Generalizing from Limited Resources in the Open World (GLOW)

2026-03
Joining CAISA as Senior Scientist

Started as Senior Scientist at the National Center for AI in Society (CAISA), co-leading a moonshot project on designing and evaluating AI for responsible use in the public sector.

Talks

Collage of Nicki Skafte Detlefsen speaking at conferences and teaching
2026-06-04
Coding with Agents PDF

DTU Biosustain A inspirational talk to my colleagues at DTU Biosustain on how to get started with coding agents

2026-05-28
Agentic mindset for startups PDF

Young Entrepreneur Researcher Bootcamp A 4 hour workshop at the Young Entrepreneur Researcher Bootcamp on how to use Agents to accelerate your startup. A good chunk of the time was spendt on hands-on 'vibe coding' a prototype of an startup idea.

2026-05-13
Fra prototype til product PDF

AIBoost accelerator workshop for SMVs A follow-up workshop for the AIBoost accelerator program for small and medium sized companies on how to take a vibe coded prototype and turn it into a product and all the problems and pitfalls that can be encountered along the way. Also provided hands-on experience with coding agents and vibe coding

2026-05-11
Fra ide til prototype PDF

AIBoost accelerator workshop for SMVs A 3 hour workshop for the AIBoost accelerator program for small and medium sized companies on how vibe coding/agentic engineering can help you go from idea to prototype in short amount of time. Also provided hands-on experience with coding agents and vibe coding

2026-02-11
Can we use LLMs to evaluating our students? PDF

DTU cogsys wednesday presentation A small 10 min talk for my colleagues in the section I work at, on a experiment we had run in course 02476 on providing feedback and evaluating the students using LLMs.

2026-01-28
Introduktion til MLOps og dens anvendelse PDF

EuroCC DK NEXT AI workshop on Data, AI factories and supercomputers A introduction presentation on how to apply MLOps in practice for HPC systems and supercomputers.

2026-01-23
MLOps as a tool in the era of AI bullshit PDF

DTU course Innovation in Engineering A small inspirational talk to students on the innovation in engineering course at DTU on how to use MLOps principles as a tool on solving their course case.

2025-11-05
From code to conscience PDF

Digital Tech Summit 2025 A talk in collaboration with my colleague Sneha Das on how how MLOps principles can be used to build AI systems that support responsible and trustworthy AI.

2025-06-04
HPC at DTU PDF

DDSA event: Computer Power in Research - Be inspired and get started A talk on my experience using many different HPC installations over the years, but locally at DTU but also at national and international HPC installations.

2025-05-27
Introduction to MLOps PDF

Young Entrepreneur Researcher Bootcamp A 3 hour workshop at the Young Entrepreneur Researcher Bootcamp on what MLOps is, what to consider when implementing ML in practice and related topics. Included a live demo session of how it works in practices

2025-05-15
AI-udvikling - fra design til virkelighed, det lange seje træk PDF

AIBoost and EuroCC DK workshop for SMVs A presentation on the challenges and pitfalls of developing AI systems in practice.

2025-01-27
Introduction to ML-Ops and its application PDF

AIBoost accelerator workshop for SMVs A 3 hour workshop on getting started with MLOps principles and how to apply them in practice. The workshop included group discussion involving a specially made MLOps canvas for the participants to use in their own projects and a live demo session of how MLOps works in practice.

2024-09-26
Introduktion til AI og MLOps PDF

AIBoost accelerator workshop for SMVs A 3 hour workshop on getting started with MLOps principles and how to apply them in practice. The workshop included group discussion involving a specially made MLOps canvas for the participants to use in their own projects and a live demo session of how MLOps works in practice.

