Assessment and feedback activities in higher education are undergoing significant changes. Many universities and institutes still rely on traditional testing and grading methods, which often fall short in supporting meaningful student learning, especially in large classes. Although educational policies, such as those promoted by the Bologna process, encourage more feedback-oriented and student-centered approaches, these practices can be difficult to implement due to time constraints and limited resources. Generative Artificial Intelligence (GenAI), particularly Large Language Models (LLMs), has shown strong potential in addressing these challenges. This review examines 21 research studies published between 2023 and 2025 that explore the use of GenAI in providing feedback and assessing student work in higher education, with some studies also comparing GenAI's performance with human instructors. Findings show that LLMs can generate personalized and constructive feedback and/or assist with fair and consistent assessment. However, in most studies, teachers still play a key role, as expert oversight is essential to ensure that grading assessments and assignment feedback are accurate, relevant, and aligned with learning objectives. For GenAI to be used effectively, educators need to understand how to work with these tools, such as learning GenAI prompt design and the basic principles behind LLMs. We recommend that academic institutions provide training for educators in AI literacy, prompt engineering, and the development of teaching strategies that combine the strengths of human judgment with AI support. By effectively integrating LLM tools, major assessment challenges, such as limited time and inconsistent feedback quality, can be addressed while also enhancing student learning and engagement. 

This study presents a statistical analysis of train delays in the Swedish railway system. The focus of the study is to identify the best-fitting probability distributions for train arrival times across different stations and travel directions. Using the Kolmogorov-Smirnov (K-S) test, we evaluate the goodness of fit for common distributions—gamma, log-normal, and inverse Gaussian—to capture delay patterns at ten stations. Our findings reveal significant variability across stations, with the log-normal distribution providing the best fit for 70% of cases. However, some stations exhibited direction-specific deviations, emphasizing the need for localized analysis. Traditionally, train delays in Sweden have been assumed to be uniformly distributed across the network, an oversimplification frequently used in generating synthetic datasets for AI-based timetable rescheduling systems. This study challenges that assumption, demonstrating that delay distributions vary by station and direction. By incorporating station- and direction-specific modeling, our results contribute to the development of more accurate synthetic datasets. These insights support data-driven approaches to predictive modeling, operational efficiency improvements, and increased reliability in railway networks. Based on the best-fitting distributions identified through statistical testing, we generate synthetic data using maximum likelihood estimates and direct sampling. Our study systematically assesses the distributional characteristics of train arrivals across stations and directions in the southern Swedish railway network, aiming both to understand operational variability and to generate realistic synthetic data for AI-based rescheduling. Building on this analysis, our method produces datasets that preserve the statistical characteristics of real train delays, ensuring they are more suitable for training and evaluating AI-based rescheduling algorithms. 

Phishing has evolved into one of the most adaptive and damaging cybersecurity threats, continually reshaping itself to exploit human behaviour, technical vulnerabilities, and, more recently, advances in artificial intelligence. As attack vectors diversify from traditional email scams to sophisticated, multi-stage, AI-generated, and hybrid phishing campaigns, defending against them has become significantly challenging. This survey provides a comprehensive and contemporary examination of the phishing landscape, tracing its evolution, analysing real-world incidents, and contextualising its growing impact through global statistics. We introduce a unified, multidimensional taxonomy that categorizes phishing attacks into distinct categories, providing a clearer understanding of how new attack techniques operate and escalate. In parallel, we review a broader range of phishing detection strategies, from list-based, heuristic, and similarity-driven techniques to modern machine learning and deep learning approaches. While these methods have advanced detection capabilities, they continue to face significant constraints related to data privacy, scalability, and the rapid emergence of novel attack patterns. Motivated by these limitations, the survey highlights the growing relevance of Federated Learning (FL) as a privacy-preserving and collaborative paradigm for phishing detection. To the best of our knowledge, this is the first comprehensive survey to examine phishing defence through the lens of FL. In particular, we examine the role of FL in enabling decentralized, privacy-aware detection without exchanging raw data, compared to centralized training in terms of performance, privacy guarantees, resilience, and scalability. Drawing from this analysis, we offer valuable insights into critical research gaps and future directions for developing robust, scalable, and privacy-aware phishing detection solutions. 

