DISA

When? November 27, 14.00-16.00
Where? Online, https://lnu-se.zoom.us/j/61812638609?pwd=zy5Z2oFW3lhVh3C0bqYAqBJFpRNwnm.1&from=addon

Abstracts

Mahalanobis Distance and Angle under a SCAD-Dual Modified-Cholesky Precision Estimator: The Gaussian case – Deliang Dai from Department of Economics and Statistics
We study finite-sample and asymptotic properties of Mahalanobis distances and angles computed with a row-separable modified-Cholesky precision estimator that combines SCAD selection and a positive-semidefinite quadratic penalty. For an independent Gaussian observation, we give the exact conditional generalized-chi-square distribution of the estimated squared distance, a finite-sample deviation inequality, and a central limit theorem on the $\sqrt{2p}$ scale. For the Mahalanobis cosine between two deterministic vectors, we obtain a finite-sample perturbation bound and an asymptotic normal approximation with an explicit direction-dependent Gaussian quadratic-form variance. Separate conditions permit centering at the true cosine and feasible studentization.  The estimator is defined by two local-linear-approximation updates, whose agreement with an oracle path is established under explicit initialization, signal-separation and tuning assumptions. Simulation and empirical studies are presented to evaluate the theoretical results. 

Nonlinear forecasting with many predictors using mixed data sampling kernel ridge regression models – Kristofer Månsson from Department of Economics and Statistics
Policy institutes such as central banks need accurate forecasts of key measures of economic activity to design stabilization policies that reduce the severity of economic fluctuations. Therefore, this paper develops a kernel ridge regression estimator in a mixed data sampling framework. Kernel ridge regression can handle many predictors with a nonlinear relationship to the target variable. Consequently, it has potential to improve the currently used principal component-based methods when the economic data follow a nonlinear factor structure. In a Monte Carlo study, we show that the kernel ridge regression approach is superior in terms of mean square error and is more robust than principal component-based methods to different nonlinear data generating processes. By using a dataset consisting of 24 economic indicators, we forecast Swedish gross domestic production. The results confirm the superiority of the kernel ridge regression approach. Therefore, we suggest that policy institutes consider the use of kernel-based approaches when forecasting key measures of economic activity.

We have an exciting few weeks ahead, with no fewer than three public PhD defenses on the horizon. We warmly invite you to join us for these important academic events and celebrate the accomplishments of our doctoral candidates. 
Farid Edrisi, 17 November  
On 17 November at 1:00 p.m., Farid will defend his doctoral thesis, Engineering Proactive Self-Adaptation in Cyber-Physical Systems under Uncertainty. The defense will take place in room N1017, N Building, at Linnaeus University in Växjö. 

Dag Björnberg, 20 November  
On 20 November, at 9:30 a.m., Dag, who combined his doctoral studies with his role at Softwerk through our industrial graduate school, DIA, will defend his thesis Machine Learning under Data Scarcity: Methods and Applications in Forestry. The public defense will take place in Newton, Building C, at Linnaeus University in Växjö. 

Rakhshanda Jabeen, 4 December   
On 4 December at 9:15 a.m., Rakhshanda, who pursued her doctoral studies in collaboration with Electrolux within DIA, will defend her thesis Representing Architectural Evidence in Software Systems: A Graph Learning Approach. The defense will take place in Room D1136, Building D, at Linnaeus University in Växjö. 

When? October 23, 14.00-16.00
Where? D2272 or online, https://lnu-se.zoom.us/j/61126230818?pwd=2ZJJtYeoTNmKaweBshZDLbN0BradBq.1&from=addon

Abstracts

Self-Adaptive Federated Domain Generalization for Privacy-Preserving Fault Diagnosis – Mehdi Saman Azari
Fault diagnosis is at the core of predictive maintenance, helping prevent costly breakdowns in industrial machines. Deep learning has made diagnosis more accurate, but it works best when a machine operates under the same conditions it was trained on. In practice, machines run at different speeds and loads, and they often meet conditions that never appeared in the training data. Since no data from these unseen conditions is available in advance, the model has to generalize to them without having seen them. One way to achieve this is to train on data from many machines, but gathering all data in one place conflicts with privacy, intellectual-property and cybersecurity requirements. Federated learning lets machines train a model together without sharing raw data, yet current methods lose accuracy compared with centralized training and cannot adapt on their own during operation.

In this talk, I present SAFDG, a framework that combines federated learning with domain generalization techniques and a self-adaptive control loop (MAPE-K). Federated domain generalization helps the model learn fault patterns that stay the same across working conditions, so it can handle conditions it has never seen without have a direct access to raw data of other machine. The control loop lets each machine monitor its own state and decide by itself whether to update its model locally, when it has enough data, or together with other machines, when it faces a new condition with too little data. Evaluated on four bearing datasets under 24 unseen conditions, SAFDG outperforms existing privacy-preserving methods and comes within 3 percentage points of centralized training.

