Parallel Sessions
Below you will find a list of all confirmed parallel sessions. No previous registration is needed, and attendance will be on a first-come first-served basis. Please note that the programme will be updated continuously, so keep an eye out for more information.
______________________________________________
MONDAY, NOVEMBER 7, 13:45 – 16:00
PS1. Machine Learning in Production: MLOps is all you need? - Meeting room C
The development and operation of machine learning systems for different system settings is difficult for a range of reasons: There are many possible system settings from wearables to super computers, types of data from open to sensitive data and response requirements from milliseconds to daily predictions. On top of this different competences need to collaborate and make proper choices of software technologies to achieve success. This session will dive into Machine Learning Operations (MLOps), and the challenges, processes and technologies that provide data scientists and software engineers with a strong background for enabling successful ML systems in production and discuss unsolved challenges in this regard.
Organised by Nicki Skafte Detlefsen (Technical University of Denmark) and Mikkel Baun Kjærgaard (University of Southern Denmark)
PS2. Generative models - Meeting room D
Deep generative models are a class of probabilistic models that use neural networks to unsupervised estimate complex and high dimensional distributions. It is a research field that has seen great process in the last years, and today these models are capable of, e.g., generating photorealistic images, and they have found application across the sciences as well as in the industry. In the two sessions on generative models (PS2 and PS10), we have invited leading national and international speakers to present their latest research. With generative models as the common theme, the talks will cover experimental design, generalisation, out-of-distribution detection, generative AI, representation learning, and few-shot learning. The talks are mainly methodological, and we aim to make them and the discussions relevant both for the machine learning specialists and the broader data science community.
Organised by Jes Frellsen (Technical University of Denmark)
Co-organised with the Pioneer Centre for Artificial Intelligence
PS3. Open science - Auditorium A
Open science is the movement to make scientific research (including publications, data and software) and its dissemination accessible to all levels of society. It encompasses practices such as publishing open research, campaigning for open access, encouraging scientists to practice open code science broader dissemination and engagement in science.
In this parallel session we will give an introduction and overview of movements within open science. We will have talks on how to access data openly and also share data yourself, showcase different tools for open science such as GitHub and scikit-learn and how those can be used on cloud services. We will finally highlight how one can publish scientific results openly.
Organised by Melanie Ganz (Rigshospitalet/University of Copenhagen) and Cyril Pernet (Rigshospitalet)
PS4. Data Quality in Data Science - Auditorium B
The session will focus on issues related to data quality in data science. It will feature several talks and a panel discussion, with speakers from both academia and industry. The session will cover topics such as what data quality is all about, how to achieve data quality in general, and how to improve data quality in different scenarios within data science.
Organised by Jacob Ramlov Jensen (Go Autonomous) and Hua Lu (Roskilde University)
PS5. Data Science in the Wild - Meeting room U.14
Scalability, missing values, small datasets, inconsistent data records, sensitive data, unreliable black-box models versus reliable yet more restrictive models, computational expense, cloud infrastructure, … — This session is for knowledge sharing about the hurdles, problems, and surprises we face when moving from theory and methodology developed on clean toy datasets to deploying data science solutions in the wild. How is data science used in practice? What issues are predominant? What are the lessons learnt and advice for fellow data scientists?
Organised by Cecilie Utke Rank (Rigshospitalet/University of Copenhagen), Federica Belmonte (Danish Cancer Research Center) and Sebastian Weichwald (University of Copenhagen)
PS6. Patient benefit from risk assessment models and tools - Meeting room U.11
Models for risk of disease, benefit from treatment, risk of outcome, among others, are increasingly published. As data are being assessable in terms of genetics, proteomics, metabolomics, radiomics, etc., we constantly gain more insight into the underlying biology and produce new models. With the use of machine learning, the models created become more complex, less transparent, and harder to validate. These are some of the reasons for modern risk modelling not being transferred at a high rate into the clinic for patient benefit. This workshop will concentrate on the modern risk model’s translation into the clinic through proper validation and quality assurance. The workshop will show examples from medicine and discuss the proper procedures for ensuring sufficient model quality to benefit patients.
Organised by Mads Nielsen (University of Copenhagen), Sisse Rye Ostrowski (Rigshospitalet/ University of Copenhagen) and Deirdre Fenton (Aarhus University)
PS7. Applying Causal Methods - Meeting room U.8
Causal inference promises the tantalizing possibility of reliably drawing causal conclusions from observational data. It generally requires strong and hard to verify assumptions, however, which makes it difficult to apply correctly in practice. In this session, we will dive into the unique promises and difficulties of applying causal methodology in a variety of research areas.