Context: The COVID-19 pandemic permanently altered workplace structures, normalizing remote work. While flexibility has well-known benefits, critical evidence highlights challenges with fully remote arrangements, particularly for software teams.

Objective: This study investigates how employee resignation patterns evolved across different work modalities - onsite, remote, and hybrid - at Ericsson, a global developer of software-intensive systems.

Method: Using HR and exit survey data from 2016 to 2025 for technical roles in Ericsson Sweden, we analyze how tenure, onboarding modality, and reasons for resignation relate to attrition trends before, during, and after the pandemic.

Results: Our findings show a marked increase in resignations from mid-2021 to mid-2023, especially among employees with less than three years of tenure. Those onboarded during the fully remote period were significantly more likely to resign early, and this pattern persisted even after Ericsson enhanced remote onboarding practices. Conversely, retention improved after the introduction of structured hybrid work policies, and survivability curves returned to pre-pandemic levels.

Conclusions: Our findings suggest that fully remote onboarding, despite structural support, may hinder the formation of organizational attachment and belonging, increasing early attrition. In contexts like software engineering, where onboarding heavily relies on in-person mentoring, frequent peer interactions, and on-the-job training, selective return-to-office practices for new hires and their teams may help restore cohesion and long-term retention. These insights can guide HR leaders and policymakers in crafting post-pandemic work practices. 

Like a winding thread passing through tryings at risk, this book is my endeavour to make explicit the situatedness and responsibility of research and researchers in tro-uble — whether in the grand challenges of our age or in the very local challenges of survival. Efforts to promote more complex and integrated understandings of society in science, or of science as a political arena, are urgent when facing the incalculabilities of our late-modern spheres of society. There is no doubt that technologies co‑evolve out of interactions in specific contexts. This implies that responsibility for where and how technologies travel, and for what uses they serve, must be collective. No innocent position exists. The demand on us as producers of knowledge and technology is focused on the reality‑producing consequences of our research and thus places us right into the context of implication.

The frames of understanding are developed within feminist technoscience and are linked to practitioners and writers of Mode 2 knowledge production. How can feminist research, and other disciplines that take a critical view of science, mobilise the transformatory potential required?

Part I presents insights into the relocations needed in (onto-)epistemological infrastructures, and Part II outlines a positioning within the fields of feminist research and feminist technoscience. Part III discusses experiences and two political dimensions — research political initiatives to support feminist research, followed by reflections on the convergence of science and politics. Part IV offers examples of research in contexts of not only application but implication.

The rapid expansion of cyberspace is fundamentally reshaping human existence, social relations, spatial structures, and mechanisms of order formation. These transformations pose significant theoretical challenges to Confucian philosophy, which has traditionally been grounded in embodied individuals, stable relational networks, and community-based ethical orders. Drawing on the framework of Cyberism, this paper re-examines Confucian philosophy through four foundational dimensions—human, relations, space, and order—and analyzes how emerging socio-technical conditions, including digital humans, algorithmic mediation, hybrid virtual–physical environments, and platform governance, destabilize its underlying assumptions. We argue that these transformations do not render Confucian ethics obsolete; rather, they call for its reinterpretation and reconstruction within a cyber-enabled context. Building on this analysis, the paper proposes three conceptual pathways for the transformation of Confucian ethics in the digital age: digital rituality as a framework for regulating interaction order in platform environments; algorithmic benevolence as a normative orientation for embedding human-centered values into technological decision-making; and platform-based community as a model for reconstituting public good under conditions of data-driven social organization. By articulating these concepts, the study contributes to bridging classical ethical traditions and contemporary digital governance, and offers a Confucian approach to addressing the ethical challenges of cyber civilization.

