Swedish e-Science Academy 2025

Satellites

Four different satellites will take place in connection with the event. Below you can find details about each of them.

All Satellites take place in different rooms of the Social Sciences Building. The specific room is listed for each session below. Coffee is always served by the respective presentation room.

1. "The science of mathematical modeling, numerical computing, optimization, and design - a workshop in honor of Martin Berggren."

October 14, 13:00 - October 15, 12:00
Room: SAM.A.230 (Tuesday), UB.A.220 (Wednesday)
Organizers: Mats G Larson, Eddie Wadbro, Paolo Bientinesi

October 14

13:15–13:30 Intro/Welcome
13:30–14:00 Niels Aage
14:00–14:30 Rainer Picard
14:30–15:00 Dan Henningson
15:00–15:30 Coffee break
15:30–16:00 Stephan Schmidt
16:00–16:30 Juliette Chabassier
16:30–17:00 Anton Evgrafov
19:00– Dinner

October 15

9:00– 9:30 Jarmo Mallinen
9:30–10:00 Eddie Wadbro
10:00–10:30 Coffee break
10:30–11:00 Ole Sigmund
11:00–11:30 Gunilla Kreiss
11:30–12:00 Dirk Pauly
12:00– Closing and lunch

You can find the full program here

---

2. "Digital Humanities"

October 15, 09:00 - 12:00
Room: Humlab one floor below the University Library
Organizer: Karin Danielsson

9:00 Intro Research infrastructures in Digital Humanities

• InfraVis, Humlab and Huminfra: Karin Danielsson • InfraVis Infrastructure development: Evelina Liliequist • MAL: Ivanka Hristova • Swedigarch: Ershad Gholamrezaie

9:45 Session 1
Swedigarch showcase SEAD: Cenk Demiroglu

10:00 Coffee break/Fika

10:30 Session 2
• InfraVis showcase: NordPow, Maria Podkrytova • InfraVis showcase: Storytelling through visualization: Interactive data visualization for education and museums, Mattis Lindmark • Humlab/Huminfra: MoCap showcase project: E-motion, Onur Kilic

11:15 Keynote speaker: Per Axelsson
”CARE principles, data and the importance of context”

You can find the full program here.

---

3. "Workshop on Advancements in Federated Machine Learning"

October 16, 13:00 - 16:00
Room: UB.A.230
Organizers: Salman Toor, Johan Ostman

13:00 - 13:15 Welcome to the Federated Learning Workshop

13:15 - 13:40 Presentation-1: Distributed Continual Learning in High-Dimensional Regression, Ayca Ozcelikkale, Uppsala University

13:45 - 14:10 Presentation -2: Fairness Regularization in Federated Learning: Connections, New Methods, and Empirical Insights, Zahra Kharaghani, Umeå University

14:10 - 14:45 Coffee break

14:45 - 15:10 Presentation -3: Federated Learning for Industrial Use Cases, Sima Sinaei, RISE Research Institutes of Sweden

15:15 - 15:40 Presentation -4: Federated Machine Learning: Opportunities and Cybersecurity Challenges, Salman Toor, Uppsala University

15:40 - 16:00 Open Discussion Session

You can view more information here pdf, 131 kB.

---

4. "Common challenges in large-scale code development"

October 16, 13:00 - 16:00
Room: UB.A.220
Organizers: Philipp Birken, Jonas Lindemann, Valera Veryazov

13.00-14.30 presentations:

1. Peter Munch (TU Berlin): deal.II: challenges and opportunities of developing an open-source FEM library

2. Luca Ferranti (Aalto University and Nordic-RSE): Overview of Nordic-RSE

3. Jonas Lindemann (Director of LUNARC, LU): From Code to Classroom:
Practical Applications of Large Language Models in Software Development

14.30-15.00 Coffee

15.00-16.00
Short presentations and discussion.

You can view the abstracts here pdf, 44 kB.

Lars Karlsson: Multi-Versioning Compilation of Linear Algebra Expressions

Numerical linear algebra expressions are commonly evaluated using high-performance kernels found in libraries such as BLAS and LAPACK. The mapping from an expression to a corresponding sequence of kernel calls is by no means unique. A standard matrix chain needs only one kernel for matrix multiplication, yet there is still an exponential number of variants. Worse, the fastest variant for a given expression will depend also on the sizes of the matrices. This fact alone makes the compilation of linear algebra expressions challenging, since the sizes of the matrices are rarely known at the time of compilation. Current approaches either fail to provide good performance on all inputs (e.g., by always using a specific variant) or they incur significant code size and runtime overheads (e.g., by relying on an interpreter or JIT compiler).

