DISA

When? Friday May 16th 14-16
Where? Onsite: D2272 at Linnaeus University in Växjö and online
Registration: Please sign up for the PhD-seminar via this link https://forms.gle/DKAh2iCN5EGEth9F6 by May 14th (especially important if you plan on attending onsite so we have fika for everyone)

Agenda
14.00-14.10 Welcome and practical information from Welf Löwe
14.10-14.55 Presentation and discussion: Secure On-Premises Deployment of Large Language Models for Enhanced Patent Drafting – Homam Mawaldi, AWA
14.55 – 15.05 Coffee break
15.05 – 15.50 Presentation and discussion: Enhancing E-commerce Personalization with a Hybrid Recommendation and Advanced Search System – Kailash Chowdary Bodduluri, Enode
15.50 -16.00 Sum up and plan for our seminars in June

Abstracts

Secure On-Premises Deployment of Large Language Models for Enhanced Patent Drafting – Homam Mawaldi, AWA

Patent drafting is a complex and high-stakes process for securing intellectual property rights. During the patent prosecution phase, maintaining confidentiality is crucial, which makes cloud-based third-party services inadequate. This study explores the feasibility of AWACopilot, a secure, on-premise solution comprising a web service that leverages open-source large language models (LLMs) to assist patent attorneys in the intricate patent application drafting process. AWACopilot generates key patent sections such as background, abstract, detailed description, etc., from human-crafted claims, addressing the data security risks posed by cloud-based AI services. Its modular architecture enables customization and adaptability to different patent tasks. Although challenges remain—including reliance on LLM capabilities and the need for rigorous content verification—this study demonstrates the potential for secure, AI-driven solutions to enhance patent drafting workflows.

Enhancing E-commerce Personalization with a Hybrid Recommendation and Advanced Search System – Kailash Chowdary Bodduluri, Enode

In the evolving landscape of e-commerce, personalizing user experience through recommendation systems has become a way to boost user satisfaction and engagement. However, small-scale e-commerce platforms struggle with significant challenges, including data sparsity and user anonymity. These issues make it hard to effectively implement recommendation systems, resulting in difficulty in recommending the right products to users. This study introduces an innovative Hybrid Recommendation System (HRS) to address challenges in e-commerce personalization caused by data sparsity and user anonymity. By blending multiple dimensions of the data into one unified system for producing recommendations, this system represents a notable advancement in achieving personalized user experiences in the context of limited data. In addition to the recommendation system, we have also developed an effective search feature with capability of leveraging fuzzy matching, TF-IDF vectorization, and a Swedish language synonym model for query expansion. Our current research focuses on integrating these two independent systems—recommendations and search—to address their individual limitations and create a unified discovery ecosystem. By combining explicit search behaviors with implicit user preferences and exploring technologies such as large language models and sequential recommendation frameworks, we aim to further improve and optimize product discovery in data-sparse environments.

When? Friday April 25th 14-16
Where? Onsite: D2272 at Linnaeus University in Växjö and online
Registration: Please sign up for the PhD-seminar via this link https://forms.gle/BTAE5wY4XW9TDB2A9 by April 23 (especially important if you plan on attending onsite so we have fika for everyone)

Agenda
14.00-14.10 Welcome and practical information from Welf Löwe
14.10-14.55 Presentation and discussion: Enhancing Efficiency in Industry: Automation, Analytics, and Digital Twins – Arslan Musaddiq
14.55 – 15.05 Coffee break
15.05 – 15.50 Presentation and discussion: Machine-Learning Models for Two-Dimensional Antiferromagnetic Materials – Shahid Sattar
15.50 -16.00 Sum up and plan for our seminars in May

Abstracts

Enhancing Efficiency in Industry: Automation, Analytics, and Digital Twins –Arslan Musaddiq
This presentation will be about ongoing industry collaborations focused on data-driven optimization in manufacturing, energy, and logistics. These projects aim to improve operational efficiency, enhance decision-making, and drive digital transformation across various sectors. The presentation will cover the automation of waste input measurement and real-time thickness monitoring in construction board production, energy data analysis for baseload identification and prediction, digital twin development for sawmill process optimization, and scheduling optimization for mobile blood donation units. The seminar will explore the challenges, methodologies, and practical impact of integrating smart systems into traditional industries.