Federated Learning (FL) is a promising distributed learning method that enables multiple clients to collaboratively train a shared model without sharing their raw data thus preserving privacy. However, in practical implementations, client data are typically non-independent and identically distributed (non-IID). This resulting in heterogeneous learning dynamics and unequal benefits across participants. Improvements in average global performance can mask performance degradation for disadvantaged clients, highlighting a structural fairness challenge in FL. This thesis argues that achieving fairness under non-IID FL requires explicit understanding and modeling of client behavioral heterogeneity rather than uniform aggregation of client updates. 

In addressing the issue of fairness in FL under data heterogeneity, the thesis first studies and analyzes clients' deviating behavior during the federated training process. An eccentricity-based approach is introduced to quantify deviations in local models and data representations within the global model, enabling systematic identification of atypical contribution and benefit patterns. The insights gained lay the foundation for our further research into developing novel, fairness-aware FL solutions for heterogeneous, distributed learning setups.

Then it proposes a fairness-aware aggregation framework called FeDABoost that adapts client influence based on local performance signals. By dynamically weighting client updates and adjusting local optimization to emphasize hard examples, the method reduces disparities across heterogeneous clients while maintaining competitive global performance. Later, the thesis introduces DEFFT, a clients distribution-aware framework that models latent similarities among clients through persistent grouping based on label distributions. Cluster-level models and hierarchical knowledge distillation integrate inter-client structure into the learning process, enhancing fairness metrics along with overall accuracy.

Across multiple benchmark datasets, the proposed approaches demonstrate that a principled way to modeling heterogeneity can lead to measurable improvements in fairness without compromising global performance. The three discussed studies together establish a structured framework for mitigating unequal benefits in FL under non-IID data distributions.

This article explores architectural space in residential care in Sweden during the Covid-19 pandemic, when residential care was subject to spatial strategies. Community, which is at the heart of the predominant collective care model and underpins the design of architectural space, was identified as a potential source of contagion and thus restricted by spatial measures. Lefebvre's spatial triad is the theoretical backbone of this study. The research design is an intuitive enquiry and qualitative data collection methods include interviews, observations and analysis of drawings. The lived space of staff and residents is primarily investigated in semi-structured individual interviews, and there is particular focus on their experiences of spatial strategies. Findings show that the collective care model and residents' everyday lives changed significantly when the virus entered residential care, creating a different lived space. The most common spatial strategies were isolation, social distance and managing movements in line with existing research. These were enacted in existing spatial conditions. The study findings challenge the value of community in the collective care model. For instance, the staff found it very awkward to isolate the residents in their flats, whereas the residents themselves did not view life during the pandemic as very different from normal, everyday life. The findings also challenge the relevance of the architectural models that are in use today. The article concludes that it is necessary to develop new architectural models, a caring architecture in which handling epidemics is less strenuous, and where residents' diverse wishes can better be met.

Objectives: Crime prevention strategies often rely on the small set of micro-places where crime is most concentrated, the so-called hotspots, yet it has remained unclear how close existing hotspot detection methods come to the maximum coverage theoretically possible. This study introduces GraphVenn, the first algorithm that identifies the globally optimal placement of N fixed-radius hotspots directly from the empirical crime distribution, without relying on heuristic or approximate approaches.

Methods: GraphVenn was evaluated on three years of crime data from Malm & ouml;, Boston, and New York City (in total 1.75 million crimes) and compared against kernel density estimation (KDE), greedy PAI maximization (PAI-Max), and GraphTrace. Both the globally optimal and the greedy (fast approximation) modes of GraphVenn were evaluated across different spatial resolutions, demonstrating scalability to large urban datasets.

Results: In optimal mode, GraphVenn identified the absolute maximum coverage of incidents achievable under fixed-radius constraints. The greedy variant reached within 0.1-1.9% of this optimum while reducing runtimes by up to two orders of magnitude. By contrast, existing methods consistently fell short, e.g., in New York City the optimal GraphVenn captured 51,522 crimes within its top-100 hotspots compared to 35,098 with KDE and 28,241 with GraphTrace, while PAI-Max was excluded due to its runtimes. In practical terms, the baselines therefore missed between 16,000 and 23,000 crime incidents that could have been covered.