To address the shortcomings of existing methods, we are investigating a multi-versioning approach where the compiler emits a few carefully selected variants. The selection can be guided by theory and heuristics to offer good performance on all inputs. At the same time, the code size and runtime overheads are kept small. In this talk, I will summarize our recent work on a subclass of expressions called generalized matrix chains and discuss some of the challenges of extending the approach to other types of expressions.

Paul Watson: From eScience to Impact

The UK National Innovation Centre for Data (NICD) was created in 2018 by the UK government and Newcastle University to generate economic and societal benefit from data science and AI. Most commercial and public organisations are drowning in data. They can see that this could help them gain insights into their business, optimise existing operations, and launch new data-driven products and services. However, few organisations have the skills needed to realise the potential of data to transform their business. In contrast, universities have these skills in abundance, due to their expertise in areas like eScience, but the issue is how to deploy those skills at scale outside of universities? Since 2019, NICD has been addressing this challenge by transferring data science and AI skills into organisations.

It does this through projects in which its team of data scientists collaborate with staff in the external organisation on projects that address a business needs. To date, it has run 160 successful projects with companies of all sizes, across a wide variety of sectors. The talk will explain the systematic way NICD scopes and runs projects to ensure that they successfully deliver impact for the external organisation.

This will be illustrated with a set of case studies.

Monika Eisenmann: Probability in Numerical Analysis: Methods and Applications

In the field of applied mathematics, stochasticity plays an increasingly important role in modeling and approximating real-world problems. The sources of randomness vary depending on the application.

In this presentation, I will discuss examples where stochasticity influences both the formulation of models and the design of numerical methods.

When modeling a process, adding noise extends deterministic (partial) differential equations to stochastic (partial) differential equations.

Examples include phase-field models and fluid flow problems, where the noise can represent small-scale structures arising from thermal fluctuations or introduce stochastic forcing that cannot be predicted in deterministic models.

On the methodological side, randomness can be integrated directly into the numerical scheme. Approaches such as Monte Carlo methods or stochastic gradient methods can reduce computational costs and, in some cases, improve accuracy compared to their deterministic counterparts.

Paul Townend: Low-Carbon Resource Management in the Edge-Cloud Continuum

Data centers are the infrastructure on which most modern distributed systems and applications are built. They deliver high levels of service at the expense of significant energy consumption and environmental impact. This talk presents an academic and industrial perspective on data centers as complex cyber-physical systems-of-systems, and describes how intelligent job scheduling can be used to improve the performance and reduce the energy consumption and carbon footprint of server clusters, finishing with some examples from Umea University's current research output.

Orcun Göksel: Continual Learning and Domain Adaptation in Image Analysis

For a long time, learning-based models were seen as a one-off train and multi-use solutions. This omits the fact that training data and annotations are commonly limited, real-world data has natural variations challenging to capture with limited training, and understanding and solutions to a problem can always be expanded given new data. Human learning is indeed based on these; i.e., we never call any time-point as the end of training, but can always integrate any new information on our existing knowledge.

In this presentation, I will start by motivating incremental knowledge accumulation in learned models. I will introduce some of our developments based on generative models for image replay and feature alignment in scenarios of domain and class incremental learning. I will exemplify these via applications of image classification and segmentation on natural and medical images.

Gemma Atkinson: Mining diversity hotspots of bacteriophage genomes to discover new microbial immune systems

Viruses that infect bacteria (bacteriophages or simply phages) often battle each other inside the same host, using genetic weapons that act as immune systems. These genes are frequently concentrated in genomic diversity hotspots—regions with unusually high turnover of gene content.

By systematically identifying and mining these hotspots with computational pipelines to cluster and annotate unknown genes, we can uncover new microbial immune systems and better understand phage–phage conflict in the battleground of the bacterial cell.

Amber Mace: Modeling Electrolytes for Solid-State Batteries

The development of novel, cost-efficient multiscale modeling strategies for ion diffusion in solid-state electrolytes offers the potential to deliver groundbreaking insights into the compositional and structural factors that govern conduction mechanisms. These strategies also enhance our understanding of how tuning parameters influence transport behavior in such materials.