Publications

12 h-index
12 i10-index
1,114 Citations
1,063 Last 5 years

2026

Exploring different approaches to customize language models for domain-specific text-to-code generation

Luís Freire, Fernanda A Andaló, Nicki Skafte Detlefsen

arXiv preprint arXiv:2603.16526 preprint 1
Decoding Depression: AI-EEG Integration for Efficient Depression Diagnosis

Bekarys Gabdrakhimov, Nicki Skafte Detlefsen, Cihan Uyanik, Muhammad Ahmed Khan, Osama Ejaz, Muhammad Abul Hasan, Saad Ahmed Qazi, Matteo Saibene, Sadasivan Puthusserypady

2026 48th Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC) conference

2025

Towards a European HPC/AI ecosystem: a community-driven report

Petr Taborsky, Iacopo Colonnelli, Krzysztof Kurowski, Rakesh Sarma, Niels Henrik Pontoppidan, Branislav Jansík, Nicki Skafte Detlefsen, Jens Egholm Pedersen, Rasmus Larsen, Lars Kai Hansen

Procedia Computer Science, Volume 255, Pages 140-149 conference 7

2024

Seeded LoRA: Collaborative Fine-Tuning Through Seed Initialization of Adapters

Alejandro R. Salamanca, Ahmet Üstün, Nicki Skafte Detlefsen, Tim Dettmers

Efficient Systems for Foundation Models workshop at ICML 2024 workshop 2

2022

TorchMetrics – Measuring Reproducibility in PyTorch

Nicki Skafte Detlefsen, Jiri Borovec, Justus Schock, Ananya Harsh Jha, Teddy Koker, Luca Di Liello, Daniel Stancl, Changsheng Quan, Maxim Grechkin, and William Falcon

Journal of Open Source Software (JOSS), Volume 7, Number 70, Pages 4101 journal 323
Code
What is a meaningful representation of protein sequences?

Nicki Skafte Detlefsen, Søren Hauberg, Wouter Boomsma

Nature Communications, Volume 13, Number 1, Pages 1914 journal 220
Code
Laplacian autoencoders for learning stochastic representations

Marco Miani, Frederik Warburg, Pablo Moreno-Muñoz, Nicki Skafte Detlefsen, Søren Hauberg

Advances in Neural Information Processing Systems (NeurIPS), Volume 35, Pages 21059-21072 conference 20
Code
Is an encoder within reach?

Helene Hauschultz, Rasmus Berg Palm, Nicki Skafte Detlefsen, Andrew Allan du Plessis, Søren Hauberg

arXiv preprint arXiv:2206.01552 preprint 1

2021

Developing and validating COVID-19 adverse outcome risk prediction models from a bi-national European cohort of 5594 patients

Espen Jimenez-Solem, Tonny S Petersen, Christina Lioma, Christian Igel, Wouter Boomsma, Oswin Krause, ... Nicki S. Detlefsen, ... Mads Nielsen, Martin Sillesen

Scientific Reports, Volume 11, Number 1, Pages 3246 journal 105

2020

Lung Segmentation from Chest X-rays using Variational Data Imputation

Raghavendra Selvan, Erik B. Dam, Nicki Skafte Detlefsen, Sofus Rischel, Kaining Sheng, Mads Nielsen, Akshay Pai

ICML Workshop on Learning from Missing Data (Artemiss) 2020 workshop 67
Code
What is a meaningful representation of protein sequences?

Nicki Skafte Detlefsen, Søren Hauberg, Wouter Boomsma

arXiv preprint arXiv:2012.02679 preprint 12
Learning invariant representations from prior knowledge in Deep Learning

Nicki Skafte Detlefsen

PhD Thesis, Technical University of Denmark thesis

2019

Reliable training and estimation of variance networks

Nicki Skafte Detlefsen, Martin Jørgensen, Søren Hauberg

Advances in Neural Information Processing Systems (NeurIPS), Vancouver, Canada conference 87
Code
Explicit Disentanglement of Appearance and Perspective in Generative Models

Nicki Skafte Detlefsen, Søren Hauberg

Advances in Neural Information Processing Systems (NeurIPS), Vancouver, Canada conference 66
Code
Diffeomorphic Temporal Alignment Nets

Ron A. Shapira Weber, Matan Eyal, Nicki Skafte Detlefsen, Oren Shriki, Oren Freifeld