Visual Analytics for Large Language Model (LLM) Evaluation – Darius Coelho
As large language models (LLMs) continue to advance, evaluating their reasoning at scale remains a key challenge. We introduce a visual analytics framework that combines automated evaluation with interactive exploration to help researchers assess model performance. By supporting simultaneous analysis across multiple models and benchmark datasets, our approach enables scalable comparisons, pattern discovery, and targeted case investigations to better understand LLM behavior.

Over three intensive days, Linnaeus University was a hub for research, digitalisation and emerging technologies as symposia, workshops, and conferences brought together altogether 150 researchers, PhD students, and representatives from industry. 

The programme included the 12th Big Data Conference, organised by the Linnaeus University Centre for Data Intensive Sciences and Applications (DISA). This year, the conference was held together with the International Symposium on Digital Transformation (ISDT), creating an even broader meeting place for researchers and professionals working with data, digital transformation, and emerging technologies. 

From AI and large language models to visualization, e-health and digital transformation, the days offered a wide range of perspectives and engaging discussions. 

Above all, the events created opportunities for people to share ideas, be inspired by each other’s research, and discover new possibilities for collaboration. Read more about three inspiring and eventful days at Linnaeus University. 

Read the whole article here

When? August 21, 14.00-15.30
Where? D2272 or online https://lnu-se.zoom.us/j/68245999256?pwd=diVc4g6cCuanYHH5aN1lrnGMUZzkuW.1&from=addon

Abstract

A Geometric Perspective on Resolving the Likelihood Out-of-Distribution Paradox – Sebastian Hönel

Deep generative models can assign higher likelihood to supposedly out-of-distribution (OOD) samples than to in-distribution data. This apparent likelihood-OOD paradox is often interpreted as a failure of density estimation or falsely attributed to model pathologies. A geometric decomposition instead reveals several distinct mechanisms: a failure to distinguish between likelihood and typicality, the effect and dominance of high codimension that gives an increasingly transverse answer to the intrinsic likelihood question, and the non-identifiability of ambient likelihood; the latter of which is predominantly caused by ignorance.

The seminar includes a primer in differential geometry; introducing, for example, tangent and normal spaces, projectors, immersions and embeddings, together with strong and weak forms of the manifold hypothesis, to distinguish intrinsic manifold densities from ambient densities. We then introduce the tubular likelihood decomposition of the log ambient likelihood as the sum of an intrinsic-density term, a normal-space term, and a volume-change correction. This decomposition serves as an exhaustive oracle likelihood. It shows that ambient likelihood may be dominated by extrinsic variation, even when the intended OOD notion concerns intrinsic (semantic) variation, causing a (likely unexpected) likelihood reversal.

A central consequence exposed by the tubular decomposition is likelihood non-identifiability. When any of the intrinsic law, decoder (representation), or normal-space model is unknown, observations alone do not identify a unique ambient likelihood. Distinct representations and normal-space completions can agree on observed or near-manifold data while inducing substantially different likelihood values and rankings.

These effects are examined in controlled settings based on injective procedural decoders and differentiable image renderers. The generators provide known intrinsic variables, deterministic decoder Jacobians, projected normal displacements, and analytically specified ambient likelihoods. This enables decomposition of ground-truth likelihood into intrinsic, normal, and geometric components, construction of controlled likelihood reversals, and direct comparison between ground-truth ambient likelihoods and likelihoods learned by deep generative models, such as normalizing flows. The resulting framework clarifies what likelihood measures, when ambient likelihood is identifiable, and why likelihood-based OOD scores can fail even when ambient density estimation is formally well-defined.

When? 5 June, 14.00-16.00
Where? Onsite: D2272 and via zoom

Abstract

Engineering Proactive Self-Adaptation in Cyber-Physical Systems under Uncertainty – Farid Edrisi
Cyber-Physical Systems (CPS) are characterized by their complex interactions between computational and physical processes, uncertainty, and continuous change. Engineering such systems is challenging because their operational circumstances cannot be fully characterized. Self-adaptation was introduced to address uncertainty and continuous change. Most existing self-adaptation approaches are reactive; the system monitors for goal violations and initiates corrective action once a violation is detected. In CPS, however, some actions are irreversible, and delayed or incorrect responses can cause safety issues. This becomes even more challenging under runtime uncertainties, such as incomplete or inaccurate information during operation, where the system may only react to visible symptoms while the cause remains undetected, leading to repeated violations instead of stable operation. 

To deal with these limitations, this thesis advances the engineering of proactive self-adaptation in CPS under uncertainty. Rather than responding to violations, proactive adaptation shifts the decision point ahead of the violation. It anticipates how conditions may evolve and acts while corrective actions are still effective and options remain open. This also gives the system the flexibility to refine adaptation decisions as new runtime information becomes available. 