Machine-Learning Models for Two-Dimensional Antiferromagnetic Materials – Shahid Sattar
Two-dimensional antiferromagnets (2D AFMs) recently gained tremendous scientific interest owing to their use in next-generation spintronic devices. In this talk, I will discuss about our recent effort to develop machine-learning (ML) models for 2D AFMs. More specifically, how ML models together with molecular dynamics simulations can be effectively used to capture surface reconstructions in topological 2D AFMs [1]. Additionally, I will show how ML models can be employed to compute thermal properties and heat transport in 2D MnX and Janus XMnY (X,Y=S, Se, Te). Finally, I will talk about potential new avenues which can be explored combining first-principles calculations and ML models.

[1] S. Sattar, D. Hedman and C. M. Canali, Phys. Rev. Research (2025).

When? Friday April 4th 14-16
Where? Onsite: D2272 at Linnaeus University in Växjö and online
Registration: Please sign up for the PhD-seminar via this link https://forms.gle/aht4pqfi4XWv76PK6 by April 2nd (especially important if you plan on attending onsite so we have fika for everyone)

Agenda
14.00-14.10 Welcome and practical information from Welf Löwe
14.10-14.55 Presentation and discussion: Data-driven Community-based Business Models for Forestry: Friends and Foes – Samin Ghalandarzadeha, Södra
14.55 – 15.05 Coffee break
15.05 – 15.50 Presentation and discussion: Generalisation of GenAI and Verification pipelines – Nemi Pelgrom Fortnox
15.50 -16.00 Sum up and plan for our seminars in May

Abstracts

Data-driven Community-based Business Models for Forestry: Friends and Foes – Samin Ghalandarzadeha, Södra

Building on our recent systematic literature review of the challenges and opportunities in data-driven and community-based business models for agriculture and forestry , this study will explore key findings and will seek to bridge the gap between theory and practice by engaging experts from a major Swedish community-based forestry association.

Through interviews with industry specialists, this study will uncover new opportunities and challenges for implementation of the abovementioned business models, as well as test these evidence-based findings. Ultimately, this research will assess the feasibility of the proposed business models, identify context-specific challenges and benefits, and strengthen the theoretical framework with real-world insights.

Generalisation of GenAI and Verification pipelines – Nemi Pelgrom Fortnox

Based on the publication of AlphaGeometry a little over a year ago, a new development in the strive towards trustworthy AI is gaining popularity; to combine generative models with automatic verification tools, as separate parts of frameworks or information pipelines. Many formats of information pipelines have been well researched before Generative AI joined the picture, but the difficulty in interpreting GenAI models into the languages (terminologies) used by those fields, makes it hard for researchers to interpret what previous results are relevant in these new contexts. In this presentation I will propose a formal terminology for describing this kind of pipeline, which may be used as guidance for how to interpret the validity, or trustworthiness, of any pipeline produced that fulfils the relevant criteria.

When? Friday March 21 14-16
Where? Onsite: D2272 and via zoom
Registration: Please sign up for the PhD-seminar via this link https://forms.gle/XmL6bguq3T4Lax71A by March 19th (especially important if you plan on attending onsite so we have fika for everyone)

Agenda

14.00-14.10 Welcome and practical information from Welf Löwe
14.10-14.55 Presentation and discussion: Immersive Analytics for Understanding Ecosystem Services Data – Benjamin Powley
14.55 – 15.05 Coffee break
15.05 – 15.50 Presentation and discussion Bridging Theory and Practice: AI-Driven Insights into Manufacturing Evolution and Industrial Maintenance Innovation – Muntaser Nuttah
15.50 -16.00 Sum up and plan for the April seminar

Abstracts

Immersive Analytics for Understanding Ecosystem Services Data – Benjamin Powley
When planning land use decisions, the input from experts in various domains is often required when making the decision. Ecosystem services analysis is often performed by expert analysts to estimate the effect of land use changes on the environment. For example, farming provides the benefit of agricultural productivity, but can negatively impact ecosystem services by reducing biodiversity, or increasing the amount of nitrogen in waterways.

In this talk, immersive VR visualization system, Immersive ESS Visualizer, is presented. The visualization system was designed for the comparison of multiple ecosystem services across different land use change scenarios. A user study was performed to evaluate the effectiveness of Immersive ESS Visualizer for ecosystem services analysis tasks compared to existing media (paper maps, and PDF’s on a 2D screen). The results of the user study will be discussed.