Advances in Neural Information Processing Systems (NeurIPS), Vancouver, Canada conference 61
Code

2018

Deep Diffeomorphic Transformer Networks

Nicki Skafte Detlefsen, Oren Freifeld, Søren Hauberg

IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Salt Lake City, USA conference 70
Code

Teaching

02476

Machine Learning Operations

I am the main lecturer on course 02476 Machine Learning Operations at Technical University of Denmark, having developed most of the material myself. The course introduces students to a number of coding practices that will help them organize, scale, monitor and deploy machine learning models either in a research or production setting. To provide hands-on experience with a number of frameworks, both local and in the cloud, for doing large scale machine learning models.

Course Website DTU Course Page

Censor

I serve as external examiner (censor) for the following Danish censor corps. If you need an examiner for your institution, feel free to reach out.

  • Civilingeniøruddannelsernes Censorkorps
  • Censorkorpset i datalogi

Previously Involved

Summer School on MLOps (MLOPSSS)

Main organizer for the summer school on Machine Learning Operations, bringing together researchers and practitioners in the MLOps field.

Summer School on Advanced ML

Organizer for the DTU summer school on advanced machine learning methods.

Continuing Education (Efteruddannelse)

Machine Learning Operations (MLOps)

Industrial continuing education (efteruddannelse) version of 02476 Machine Learning Operations, adapted for industry professionals.

AI Integration

Continuing education (efteruddannelse) course on integrating AI into organizational workflows and decision-making processes.

Supervision

Feel free to reach out to me if you have an interesting project proposal and are missing a supervisor. I have put together a website with good advice before, during and after writing your thesis. Thesis advice website →

PhD & Postdocs

Rasmus Aagaard
2026 – 2029

Industrial PhD: Edge-deployment of Deep Neural Networks via Model Compression in Healthcare Applications

Students

95 students supervised (10 current, 85 former)