Following Design Science Research methodology, the thesis introduces three architectural contributions. PANCS is a reusable reference architecture that integrates Adaptive Model Predictive Control into the MAPE-K loop, enabling anticipatory adaptation in nonlinear CPS by reasoning predictively within physical and temporal constraints while continuously refining its models as system dynamics evolve. A Digital Twin-augmented MAPE-K architecture extends this capability to the managing system level, introducing a continuously updated virtual replica of the complete system (managed system, managing system, and execution environment) that supports runtime revision of adaptation rules and reasoning under some conditions that were never anticipated at design time. The EA Blueprint Pattern addresses the collaborative CPS scale by providing an architectural pattern for constructing and evolving digital twin of the organization from existing enterprise architecture models, enabling proactive reasoning about structural dynamics, interdependencies, and organizational change at an abstraction level where dynamic models are insufficient. 

The contributions are demonstrated and evaluated across multiple CPS domains, including autonomous ground vehicles, robotic manipulators, and collaborative CPS settings. Results show that under proactive self-adaptation, when supported by predictive reasoning, continuous model maintenance, and principled architectural structure, operational integrity can be effectively sustained over time as system and environmental conditions evolve.

When? May 29, 14.00-15.00
Where? Onsite: D2272 and via zoom
Registration: Please sign up for the seminar via this link https://forms.gle/VXDXziHks8ZYC7kz5
by May 27. This is especially important if you plan to attend onsite so we can make sure there is fika for everyone.

Agenda
14.00-14.10 Welcome and practical information
14.10-14.55 FAVE: A Visual Analytics System for Urban Accessibility Fairness – Parisa Salmanian
14.55 – 15.05 Coffee break
15.05 – 15.50 Security and Digital Forensics for Industrial IoT and Connected Systems – Mohamed Eldefrawy
15.50 -16.00 Sum up and plan for our upcoming seminars

Abstract

FAVE: A Visual Analytics System for Urban Accessibility Fairness – Parisa Salmanian
Who has fair access to essential daily services such as schools, hospitals, and grocery stores, and who does not? In rapidly growing cities, answering this question requires more than aggregate statistics; it requires methods that reveal spatial inequalities across multiple urban scales. This talk presents FAVE (Fairness Accessibility Visual Explorer), an interactive visual analytics system for analyzing urban service accessibility fairness in Swedish cities. FAVE combines gravity-based and distance-based accessibility models with inequality measures, including the Gini coefficient and generalized entropy, in a coordinated interface that also integrates dimensionality reduction, parallel coordinates, and explainable boosting machines. 

The system supports the exploration of accessibility patterns at the district, neighborhood, and building levels, and enables exploratory what-if analysis through urban scenario editing, including additions with the ability to adjust attributes such as footprint, floor area, number of floors, floor height, and shape, modification of POIs and buildings, and changes in building type. It also incorporates LLM-based explanations and intervention suggestions. We demonstrate FAVE using building and OpenStreetMap data together with official statistics from Statistics Sweden (SCB) for Växjö and other Swedish municipalities. The results show how the system reveals spatial inequalities in access to healthcare, education, and other essential services, supporting more informed and evidence-based urban planning. 

Security and Digital Forensics for Industrial IoT and Connected Systems – Mohamed Eldefrawy
Mohamed Eldefrawy will present his current research activities with a focus on Industrial Internet of Things (IIoT) security, digital forensics, lightweight authentication, and secure connected systems. He will also introduce the KK-stiftelsen-funded CyberGuard Academy project, which aims to strengthen practical competence in cybersecurity and digital forensics in collaboration with Swedish industry. The talk will highlight selected past and ongoing work on formal security analysis, lightweight authentication protocols, Internet of Vehicles security, and digital forensics. It will also outline possible areas for future collaboration between Halmstad University and Linnaeus University.

Short bio:
Mohamed Eldefrawy is a Senior Lecturer in Cybersecurity at the School of Information Technology, Halmstad University. His research focuses on IIoT security, lightweight authentication, formal security analysis, and digital forensics. He is currently PI of the KK-stiftelsen-funded CyberGuard Academy project and has published more than 30 peer-reviewed papers. He has served as Guest Editor for IEEE Transactions on Industrial Informatics and Big Data and Cognitive Computing and holds two US patents.

When? May 22, 14.00-16.00
Where? Onsite: D2272 and via zoom
Registration: Please sign up for the seminar via this link, https://forms.gle/6NV1b1tjJ2zUndcn8 by May 20. This is especially important if you plan to attend onsite so we can make sure there is fika for everyone.

