Research projects AI4Research

Ten research projects associated with artificial intelligence and machine learning have been awarded funding within AI4Research 2024. Read more about the projects below.

Secure and Private AI: Federated Machine Learning

Andreas Hellander, Department of Information Technology

Representative data of good quality is the most important component in machine learning. There are many datasets that today are not utilized because there are challenges and risks associated with uploading data to central clouds or supercomputers. This can be due to data being private, sensitive, or too large to move. Federated machine learning provides a way to circumvent this problem by coordinating the training of models directly on local data without having to move it. This can enable widely different applications such as collaboration between various hospitals on AI-based cancer diagnostics, privacy-preserving updates of models used in mobile applications, and predictive maintenance of vehicles where models are trained directly on the vehicles in large fleets.

However, there are challenges with federated machine learning. We now have to manage a distributed system where we do not have control over how training data is distributed across different clients. This poses new demands on both the software and the algorithms compared to traditional centralized ML. The goal of our project is to study these differences and propose new improved algorithms that can enable federated learning for larger models than what is practical today.

What do you look forward to the most during your sabbatical?

To establish new contacts at UU and explore new possible collaborations around privacy-preserving training of machine learning models in various application areas.

AI methods to decipher the patient-specific dynamics of nervous system cancers

Sven Nelander, Department of Immunology, Genetics and Pathology

Summary of your project: In the last two decades, advancements in cancer therapies have shown varied success. While some cancers can be treated by targeted or immune-based treatments, challenges persist for central nervous system (CNS) cancers. Leading a team specializing in CNS cancer studies, I plan to spend my research time on AI4research to develop new approaches to probe how brain tumors alter their behavior over time. A key focus of the project will be to adapt AI strategies to model assess cell movement through brain tissue and models for targeted intervention in neural cancer cells, using new measurements generated by my team.

What do you look forward to the most during your sabbatical?

To deepen my skills in AI modeling and to interact with interesting colleagues.


Using AI for spectroscopic characterisation of cool dwarf stars

Ulrike Heiter, Department of Physics and Astronomy

The properties of stars determine the structure and evolution of galaxies and the formation and evolution of planetary systems. Thus, to study galaxies or extrasolar planets we need to be able to characterise stars to high precision. The stellar properties include quantities like surface temperature, surface gravity, and the abundances of chemical elements. These are derived in the best way from spectroscopic observations obtained with ground-based or space telescopes. Stellar spectra, i.e., the amount of radiation emitted by a star at its surface as a function of wavelength, depend in a non-linear way on the physical conditions in the stellar atmosphere. The shape of the spectra can be predicted by solving the equation of radiative transfer through a model atmosphere characterised by a specific combination of stellar properties. The calculated spectra are then compared to the observations to infer the stellar properties. Machine-learning techniques using artificial neural networks (ANNs) are ideally suited for an efficient implementation of the spectroscopic characterisation of stars. This applies in particular to the large-scale spectroscopic stellar surveys which have provided high-quality spectra for millions of stars, and to stars with low surface temperatures (from about 3000 to 4000 K), so-called "M dwarfs", which emit more complex spectra than warmer stars.

This project will focus on the investigation of ANNs that have been used for the spectroscopic characterisation of M dwarfs. Different kinds of input data used for training the ANN models, together with different set-ups and choices of hyperparameters will be explored and the performances resulting from the different alternatives will be compared. The aim is to provide consolidated guidelines for the usage of AI/ML methods to characterise cool dwarf stars, be it planet hosts or tracers of Galactic structure, enabling the best choice of inputs and set-up of ANNs. The project is closely related to the Plato mission, a space telescope designed to detect extrasolar planets, to be launched by the European Space Agency in 2026. The stars to be observed by Plato include several thousand M dwarfs.

What do you look forward to the most during your sabbatical?

Besides working on my own project, I am looking forward to discussing ideas and experiences related to AI methods and applications with the other participants of the AI4Research project.

AI-enhanced virtual screening to accelerate drug discovery

Jens Carlsson, Department of Cell and Molecular Biology

The AI4Research project is focused on development of an AI-enhanced platform for computer-aided drug design with the aim to discover novel therapeutic targets.

What do you look forward to the most during your sabbatical?

I look forward to discussions with other researchers that are interested in artificial intelligence and to initiate new projects in this area.

Exploring the Dark Side of the Universe with Machine Learning at Particle Colliders

Rebeca Gonzalez Suarez, Department of Physics and Astronomy

The goal of this project is to uncover clues of a dark sector parallel to the standard model (SM) of particle physics that is weakly connected to it via what is called the Higgs portal and can potentially explain dark matter (DM). To achieve this, my team and I will develop machine learning (ML) techniques to use in LHC data and beyond.

What do you look forward to the most during your sabbatical?

I am looking forward to the AI4Research sabbatical for two reasons: to find the time to focus on the machine learning aspects of my current research, which sometimes take a backseat with respect to other areas, and to expand my network inside Uppsala University, to find new collaborators and get ideas for new interdisciplinary projects.