2026 10 current / 12 former
MSc Sebastian Nicolai Fabricius Grut current ML models for improved patient diagnostics on Danish Hospital Data
MSc George Georgiou current An Empirical Study of Aggregation Strategies for LLM-as-a-Judge Evaluation?
MSc Josefien Kam current Cross-Domain Generalization for Speech-Based Dementia Detection Across Languages
MSc Philip Johannes Fog Helsted current Learning Robust Multimodal Representations as a Basis for Pattern Discovery in Repeated User Workflows
MSc Wenji Xie current Enhancing Agentic Capabilities in Task Automation Systems through Post-Training and Orchestration of Small Language Models
MSc Shah Bekhsh current Towards Intelligent Public Assistance: A Framework and Pilot Deployment of Context-Aware Agentic AI For Denmark’s National Well Database Using Model Context Protocol
MSc Morten Lindhardt Helsø current Game Recommendation in Online Casino Sessions Using Deep Learning
MSc Alexander Valentini current Building RAG-Based AI Systems for Scalable Resume Screening using LLM’s
MSc Edrin Molla current Development of an Operationally Usable Data-Driven Performance Model Based on Engine Simulation Data
MSc Oliver Birkmose Broager current Explainability of Speech Embeddings for Cognitive Impairment Detection
MSc Arnór Daníel Moncada former PyPyrus: A Data Provenance Layer for Transparent and Reproducible Machine Learning Systems
MSc Magnus Søndergaard Vinjebo former An Exploration of End-to-End Machine Learning Deployment and Maintenance under Drift
MSc Benedicte Maul Vilhjálmsson · Emilie Leonora Wenner former Multi-Modal Representation Learning with LEGO: Constructing a Shared Embedding Space for Segmented Human-to-Component Mapping
BSc Lucas Emil Bülow Mortensen · Elias Dahl Lunøe former Automated Coding of Historical Causes of Death Using Fine-Tuned Large Language Models
MSc Lukas Kofoed Petersen former Implementation and evaluation of a Neural Target Speaker Extraction Model for Resource-Constrained Hearing Instruments
BSc Clara Louise Brodt · Mads Helle Højgaard · Julius Gregers Gliese Winkel former Implementation and Quantitative Validation of a Retrieval-Augmented Generation System for DTU’s Information Retrieval
MSc Anton Egeskov Grier · Frederik Valdemar Holck Reimert · Lasse Schnell Danielsen former Development of an Integrated Computer Vision System for Semi-Automated Microbial Colony Analysis
MSc Mathilde Rosholm Block former Adapting Large Language Models for Structured Information Extraction from Sleep Study Reports
BSc Sofus Alexander Kjelgaard Carstens former Automatic Joint Structured Pruning and Quantization for Efficient Neural Network Training and Compression
MSc Paulo Ricardo Beckhauser de Araujo former Exploring methods for building reliable agentic systems using constrained small language models
BSc Carl Schmidt-Svejstrup · David Lindahl former Automatic Speech Assessment Using Transformer-Based Models: Reproduction and Extension of Recent Methods
BSc Mia Isabella Lund former Predicting account number based on extracted address
2025 23 former
MSc Christian Raasteen former AI Agents in Contract Management
MSc Viktor Isaksen former Improving Video Conferencing Experience
MSc Johan Böcher Hanehøj former Applied AI Personalization: Learning User Preferences from Revision Data
MSc Rasmus Jehn Dedenroth Pedersen former Design and Development of a Scalable Volume Transaction Model and Rebalancing Engine for Pension Investments
MSc Christoffer Høeg Wejendorp former Dynamic Neural Networks for Efficient Speech and Audio Processing
MSc Inês da Fonseca Tacanho · Ana Marija Pavicic former Multi-Factor Financial Forecasting with Machine Learning Techniques
MSc Karl Meisner-Jensen former P-PIPE: A learning platform for MLOps
MSc Fabian Scott former Prompt Engineering for Vision-Language Models: Techniques and Evaluation
MSc Adrian Valentin Kragh-Hillers former Last mile problem solved by agentic systems
MSc Gabriel Lanaro former Improving Accuracy of LLM Tool Calling through Prompt Engineering and Reasoning Models
MSc Arnau Molins Carbelo former Intelligent Automation of Invoice Processing: Leveraging Microsoft Power Platform and Cloud Services
MSc Eysteinn Högnason former Integrating Technical Indicators into ARIMA–LSTM Hybrids: A Nordic Market Case Study
MSc Paula Gambús i Moreno former Multi-Modal Deep Learning for Clinical Question Answering on Brain MRI
MSc Laura Pascual Hebrero former Machine Learning for Process Optimization in Pharmaceutical Manufacturing
MSc Spyridon Pikoulas former Orchestration of LLM agents for complex tasks
MSc Juan Iglesias Trebolle former AI-Powered Analysis of Cognitive Test Drawings for Dementia Detection and Deployment
MSc Andreas Østerby Holst Rasmussen former Locally Hosted AI-Based Summarization of Patient Records in Danish General Practice
MSc Bozhi Lyu former From Training to Deployment: Quantization Strategies for Efficient Audio Deep Neural Networks
MSc Luís Miguel Ferreira Freire former Exploring different approaches to customize LLMs for text-to-code generation
MSc Esben Damkjær Sørensen former Multimodal RAG for Answering Queries Across Diverse Data Sources
BSc William Henrik Klingsten Peytz former Adaptive Learning Tool for Middle School Math
MSc Alexandra Bøje Inselmann · Linea Bartholdy former Active Learning for Industrial Quality Assesment
BSc Christian Rahbæk Warburg former Investigating 3D Gaussian Splatting for Efficient Real-Time Radiance Field Rendering
2024 18 former