Bridging Theory and Practice: AI-Driven Insights into Manufacturing Evolution and Industrial Maintenance Innovation – Muntaser Nuttah
“In today’s industrial landscape, artificial intelligence (AI) is critical for transforming data into actionable knowledge. This talk highlights two innovative studies that leverage AI to decode complex unstructured datasets. The first study employs Natural Language Processing (NLP), Large Language Models, and Dynamic Topic Modeling to conduct a large-scale review of over 35,000 publications in manufacturing digitalization and automation from 1970 to 2023. This approach not only structures a fragmented body of knowledge but also tracks thematic evolutions—from early simulation and scheduling studies to emerging trends in energy efficiency, composite materials, cybersecurity, robotics, and AI—offering empirical support to creative destruction and technological paradigm theories. Similarly, the second study transitions to practical application, demonstrating how NLP-driven text mining could be used to deal with unstructured maintenance logs, claims, and work orders from Volvo CE. By converting raw text into structured insights, the framework enables proactive maintenance planning, system optimization, and knowledge transfer—showcasing AI’s capacity to bridge data volume and expert interpretation in industrial settings.”

When? Friday March 7th 14-16
Where? Onsite: D1140 at Linnaeus University in Växjö and online
Registration: Please sign up for the PhD-seminar via this link https://forms.gle/kydkchwh92y9g2RC9 by March 5th (especially important if you plan on attending onsite so we have fika for everyone)

14.00-14.10 Welcome and practical information from Welf Löwe
14.10-14.55 Presentation and discussion: Producing the Next Generation of Forest Attribute Maps – the Swedish Case – Dag Björnberg
14.55 – 15.05 Coffee break
15.05 – 15.50 Presentation and discussion: Sound, Precise, Memory Efficient P2A and beyond – Mathias Hedenborg
15.50 -16.00 Sum up and plan for our seminars in April

Abstracts

Producing the Next Generation of Forest Attribute Maps – the Swedish Case – Dag Björnberg
Remote sensing techniques are widely used for mapping and monitoring forest attributes, providing valuable information on forest cover, biomass, and overall forest health. In recent years, national airborne laser scanning (ALS) campaigns have been conducted in several countries to map forest resources. When combining ALS data with field inventory data, these datasets enable the development of nationwide models for prediction of forest attributes. In this talk, we discuss the potential of machine learning (ML) to enhance existing modeling approaches for nationwide forest attribute mapping in Sweden, and show prediction results on five forest variables.

Sound, Precise, Memory Efficient P2A and beyond – Mathias Hedenborg
Points-to analysis can be used as a helping tool, but then it needs to be sound, fast, and precise.
The Points-to information can be useful in Compiler Optimization and Software Engineering.

In this thesis, an approach is presented that fulfills all of these requirements. The approach is flow-sensitive since it is an SSA-based data-flow analysis.
By using X-terms (chi-terms) for saving context data, the approach will be context-sensitive.

We describe how the soundness is reached, by relate the use of X-terms to a conservative data-flow analysis.
The proof will show that the steps in creating X-term based representation will guarantee the soundness, if the conservative data-analysis is sound.

We will also show that the use of X-terms out-range other traditional representation for the context information needed.

There will also be a discussion about the precision in a system using X-terms.

In addition to this, the thesis discusses how points-to analysis can be used in other areas like program/system understandability and Compiler Optimization.
Future work will point out areas like result prognosis, alias, reachability, security, and more areas related to Software Engineering.

When? Friday December 13th 14-16
Where? Onsite: D2272 and via zoom
Registration: Please sign up for the PhD-seminar via this link by https://forms.gle/94Gb6pGdQ5qj2BeD7 December 11th (especially important if you plan on attending onsite so we have fika for everyone)

Agenda

14.00-14.10 Welcome and practical information from Welf Löwe
14.10-14.55 Presentation and discussion: The deterministic pancake forest – Jonas Nordqvist
14.55 – 15.05 Coffee break
15.05 – 15.50 Presentation and discussion – “Will It Hold? Predicting the Joinability of Metals Before Welding Them” and “A Foundational Approach for Fine-Grained Commit Quantification” – Sebastian Hönel
15.50 -16.00 Sum up and plan for the spring seminars