Conclusions: Globally optimal detection of fixed-radius hotspots that maximize the distinct crime count is now computationally feasible at city scale. GraphVenn offers (i) a practical tool for researchers, law enforcement, and crime analysts to identify the most effective fixed-radius hotspot locations with confidence that no better configuration exists, and (ii) a benchmark for evaluating approximate methods against the true maximum crime count. Open-source code is provided to support replication and further research.

Light Fidelity (LiFi) offers a high-speed, interference-resistant alternative to conventional wireless communication, making it well-suited for sensitive environments such as healthcare, defense, and industrial systems. While LiFi's confinement to line-of-sight communication provides a natural layer of physical security, it remains susceptible to local eavesdropping and insider interception within its coverage area. These limitations underscore the need for additional data-level protection strategies that align with LiFi's operational constraints. This paper introduces a novel steganographic method tailored for data structured in matrix form, a common representation in many digital systems. To demonstrate the effectiveness of the proposed technique, images-naturally represented as two-dimensional matrices-are used as test cases. The approach avoids traditional payload embedding, which can be statistically detectable, and instead applies recursive segmentation, matrix reshaping, and hierarchical tree-based indexing to transform the structure of the data itself. This process produces encrypted outputs that appear statistically random and visually unstructured (i.e., noise-like), concealing both the data content and the presence of hidden communication. Quantitative evaluations using metrics such as entropy, correlation coefficients, contrast, homogeneity, and Bhattacharya distance confirm that while the encrypted data is statistically obfuscated, the original matrix can be losslessly reconstructed through inverse recursion. The method's design ensures lightweight processing is suitable for resource-constrained LiFi-enabled sensor nodes while significantly enhancing communication confidentiality. By restructuring data at the matrix level rather than embedding within it, this approach provides an effective and generalizable framework for secure transmission in physically exposed but bandwidth-rich LiFi networks. 

Healthy aging implies physical and psychological well-being, maintaining social relationships and engaging in meaningful activities. Thus, socially inclusive initiatives that address loneliness are needed. The aim of this ethnographic study was to describe interpersonal interactions in a community-based activity program targeting loneliness among older adults. Over 10 weeks, 10 participants were observed during interpersonal interactions on 20 occasions, focusing on when, where, and how the interactions occurred. Field and reflective notes were analyzed using an ethnographic approach. The findings show that support was promoted through communication, which created togetherness in the activities by encouraging each other and exchanging knowledge and information. Furthermore, connecting by embracing openness highlighted a willingness and courage to share life experiences and bring memories into conversations. These findings provide valuable insights for designing future activity programs that reduce loneliness and promote social connectedness among older adults. 

Modern software-intensive systems must address a wide range of nonfunctional requirements (NFRs) - such as security, compliance, and maintainability - that are critical for the long-term success of the system. With the rise of large-language-model-based agents, software engineering is entering an 'agentic' era where AI components are not only tools but collaborators in development processes. However, leveraging these agents introduces dual challenges: ensuring that AI components themselves meet quality standards (e.g., compliance, security, maintainability), and harnessing AI effectively to support systemlevel NFR assurance. Our perspective explicitly spans both SE4AI, where AI components such as agents are engineered and subjected to quality assurance and AI4SE, where AI agents support the engineering of software-intensive systems. While these are conceptually distinct, our model addresses both in a unified way. This position paper introduces a conceptual, domain-agnostic three-layer model - comprising Data, Agent, and Perspective layers - for systematically embedding AI agents into NFR assurance across the software lifecycle. The model explicitly captures two complementary viewpoints: Quality for AI (ensuring AI agents are trustworthy and maintainable) and AI for Quality (using agents to support system NFRs). Through illustrative examples in compliance, security, and maintainability, the paper demonstrates how this model can guide researchers and practitioners in designing agent-based approaches to software quality. We argue that this model not only clarifies the dual roles of AI in software engineering but also provides a foundation for responsible, scalable, and effective integration of AI into NFR assurance.