MSc Mathias Bang Kristensen former Cloud resource cost optimisation for machine learning model training using Kubeflow in Kubernetes Clusters
MSc Pierre Høgenhaug former Analysts vs. Algorithms: Evaluating LLMs’ Ability to Assess Credit Risk Like Human Analysts
MSc Magnús Sigurðarson former Fine-Tuning and evaluating large language models for Icelandic language comprehension
MSc Laura Andrea Paz Salas former Machine Learning and MLOps for Process Mode Analysis in Medical Device Assembly
MSc Hyunho Shin former AI-assisted customer service support and solution proposal
MSc Nael Rashdeen former Backtesting and Deploying Systematic Trading Strategies with Machine Learning and MLOps
MSc Eleftherios Katiforis former Logical Fallacy detection using LLMs
MSc Magnus Nikolaj Nyholm Jensen former Enhancing Regulatory Intelligence with Automated Question-Answering
MSc Tobias Christian Ancher Just Munch former Towards Better Label Error Detection
MSc Jakub Solis former Exploring Modern Cloud Architectures for AI Development and Deployment
MSc Mikkel Theiss Westermann former Alignment of Unstructured Data with Structured Models in Accounting Software Using Artificial Intelligence
MSc Jacob August Xander Ottensten · Gustav Sandholm Franck former Automatic Label Error Detection in MLOps Pipelines
MSc Nima Taghidoust Sourkouhi former Harnessing LLMs for Customer Support and Financial Inquiry Automation
MSc Navaneeth Kizhakkumbadan Pacha former Investigating Deep Learning based methodologies to automate program Synthesis for Biostatistics ADAM programming
MSc Mathias Kristensen former Cloud resource cost optimisation for machine learning model training using Kubeflow in Kubernetes Clusters
MSc Freya Gerup Helstrup former A framework for automated testing of ETL pipelines
MSc Vivian Winther Jacobsen former Generative AI of flexible reports
MSc Kristof Kenez Drexler former Locus: Global Localization of Earthly Images powered by Deep Learning
2023 17 former
BSc Nikolaj Søndergaard Povlsen former Large-scale source-to-model data preparation
MSc Róbert Gers former High performance extraction of financial market data
MSc Andreas Piper Mårtensson former Investigation and implementation of Machine Learning Pipelines within the area of image analysis and classification
MSc Tamas Janos Paulik former Implementation and orchestration of a scalable and automated Machine Learning Operation pipeline
MSc Stefanos Rodopoulos former Exploring Bio relationships through Counterfactual Optimisation Analysis using VAE
MSc Joshua Sebastian former Using Deep Learning and MLOps to detect Dementia through Speech Analysis
MSc Alejandro Rodriguez Salamanca former Memory Efficient Methods for Large Models
MSc Anna Bzinkowska former Machine Learning approach to product categorization in the manufacturing industry
MSc Melina Siskou former Optimizing Machine Learning Operations in Logistics: Exploring Best Practices and Evaluating Tools
MSc Milad Taghikhani former Exploring Multimodal Data Integration to Large Language Models and LLM-based generative AI modeling
MSc Thomas Spyrou former Monitoring black-box classification models in machine learning systems
MSc Laurine Marie Celine Dargaud former Developing a spontaneous speech-based machine learning model for the early detection of dementia
MSc Spyridon Vlachospyros former Context-aware object detection using deep learning
MSc Jonah Jad Tabbal former AutoML and Meta-learning through data science competitions
MSc Mads Dudzik Møller former MLOps in a Deep Learning enabled production environment
BSc Joachim Schrøder Andersson · Jonas Hoffmann former Exploring the Extensions and Limitations of Metadata Archaeology via Probe Dynamics
BSc Julius Holbech Radzikowski · Carl Anton Schmidt former Using meta-labeled data to improve deep-learning classification model robustness, and boost data quality.
2022 10 former
MSc Simon Moe Sørensen former Deep multimodel modelling of images and text from wine
MSc Victor Girardin Flindt former Fault detection in industrial production processes using Deep learning methods
MSc Jakub Wladyslaw Szreder former Creating a Machine learning pipeline for training and evaluation of models for medical data
MSc Nicolai Weisbjerg · Kelvin Foster former Quantifying Uncertainty in Semantic Segmentation using Bayesian Deep Metric Learning
MSc Frederik Kjær · Jonas Christian Rask Levin former Construction of a Recommendation system for best next buy recommendations
MSc Bekarys Gabdrakhimov former End-to-end machine learning project on classification of EEG signal of depressed patients
MSc Xianhao Liu former CPAB flows
MSc Asger Frederik Græsholt · Andri Geir Arnarson former Automation of a fruit sorting system
MSc Gustav Selfort Hartz former Improving searching and tagging in legal contract corpuses using NLP
MSc Marco Placenti former Machine Learning Pipeline Engineering with Amazon Web Services
2021 2 former
MSc Frederik Kromann Hansen · Jonas Søbro Christophersen former Investigating osteoarthritis via x-ray images using deep learning
MSc Morten Holm Thomsen · Simon Kristian Jacobsen former Forecasting of bonds in Emerging Markets using Graph Neural Networks
2020 2 former
MSc Kathrine Thorup Hagedorn former An approach to an Entertainment Recommender System complying to the GDPR ruleset
BSc Dominik Mate Kovacs · Zoltán Kovács former Generating and Detecting Deepfakes
2018 1 former
MSc Julie Liv Cetti Hansen former Foci detection in cells using deep learning