Abstracts

The deterministic pancake forest – Jonas Nordqvist

In this talk, we will discuss a classical problem in computer science, namely sorting by prefix reversal. Its more popular name, Pancake Sorting, aside, it is actually more than just a toy problem. The long-standing question is: given a list of length n, what is the minimal number of prefix reversals needed to sort it? However, in the 70s, Conway proposed that one might study a deterministic version of this problem. Doing so, the problem, formulated as a discrete dynamical system, gives rise to an adjacency graph that is a collection of trees, i.e., a forest; more precisely, the deterministic pancake forest. Besides discussing the problem in general, I will present some results on the pancake forest and how this relates back to the original problem.

Will It Hold? Predicting the Joinability of Metals Before Welding Them – Sebastian Hönel

In the context of automotive applications, a common task is to join two or more parts, such as sections of a car’s frame.The joining of dissimilar metals presents a critical challenge in automotive manufacturing due to the differing thickness, as well as thermal, mechanical, and electrical properties of the base materials.

The challenge further lies in joining a varying number of materials reliably, that is, obtaining a joint that is sufficiently large and stable. Extensive laboratory tests using spot welding were conducted to gather an understanding of which materials using which parameters can be welded together. However, performing these tests is costly and trials need to be repeated multiple times to get robust and dependable estimates.

This study focuses on A) establishing a probabilistic understanding of selected parameters, materials, and welding outcome, and B) prediction of joint quality given the desired materials and parameters. To address these challenges, we employ deep conditional density estimation in conjunction with regression models.

Some preliminary results show that predicting joint size is within a reasonable error of margin, especially since we have not yet considered material properties just yet. Furthermore, a conditional normalizing flow was able to accurately capture the joint density of our dataset, allowing us to estimate the probability that a joint is sufficiently stable and to efficiently oversample the underrepresented test cases.

A Foundational Approach for Fine-Grained Commit Quantification – Sebastian Hönel

Commits are sets of changed made continuously to a software repository. Understanding commits and the purpose behind them is crucial for a wider range of applications, such as commit classification, fault prediction and -localization, or automated commit message generation.
Extracting features from commits is and has historically been a challenging task. In the past, many studies were limited to commit metadata or human-engineered features specific to the downstream task at hand. Such features are almost always far inferior to semi- or unsupervised approaches used in representation learning.

With the recent advent of large language models (LLMs), the ability to largely capture the underlying (changed) source code in a commit has significantly improved. However, the inherent tree-like structure of a commit, together with a variable number of affected files, hunks, etc., which are also of variable length, poses a challenge for, e.g., regression- or discriminative models.

We attempt to alleviate these challenges once and for all by suggesting a foundational approach that consists of A) a language-agnostic, fine-grained, and multi-scale source code and metadata commit extraction, and B) a flexible deep-learning-based framework for the embedding, reduction, and projection of commits. The framework is agnostic with regard to the choice of LLM(s) and exploits transformers as well as recurrence-based architectures.

We evaluate our framework using an enhanced version of the downstream task of commit classification. We add uncertainty estimation which allows the trained model to quantify the risk of misclassification. The model exploits multiple-instance learning and optionally a stochastic version of what constitutes a commit to not only allow classification, but to also enable intent-disentanglement of merge- and ordinary commits and classification of fractional commits.

When? Friday December 6th 14-16
Where? Onsite: D2272 at Linnaeus University in Växjö and online
Registration: Please sign up for the PhD-seminar via this link https://forms.gle/vTTmpqc19hutU3Dg6 by December 4th (especially important if you plan on attending onsite so we have fika for everyone)

14.00-14.10 Welcome and practical information from Welf Löwe
14.10-14.55 Presentation and discussion: Digital twin development for wheel loader – Manoranjan Kumar, industrial PhD-student Volvo CE
14.55 – 15.05 Coffee break
15.05 – 15.50 Presentation and discussion: Designing for thinking and engagement: challenges of teaching and learning Computational Thinking in K-12 – Rafael Zerga, PhD student LNU
15.50 -16.00 Sum up and plan for our seminars during the spring semester

Abstracts

Digital twin development for wheel loader – Manoranjan Kumar, industrial PhD-student Volvo CE