Unmanned aerial vehicles (UAVs) are foreseen to take on an important role in 6 G networks. UAVs can be used as aerial base stations (BSs) and aerial relays that offer flexibility in positioning, extended coverage, and reliable connectivity. A challenge in the deployment of UAVs is determining UAV placements and finding a suitable clustering of user equipment (UE) around the UAVs. In this paper, the K-means, K-means++, K-medians, and bisecting K-means algorithms are considered to obtain the UAV placements and the associated clustering of the UEs in non-orthogonal multiple access (NOMA) based cooperative UAV networks. Given a large set of UEs, the four clustering algorithms are executed for up to 1 0 0 different network topologies. Box plots are used to reveal the impact of clustering dispersion on UAV network performance. In addition, the performance supported by each of the four clustering algorithms in terms of outage probability and sum rate is assessed in the presence of Nakagamim fading. Simulation results are provided subject to the number of topologies, the BS transmit signal-to-noise ratio, the fading severity parameter m, and the number of clusters. The results illustrate the impact of the clustering dispersion associated with each clustering algorithm and system parameters on the outage probability and the sum rate of the considered NOMA-based cooperative UAV networks.

Context: Software systems in regulated domains must continually adapt to legal changes, yet practitioners often handle updates manually with limited support, making compliance work costly and error prone. Recent advances in LLMs prompt the question of how automation can reliably assist this process.

Objectives: We aim to (1) characterize the nature of regulatory changes and derive a systematic taxonomy, (2) understand through the lens of practitioners where automation is most useful, and (3) assess the feasibility of using LLMs for detecting and classifying regulatory changes.

Method: We conducted a mixed-methods study grounded in the German social security (DEÜV) in collaboration with practitioners from a FinTech company. First, we developed a taxonomy of regulatory changes through manual document analysis of four Regulatory Implementation Specifications (RIS), followed by a workshop and expert interviews. Second, we validated the taxonomy and elicited challenges through semi-structured practitioner interviews. Third, we built a gold-standard dataset of 93 annotated change instances and evaluated seven state-of-the-art LLMs within an automated detection and classification pipeline.

Results: The taxonomy defines five change scopes and four optional context dimensions. Practitioners found it intuitive and useful for filtering relevant changes, particularly Data and Field updates, but reported challenges such as tight deadlines, legal ambiguity, limited traceability, and overlapping categories. In automation, proprietary LLMs performed best, while performance dropped on narrative or weakly structured documents, highlighting sensitivity to document format.

Conclusion: The proposed taxonomy provides a practical lens for organizing regulatory change information, and LLMs can support the identification and classification of recurring, structurally explicit changes. Their limitations on context-dependent and infrequent categories suggest that automation should complement, rather than replace, expert assessment, motivating future work on human-in-the-loop compliance tooling across broader regulatory ecosystems. 

District heating (DH) networks are vital in the transition to sustainable energy systems. Maintaining their performance requires continuous monitoring to reduce heat losses, avoid user discomfort, and support efficient operation. However, automatic fault detection and diagnosis (FDD) in DH substations remains challenging due to limited labelled data, class imbalance, heterogeneous operating conditions, privacy constraints, and the lack of standardisation. This thesis aims to develop learning-based approaches for automatic FDD in DH substations that can operate under these real-world industry constraints. Therefore, this work focuses on scalable representations, transferability across domains, and robustness against scarce and imbalanced fault data, and includes eight studies that jointly develop and validate learning-based FDD methods for DH substations. Across these studies, the work advances unsupervised locality-based anomaly detection, hybrid and augmentation-enhanced fault diagnosis, transfer and cross-modal learning for label-scarce settings, privacy-preserving semi-supervised federated learning, and streaming representations for continuous fleet-scale monitoring. The results demonstrate that reliable, scalable FDD in DH systems can be achieved despite severe field constraints. By combining topology inference for local analysis, data representation, knowledge transfer and data augmentation, this thesis advances practical, deployable FDD intelligence to support more resilient, efficient, and data-smart DH networks aligned with the climate and digitalisation goals.