To enable computationally driven materials discovery, large-scale studies across a wide range of materials are essential. These efforts require automated, efficient, and robust modeling frameworks that go beyond current state-of-the-art methods such as molecular dynamics.

In my talk, I will present my team’s work on a multiscale modeling methodology developed specifically for this purpose. Our approach combines topological analysis of the potential energy landscape with efficient statistical models to predict the ionic diffusivity of crystalline materials. These models achieve accuracy comparable to molecular dynamics simulations, but at only a fraction of the computational cost.

I will also discuss our ongoing efforts to extend this framework to non-crystalline polymer-based electrolytes and to composite materials composed of multiple phases.

Mats Holmström: Particle-in-cell simulations of the interactions between the solar wind and planets

Why are Venus, Earth, and Mars so different? When the planets were formed, they were probably quite similar, but not today. One explanation may be the interaction of the solar wind with the upper parts of the planets' atmospheres. The solar wind is a stream of charged particles from the Sun.

We want to better understand the physics of the interactions between the solar wind and solar system objects by computer simulations. The main tool to accomplish this is an open source hybrid plasma solver that we have developed. This is a particle-in-cell model that treats ions as particles and electrons as a fluid. High resolution simulations require high performance computing. The modeling is done in close interaction with the analysis of observational data (ions, electrons and neutral atoms) from our instruments.

We can then study which parameters are important for solar wind interactions today, as well as what the interaction looked like in the past, for planets in our solar system, and for planets at other stars, exoplanets.

Locations

All presentation sessions will take place in the Umeå university’s Social Sciences Building (Samhällsvetarhuset) with exact room allocations presented in the program. All rooms are on entrance floor (numbered 2) except for Humlab, being one floor below. Registration will take place adjacent to, or inside, the presentation rooms.

Lunches will be served in Restaurant Hjortron, in building Universum, top floor. For those that have chosen packed lunch, that will be delivered at the registration desk.

The Poster Session will be held just opposite restaurant Hjortron, also on the 2nd floor of Universum.

The Academy Dinner at 19.00 on October 15 will be held at Restaurant Artonnittiosju (1897) at Hotel Mimer, downtown Umeå.

Find your room using Mazemap:
https://use.mazemap.com/#v=1&campusid=289&zlevel=1&center=20.309684,63.821760&zoom=15.5

Hotels and transportation

Participants for which eSSENCE has reserved hotel rooms are all booked at Hotel Scandic Plaza downtown Umeå.

There are plenty of buses between Umeå University and downtown (central stop named Vasaplan). Tickets can be bought via the app “Ultra – Umeås lokaltrafik” or with credit/debit card on the bus.

For your conveniency, we provide a chartered bus (free of charge) from Scandic Plaza to Samhällsvetarhuset on Thursday morning, leaving Scandic Plaza at 8.30. Please, make use of that opportunity and make sure to be on time for the morning conference session.

Food Restrictions and Drinks

Participants’ food restrictions have been forwarded to providers of lunches, coffees, and dinner. For dinner, people with registered food restrictions will receive a card to place on top of their glass at the table to facilitate identification of people with restrictions. At lunches, just tell the staff at time of serving.

For the dinner, all participants will receive two drink tickets to be exchanged for alcoholic or non-alcoholic drinks.

Poster Presentations

Poster presenters are advised to hang their posters on designated poster stands in Brashörnan during lunch-time (12 – 13) on Wednesday October 15. In the last session before the poster presentation, all poster presenters will be allocated 45 seconds to introduce their poster orally, presenting one slide. Detailed instructions are provided via email.

Fragrance Consumer Products

When visiting Umeå University, we urge you to avoid wearing strong scents, such as perfumes and aftershaves. Show consideration for people suffering from allergies and fragrance hypersensitivity, as they can suffer adverse reactions to strong scents.

Acknowledgements

For the organization of the Swedish eScience Academy, we express our deepest gratitude to many people that have contributed with their time and skills. In particular, we mention Carl Christian Kjelgaard Mikkelsen, Josefin Starkhammar, Anne-Lie Persson, Lena Hellman, Per Sehlstedt, Antonio Seo, and Chanh Nguyen Le Tan.