The need to virtually understand the machine usage is an important step in building the digital twin framework of a wheel loader (WL). Volvo Construction Equipment (VCE) has developed such a framework which includes data logging, complete vehicle simulations, and data analytics. Co-simulation is used in complete vehicle simulation to increase the simulation data accuracies. The framework also supports a variety of operator-driving simulations to mimic the real operator’s behaviors. This is achieved by integrating the operators’ model of the WL and its interaction with the power source model, i.e., the drive train, the hydraulics, and the material. The validation is done using real measurements which shows a good accuracy of the simulation. The results will be very useful for engineers in product development to improve WL design and controls using digital twins. The successful validation of the framework also paves the way for future research to enhance the virtual simulation techniques.

Designing for thinking and engagement: challenges of teaching and learning Computational Thinking in K-12 – Rafael Zerga, Phd student LNU

Computational Thinking is an approach for effective problem-solving which is being incorporated in the study curriculum of K-12 education in several countries in different regions of the world. Programming is considered a relevant skill in our digital society as it facilitates the process of solving problems. Sweden has introduced the teaching of programming in the subject matter of Mathematics and Technology since 2018. As technology advences newer and more natural ways of user interfaces come along which let the user interact with the computer in easier and more intuitive ways. The introduction of visual programming methods such as block-based programming has made a big impact in the way young students build algorithms without the need to learn complex programming syntax. However, students are still facing some challenges when learning basic programming concepts such as conditionals, variables and logic operators. The advent of emerging technologies such as generative AI based on the use of large language models (LLM) could allow for an even more natural form of interaction where the student would define algorithmic instructions using natural language. This approach to programming could increase the level of engagement in students when doing programming and it could facilitate a higher level of thinking in the process of solving a given problem, which is the essence of Computational Thinking.

When? Friday November 15th 14-16
Where? Onsite: D2272 and via zoom
Registration: Please sign up for the PhD-seminar via this link by https://forms.gle/GdaiE6W6J1RLPWa7A November 13th (especially important if you plan on attending onsite so we have fika for everyone)

Agenda

14.00-14.10 Welcome and practical information from Welf Löwe
14.10-14.55 Presentation and discussion: Tower-based radar observations of sub-daily water dynamics in boreal forests – Johan Fransson
14.55 – 15.05 Coffee break
15.05 – 15.50 Presentation and discussion – Enhancing Forest Attribute Prediction Using ResNet and DeepLab Architectures with Airborne Laser Scanning Data – Shafiullah Soomro
15.50 -16.00 Sum up and plan for the December seminar

Abstracts

Abstracts
Tower-based radar observations of sub-daily water dynamics in boreal forests – Johan Fransson

Radar remote sensing observations are predominantly affected by the concentration and spatial distribution of water in natural scenes. This motivates the utilization of high-resolution spaceborne radar observations for monitoring the water status of vegetation and the impacts of climate change on forests globally. While current satellite-based synthetic aperture radar observations are limited to temporal resolutions of days, tower-based radar observations of forests are capable of capturing detailed sub-daily physiological responses to variations in soil water availability and meteorological conditions. Such experiments demonstrate the scientific value of prospective sub-daily space-borne observations in the future.

The BorealScat tower-based radar experiment conducted in southern Sweden from 2017 to 2021 has captured various ecophysiological phenomena in a boreo-nemoral forest, including water stress and degradation induced by spruce bark beetles (Ips typographus). To gain a deeper insight into the sub-daily impacts of forest water dynamics on radar observations, the BorealScat-2 tower-based radar experiment was initiated in a boreal forest, located in northern Sweden in 2022. Along with in-situ sensors characterizing the water status on the tree level and an eddy-covariance flux tower, this initiative aims to compile a comprehensive and open dataset. The goal is to enhance our understanding and modelling of the relationship between traditional ground-based forest information, eddy-covariance flux measurements and radar remote sensing observables.

The data gathered by BorealScat-2 stands out as the most radiometrically precise high-resolution time series ever recorded in forest environments, resolving the subtle water content-induced signatures in radar measurements. Preliminary findings from the 2022 growing season, highlight the detectability of a diurnal radar signature across all conventional radar remote sensing bands (i.e. C-, L- and P-band). Moreover, metrics akin to tree water deficit, as measured by high-resolution point dendrometers, can be derived from interferometric radar observations. The fine temporal resolution of the data also unveils distinct signatures corresponding to intercepted precipitation in time series measurements. These findings underscore the need for sub-daily observations from space-borne satellites to monitor vegetation water status.