Organised by Leonard Henckel (University of Copenhagen)
PS8. Geometry and Topology in Machine Learning - Meeting room U.3
Much of modern data is high dimensional, very complex, and highly nonlinear. Recently, methods building on the centuries of knowledge in mathematics of such structures have been successfully brought in.
This session gathers the Danish community in the area and also invites the rest of the Data Science community to learn about these methods and ideas.
Organised by Lisbeth Fajstrup (Aalborg University) and Søren Hauberg (Technical University of Denmark)
TUESDAY, NOVEMBER 8, 10:15 - 12:15
PS9. Reproducible AI & Experiment Tracking using MLOps - Auditorium A
Being able to replicate an experiment and obtain the same results is a crucial part of the scientific method, though when it comes to machine learning applications, machine learning practitioners might often find themselves with a plethora of confusing Jupyter notebooks, where executing the cells in a very specific order is the best hope of reproducibility.
In this session, Weights and Biases will tell about their MLOps platform for experiment tracking, dataset versioning and model management. Then, all attendees will invited to be putting their ML skills to use in a Kaggle InClass competition. Finally, Danish startup Alvenir will tell about how they manage their experimental budget when training large models.
Organised by Viktor Stenby Johanson (Technical University of Denmark), Joakim Bruslund Haurum (Aalborg University) and Kenneth Borup (Aarhus University)
PS10. Generative models - Meeting room C
Deep generative models are a class of probabilistic models that use neural networks to unsupervised estimate complex and high dimensional distributions. It is a research field that has seen great process in the last years, and today these models are capable of, e.g., generating photorealistic images, and they have found application across the sciences as well as in the industry. In the two sessions on generative models (PS2 and PS10), we have invited leading national and international speakers to present their latest research. With generative models as the common theme, the talks will cover experimental design, generalisation, out-of-distribution detection, generative AI, representation learning, and few-shot learning. The talks are mainly methodological, and we aim to make them and the discussions relevant both for the machine learning specialists and the broader data science community.
Organised by Jes Frellsen (Technical University of Denmark)
Co-organised with the Pioneer Centre for Artificial Intelligence
PS11. Data Science Educations: What is the Road to Success? - Meeting room U.14
How do we develop better data science educations? This session will provide an overview of the status of data science educations in Denmark. Based on the overview the session will explore how to further develop the educational programs and provide time for networking.
Organised by Mikkel Baun Kjærgaard (University of Southern Denmark) and Hua Lu (Roskilde University)
PS12. (Un)code the bias: Algorithmic Fairness in Data Science - Meeting room D
Over the last decade, scientists, activists and journalists have repeatedly dismantled the idea of a ’neutral’ algorithm: whenever algorithms are involved in decision-making, the question of fairness arises. It is not a novelty, either, that there is no such thing as inherently unbiased data or unbiased use of an algorithm. What is it, then, that a fairness-aspiring data scientist can – and should – do? And what is the role of data science in this inherently interdisciplinary challenge?
In this session, we aim to tackle these questions on several interconnected levels: data collection; algorithmic design; usage and implementation of algorithms; and finally, societal action and response. Our three speakers will not only share their expertise on these topics, but also invite you to reflect on everyday life dilemmas, actively engage with each other and co-create a discussion on fairness in data science.
Will this session uncode the algorithmic bias? Probably not, but at least you will be more aware of it.
Organised by Anastassia Vybornova (IT University of Copenhagen), Jens Ulrik Hansen (Roskilde University) and Federica Belmonte (Danish Cancer Research Center)
PS13. Bioinformatics - Auditorium B
Over the course of the last 20 years changes in assay and sensor technologies have transformed life-science to become one of the most data rich disciplines that exist. This development affects all areas of life-science from genetics to agriculture and has brought data to the centre stage of all activities.
In this session we have invited a broad group of experts from both academia and industry. The aim is to give you a flavour of the diversity and depth of activities involved in the modelling, analysis, and application of life-science data ranging from precision medicine and vaccine development to precision agriculture.
Organised by Roald Forsberg (Raven Biosciences and JADBio)
PS14. Open Source - Meeting room U.11
We take a dive into the world of open-source data science. Firstly, we will talk about what Danish foundation models are and why we need them. Secondly, we discuss how open-source development can be incorporated when thinking about business development. Finally, we show concrete examples of how benefits can be drawn from open data.
Organised by Andreas Tind Damgaard (Region Midtjylland/Danish Data Science Community)