Agenda
14.00-14.10 Welcome and practical information
14.10-14.55 Multiscale Mathematics for Complex Systems: From Individuals to Digital Twins – Wolfgang Bock 
14.55 – 15.05 Coffee break
15.05 – 15.50 Multimodal Assessment of Cognitive Load in Technology-Enhanced Learning-Julia Bend
15.50 -16.00 Sum up and plan for our upcoming seminars

Abstracts

Multiscale Mathematics for Complex Systems: From Individuals to Digital Twins – Wolfgang Bock 
Complex systems are often governed by mechanisms acting simultaneously on several spatial, temporal, and organizational scales. In such settings, neither a purely microscopic description nor a purely aggregate model is sufficient on its own. The key difficulty is to relate local interactions, heterogeneity, and stochasticity to robust macroscopic behavior. 

In this talk, I will discuss the mathematics behind multiscale models for complex systems, focusing on the transition from microscopic agent-based or synthetic population descriptions to effective system-level dynamics. This perspective is particularly relevant in applications where one aims to combine structural modeling, uncertainty, and data, for example in the development of digital twins. Synthetic populations serve here as a useful modeling framework in which individual variability and interaction patterns can be represented explicitly, while also raising mathematical questions about emergence, approximation, and the identification of relevant scales. 

The talk is intended as a conceptual introduction to this multiscale viewpoint. Rather than emphasizing technical formalism, I will focus on the key mathematical principles that help make such complex models understandable, interpretable, and computationally tractable. 

Multimodal Assessment of Cognitive Load in Technology-Enhanced Learning-Julia Bend
Technology-enhanced learning (TEL) systems increasingly incorporate adaptive features aimed at improving learning outcomes and supporting individual learners’ needs. However, adaptive learning requires reliable methods for assessing learners’ cognitive states during interaction with digital learning environments.

This presentation focuses on multimodal assessment of cognitive load using eye tracking, EEG, and functional near-infrared spectroscopy (fNIRS). Drawing on cognitive load theory and load theory of selective attention, the work examines how perceptual load and working memory load influence visual attention and cognitive processing during learning tasks. Eye-tracking metrics, such as fixation duration, fixation rate, and blink-related measures, provide behavioral indicators of attentional allocation and processing effort, while EEG and fNIRS offer complementary insight into neural dynamics and cortical activation associated with attention, working memory, and cognitive load.

The presentation discusses current and ongoing work related to distinguishing different types of cognitive load, investigating individual differences in cognitive processing, and exploring how multimodal measures can contribute to adaptive learning systems. Particular attention is given to the potential of combining behavioral and neural measures to support the development of more personalized and cognitively efficient educational technologies.

When? April 10, 14.00-16.00
Where? Onsite: D2272 and via zoom
Registration: Please sign up for the seminar via this link https://forms.gle/gnEjrVfnZRtJgjKYA  by April 8. This is especially important if you plan to attend onsite so we can make sure there is fika for everyone.

Agenda
14.00-14.10 Welcome and practical information
14.10-14.55 Architecting Carbon-Aware Software-as-a-Service – Samuele Giussani
14.55 – 15.05 Coffee break
15.05 – 15.50 Digital product passports: A value perspective – Timmy Öberg
15.50 -16.00 Sum up and plan for our upcoming seminars

Abstracts

Architecting Carbon-Aware Software-as-a-Service – Samuele Giussani
Software-as-a-Service (SaaS) solutions are increasingly adopted due to their scalability, cost-effectiveness, and ability to support rapid deployment while ensuring high availability and flexibility. However, carbon emissions are still rarely treated as a first-class design concern. Existing approaches predominantly focus on post-deployment measurement and mitigation, overlooking the fact that architectural decisions determine a significant share of a system’s environmental impact. This is particularly evident in multi-region and multi-tenant settings, where carbon intensity varies across locations and over time, and tenants contribute unevenly to overall resource consumption. Supporting architects in evaluating these factors early in the design process requires systematic reasoning mechanisms that can combine heterogeneous architectural, infrastructural, and operational information into actionable insights.

In this talk, the conceptual foundations of a Carbon-Aware Reasoning Framework are presented, enabling the design-time evaluation of multi-tenant SaaS architectures with respect to both carbon impact and performance-related quality attributes. A model-driven, tool-supported implementation of the Framework is then introduced, illustrating how carbon-aware reasoning can complement traditional architectural evaluation practices and inform environmentally conscious deployment decisions.

Digital product passports: A value perspective – Timmy Öberg
This presentation provides insights into an ongoing systematic mapping review of digital product passports (DPPs) from a value and ecosystem perspective. Based on existing and emerging classification schemes, the presentation maps the DPP research landscape and discusses current trends. The study explores where current DPP research is focused, including publication venues, research methods, contribution types, and industry sectors. It further analyzes the value dimensions the literature address, including value outcomes, value creation capabilities, and the value chain actors considered.