Enhancing Forest Attribute Prediction Using ResNet and DeepLab Architectures with Airborne Laser Scanning Data – Shafiullah Soomro

This study explores the application of advanced deep learning architectures, including ResNet and DeepLab, in conjunction with Airborne Laser Scanning (ALS) data for predicting forest attributes in Sweden. Utilizing a high-precision Digital Elevation Model (DEM) generated from ALS surveys conducted between 2016 and 2020, we integrated raster data and laser metric data, including point clouds, RGB imagery, and infrared imagery. We employed pre-trained model architectures, leveraging Transfer Learning to enhance model performance on a dataset comprising approximately 18,435 plots from the Swedish National Forest Inventory (NFI). The models were trained to predict key forest metrics such as stem volume, basal area, mean tree height, and mean stem diameter. Performance was evaluated through Root Mean Square Error (RMSE) calculations, revealing significant advancements over traditional modeling approaches. The results underscore the potential of employing deep learning techniques for improved forest planning and management in Sweden.

When? Friday October 18th 14-16
Where? Onsite: D2272 and via zoom
Registration: Please sign up for the PhD-seminar via this link https://forms.gle/CLBLYvcFgFSXAkBr8 October 17th (especially important if you plan on attending onsite so we have fika for everyone)

Agenda

14.00-14.10 Welcome and practical information from Welf Löwe
14.10-14.55 Presentation and discussion: Design of an intelligent wearable for activity and health – the DIWAH study – Patrick Bergman
14.55 – 15.05 Coffee break
15.05 – 15.50 Presentation and discussion – Weakly supervised learning for dendritic cells image-segmentation – Jorge Lazo
15.50 -16.00 Sum up and plan for the November seminar

Abstracts

Design of an intelligent wearable for activity and health – the DIWAH study – Patrick Bergman

The overarching goal of the DIWAH-study is to create algorithms that can be implemented in welfare technology, specifically wearables, for self-monitoring and use within in the healthcare system. By utilizing the computing power of artificial intelligence (AI), we will develop and validate algorithms to assess physical activity (PA), energy expenditure (EE), and blood pressure (BP) at an individual level. Building on that information will develop self-learning AI algorithms that provides tailored PA recommendations on the appropriate dose to optimize an individual’s BP in real-time and without human intervention. This is a new approach since it not only gives information of the outcomes separately but also on the effect of PA has on health in real time. In the long run the AI will learn from its user and be able to provide personalized tailored activity advice to optimize the individual’s health. This has not been studied before, but the field has now reached a point where the technology exists along with processing power and analytical tools. Therefore, it is time to fully explore the possibility of combining real-time data on PA, health-related outcomes, and AI, so that future citizens can maintain or improve their health by making informed decisions based on personal data. For the older individuals in society this is extra important since they are at a higher risk of developing physical, mental, and social health issues where a maintained or increased PA-level is known to prevent and reduce the effects of such health problems. From a societal perspective a change from a reactive to a proactive health care system is necessary considering the demographic shift towards an aging population and an expected increased load on the healthcare system. Thus, it is necessary to develop evidence-based strategies that can relieve the healthcare system by promoting health, preventing diseases, and treating people in the best possible way when they do get sick. However, the current wearables available do not produce valid output especially among elderly individuals, therefore in this proposal we will

1. Identify and adapt open-source wearables suitable for assessing PA, energy expenditure and blood pressure
2. Collect data from the wearables in a controlled environment, using AI develop algorithms and to compare them against golden standard method
3. Test the developed algorithms during free-living and compare them against golden standard method

Weakly supervised learning for dendritic cells image-segmentation – Jorge Lazo

Manual detection and classification of immune cells in In-Vivo Confocal Microscopy (IVCM) images is a highly time-consuming task, and prone to subjective decisions and levels of expertise; therefore, it becomes a bottleneck in the detection and assessment of different ophthalmic disorders.

In the last few years, Deep-Learning (DL) based methods have been explored for the task of dendritic-cell segmentation in IVCM images, with fully-supervised strategies as the most common ones. These methods, despite showing promising results, rely on the availability of a large amount of detailed pixel-level labels, necessary to train the DL models.