Rebeca Gonzalez Suarez

david

Development of educational material for AI-based social science

Matteo Magnani, Department of Information technology

Summary of your project: During this sabbatical I will develop advanced-level educational material on AI-based social science, which are a fundamental part of the emerging discipline known as Computational Social Science (CSS). While specific types of AI have been used in specific sub-fields of the social sciences for a long time, most notably Agent-Based Modelling and Natural Language Processing, only recently AI has been broadly recognised as a new paradigm applicable across all the social sciences. Despite this recognition, there is still no textbook or equivalent materials describing this paradigm in a way that is comprehensive in terms of covered methods, technically advanced, and clearly connected to social science questions and methodology. Through the developed materials, my ambition is to play an important role in the shaping of AI-based social science (a topic that will soon be a fundamental part of any social science research education), and to help establish Uppsala University as one of the leading centres in CSS. This project connects the AI4Research initiative with the Information Laboratory (InfoLab) at the Department of Information Technology and the cross-faculty Computational Social Science Laboratory (CSSLab).

What do you look forward to the most during your sabbatical?

To develop educational material, I will have to read a lot and interact with other experts in different AI methods. AI4Research is a perfect environment for both.

Digging deeper into anxiety mechanisms using multimodal data and AI

Andreas Frick, Department of Medical Sciences

How do we learn which situations and cues in the environment that predict danger and which are safe? How do we update our response to a once-threatening situation that is now safe? To answer these questions, we combine psychological and neurobiological perspectives using data from brain imaging, psychological experiments, and self-reports. Current approaches to analyze these multimodal data do not make use of the rich information and structures inherent in the data. We aim to develop a novel framework using machine learning to better harness the data and further our understanding of the basic questions of how we safely navigate the world. This understanding can then be used to improve treatments for anxiety disorders, where certain safe situations evoke strong fear reactions.

What do you look forward to the most during your sabbatical?

To be in an inspiring environment where I can focus on learning more about AI and devote time to develop our analysis methods. Spend time with and learn from other researchers working with similar methods and have similar interests.

Automating the Exercise of Public Authority

Jenny Eriksson Lundström, Department of Informatics and Media

Decision-making and especially the exercise of authority against an individual has a pervasive significance both for the individual and for society at large. This project aims to study professional, practical and technical aspects of AI-based automated decision-making, especially regarding the exercise of authority against individuals, with the aim of highlighting the risks and opportunities that arise for the organization, individual and society when professional judgment is increasingly replaced by AI solutions. This focus means that this project constitutes a sought-after contribution to individuals (decision-makers, authority practitioners and the individual) and organizations.

What do you look forward to the most during your sabbatical?

I look forward to meeting and exchanging experiences with colleagues at UU to hopefully strengthen collaborations on AI within and between our various research fields and disciplines. I also look forward to focused time for commitments in and the implementation of my project.


Jenny Eriksson Lundström

david

Text-as-Data and Statistical Inference

Måns Magnusson, Department of Statistics

My project addresses the growing availability of digital textual data in fields like social sciences and humanities (SSH), such as parliamentary proceedings, court cases and newspapers. The project focuses on advancing statistical inference for analysing textual data. It seeks to combine probabilistic programming frameworks with specialised models, such as pre-trained language models, to facilitate flexible modelling of textual data in SSH, incorporate prior knowledge into textual models for capturing relevant concepts, and develop new methods for evaluating and diagnosing probabilistic models for textual data. The ultimate goal is to enhance the scientific and statistical validity and interpretability of analyses conducted on large corpora of textual data.

What do you look forward to the most during your sabbatical?

Discussing my issues with other researchers, especially how we can combine machine learning models and algorithms with more traditional statistical inference. Of particular interest is understanding how others integrate these models and methods into the research process to draw valid conclusions.

The impact of AI on the work environment

Åsa Cajander, Department of Information Technology

In a rapidly changing world, artificial intelligence (AI) has become a force to be considered, and its consequence on how we work is becoming increasingly clear. This development raises interesting questions about how AI affects the work environment and our well-being.

In this research project, our main objective is to explore the consequences of AI usage in the workplace. The project aims to understand how early adopters of AI in the IT industry worldwide are affected in their work environment and how the technology affects the working conditions of healthcare professionals.

Through collaboration with other researchers and an interdisciplinary approach, the project promotes positive changes in the work environment. It is an important effort to create a more inclusive workplace and a society that harnesses the opportunities that AI can offer.

What do you look forward to the most during your sabbatical?

Having the time and resources to explore the impact of AI on the work environment and to collaborate with other researchers is a great opportunity! I want to contribute to a better understanding of how AI can improve working conditions and promote positive changes in the world of work. It is an exciting and meaningful challenge that I look forward to tackling.

FOLLOW UPPSALA UNIVERSITY ON

facebook
instagram
twitter
youtube
linkedin