Code

libcpab Public

Library for working with diffeomorphic transformations in numpy, TensorFlow, and PyTorch. Designed to be user-friendly — you don't need to know the slightest thing about diffeomorphic transformations to get started.

Python ★ 42 ⑂ 8
ddtn Public

Code for the CVPR 2018 paper on Deep Diffeomorphic Transformer Networks. Implemented in C++ and TensorFlow with an easy-to-use diffeomorphic transformer layer.

Python ★ 28 ⑂ 6
torchplot Public

A simple plug-in replacement for matplotlib.pyplot that automatically handles .cpu().detach().numpy() before plotting PyTorch tensors.

Python ★ 15 ⑂ 2
TorchMetrics Public

Machine learning metrics for distributed, scalable PyTorch applications. Core developer and maintainer at Lightning AI.

Python ★ 2,200 ⑂ 420
PyTorch Lightning Public

The lightweight PyTorch wrapper for high-performance AI research. Core contributor and maintainer.

Python ★ 29,000 ⑂ 3,200
See all on GitHub →

Grants

Active and recent research grants totalling ~DKK 7M.

Next-Generation VMS: Agentic AI and MLOps for Automated Cross-Camera Tracking and Re-ID

2025

Innovation Foundation Denmark · with Kamal Nasrollahi

DKK 828,000

Next Generation VMS: Contextualized Vision-Language Models for Alarm Refinement and Executive Summaries

2025

Innovation Foundation Denmark · with Kamal Nasrollahi

DKK 1,070,000

General Pretrained EEG Model

2025

Gefion Voucher Grant, Novo Nordisk Foundation · with Lars K. Hansen

DKK 1,400,000

Industrial PhD: Edge-deployment of Deep Neural Networks via Model Compression in Healthcare Applications

2025

Innovation Foundation Denmark · with Rasmus Aagaard, Kaare B. Petersen, Lars K. Hansen

DKK 1,070,000

Fremtidens digitale integratorer: National efteruddannelsesplatform for bæredygtig AI

2024

Danish Agency for Higher Education and Science · with Mathilde Sjølander Andresen, Mark Riis

DKK 2,485,000

Media

I occasionally contribute as an expert source for Danish and international media on topics related to AI, machine learning, and the technology industry.

Børsen 17 Mar 2024

It-rigmand og DTU-forsker: Så længe vil Nvidia dominere AI-kapløbet

Expert source on Nvidia's AI chip dominance
Børsen 28 Jan 2025

Lyver Deepseek om deres AI? Her er eksperters bud

Expert source on DeepSeek's AI claims
Børsen 29 Jan 2025

Topchef i hollandsk chipgigant afviser bekymring om kinesisk AI-robot

Expert source on ASML and AI chip landscape
Videnskab.dk 1 Jul 2022

"Kunstig intelligens tegner verdenskort for proteiner"

Explainer on Nature Communications paper
Videnskab.dk 1 Jun 2019

"Computerne skal lære at tage fejl"

Explainer on NeurIPS paper