Weakly supervised image segmentation is a DL technique where models are trained to segment images into meaningful regions or objects using limited, imprecise, or incomplete labels. In contrast to fully supervised learning, where detailed pixel-level annotations (or “masks”) are provided for every image, weakly supervised methods rely on weaker forms of supervision. Foundation models have made their appearance in the DL landscape offering powerful feature extraction and transfer learning capabilities. Even though their performance in medical imaging applications is still far from optimal, they offer a powerful tool for generating pseudo-labels that could be used in weakly supervised set-ups for training specialized models.

By utilizing weak supervision, these models can reduce the burden of manual annotation in medical imaging while still achieving high performance in identifying and segmenting key structures like dendritic cells.

When? Friday September 20th 14-16
Where? Onsite: D0073 at Linnaeus University in Växjö and online
Registration: Please sign up for the PhD-seminar via this link https://forms.gle/h8oQ9VVYJUFu89i77 by September 18th (especially important if you plan on attending onsite so we have fika for everyone)

Agenda

14.00-14.10 Welcome and practical information from Welf Löwe
14.10-14.55 Presentation and discussion: Improving medication safety though the collaboration between researchers from medicine and computer science – Tora Hammar
14.55 – 15.05 Coffee break
15.05 – 15.50 Presentation and discussion – A machine learning approach to improving drug risk assessment – Daniel Nilsson
15.50 -16.00 Sum up and plan for the October seminar

Abstracts

Improving medication safety though the collaboration between researchers from medicine and computer science – Tora Hammar
In my presentation I will talk about our research project where we use health data to see if we can predict medication related problems and improve predictions compared with the current clinical decision support systems (CDSS) used in health care. The project is an interdisciplinary collaboration between researchers from medicine, pharmacy and computer science.

Medication usage and the simultaneous use of many medications is increasing world-wide. Problems with side-effects (adverse drug events) are common and cause suffering and even death, as well as substantial costs for society. One method to prevent harmful combinations of medications is by using CDSS in health care or at pharmacies that can detect potential ADEs. Todays CDSS are often based on rules written by humans (so called knowledge databases). Although we have high quality knowledge databases in Sweden these systems have known weaknesses such as having to many non-relevant alerts causing alert fatigue among users. One reason is that the rules are often very simple, another reason is that they have not been validated in large populations. In our research project we use data from health care in the region of Kalmar County for over ten years of time, including data on all medications being used during that time. We also have the rules and algorithms from the Swedish knowledge databases called Janusmed which is used in health care to give warnings about potentially harmful combinations of medications. In the project we aim to:

• Increase knowledge of effects when combining many different medications
• Study how well current CDSS can predict adverse drug events (Spoiler alert! Not very well.)
• Examine if we can improve predictions compared with the current CDSS by using machine learning.
• Develop methods to identify adverse drug events in clinical notes (unstructured data) using large language models.

Tora is an associate professor in health informatics at the eHealth Institute at Linnaeus University. She is a pharmacist, with a master and PhD in Biomedical Science. In her research she is using different methods to improve information systems and decision support in the medication management process. Much of her research is done in collaboration with computer science as a part of LnuC DISA, and she is the research leader for the DISA eHealth group.

Daniel Nilsson who is presenting after Tora is working in the research project Tora is presenting and will dive deeper into some of the questions.

A machine learning approach to improving drug risk assessment – Daniel Nilsson

Many medications are associated with adverse side effects. An understanding of what factors influence the risk of adverse drug events is important for managing this risk. In this presentation I will describe the results of a project to investigate ways to improve risk assessments for two categories of adverse drug events (bleedings and QT-prolongation) compared to the currently used knowledge database Janusmed. Using data on adverse drug events from the healthcare information system from Kalmar region (comprising ten years of event data, and ~280 000 patients), we seek to use machine learning methods to answer questions such as:

• Do the Janusmed risk values provide predictive information?
• Can we combine the Janusmed risk values (which only contain information of current medication) with demographic information and additional health data to improve predictions?
• Are the risk values assigned to different medications in Janusmed in alignment with the risks observed in the data?

Daniel is working in the DISA eHealth group as part of the AI Sweden program Eye for AI. He has a PhD in computational biology from Lund University, where he studied computational methods for protein simulation.