Internal hackathons are widely used to drive innovation and collaboration in software companies, yet not all succeed. Many fall short due to low employee engagement. Why do some employees dive in, while others hold back? This article explores what drives and hinders engagement based on insights from five companies practicing regular hackathons. Using Social-Cognitive Theory, we show that engagement depends on two key beliefs: one’s ability to contribute (self-efficacy) and the value of participation (outcome expectations). These drivers are not arbitrary; they are predictable, and thus manageable. By understanding the psychology behind engagement, organizers can create hackathons that help more employees to feel confident and inspired to participate. We offer practical guidance to make your next hackathon truly engaging. 

Background: As paediatric day surgery becomes increasingly common, postoperative care is transferred to the home setting where parents play a central role in assessing and managing their child's postoperative pain. No validated Swedish instrument currently exists to support parents in evaluating their child's pain at home. Aim: To assess the reliability and validity of the Swedish translation of the PPPM-S in children aged 2–12 years during the first two postoperative days.

Method: The instrument was earlier translated from English into Swedish in accordance with the WHO Guidelines for translation and adaptation of an instrument. A backward-forward translation was done with a bilingual expert panel, and cognitive interviews were done in the target population. This study was conducted at three Swedish hospitals between 2022 and 2025 involving 80 parents of children aged 2–12 years who underwent day surgery. Parents completed the PPPM-S on postoperative days 1 and 2. To evaluate the accuracy of the instrument, the results were compared with scores from an established pain rating scale, the Coloured Analogue Scale (CAS).

Findings: PPPM-S demonstrated good psychometric properties: good internal consistency (Cronbach's alpha 0.842–0.851) and satisfactory criterion validity demonstrated by strong correlations with CAS (Spearman's rho = 0.683–0.630, p < 0.01). ROC analysis identified 5/15 as an optimal cut-off, with acceptable sensitivity and specificity. Parents reported high levels of satisfaction and found it easy to use at home.

Conclusion: The PPPM-S is a valid and practical tool for assessing children's postoperative pain at home. It can help parents better understand and evaluate their child's pain, potentially improving postoperative care in the home setting. 

This paper introduces a generative framework for individual agent behavioral modelling based on artificial personalities (APs), which has been defined as distinct distributions over decision strategies in the given task. Using Markov Chain Monte Carlo (MCMC) sampling method, we generate action sequences that embody each AP's internal decision dynamics. The framework is evaluated through a numberguessing task, where five APs are compared based on two performance metrics: the length of the action sequence and a curiosity score, which quantifies deviation from the optimal solution. Results show that while agents demonstrated comparable task efficiency, their curiosity levels varied significantly, reflecting diverse exploratory tendencies. These findings support the feasibility of encoding strategic individuality in artificial agents, enabling richer behavioral modeling for applications in simulation, multi-agent coordination, and human-AI interaction. 

Federated learning (FL) is a distributed machine learning paradigm that enables multiple participants to collaboratively train a model without sharing raw data to preserve privacy. However, traditional FL frameworks remain vulnerable to integrity and accountability issues at both the global and local levels. Blockchain (BC), known for its decentralization, transparency, immutability, and cryptographic security, has been explored to enhance trust in FL. Yet, prior BC-integrated FL approaches often suffer from limitations such as resource-heavy consensus algorithms, on-chain storage of model weights, and unenforced incentive schemes in semi-decentralized settings. In this work, we propose a lightweight decentralized FL framework for supply chains. Our design features role-based governance, quorum-based model approval, stake-and-slash incentives, and off-chain model storage via the InterPlanetary File System (IPFS) to minimize gas costs. We validate the framework through simulation and perform BC performance evaluation, demonstrating its efficiency in terms of gas usage and latency across 100 FL rounds. 

Organizations often migrate monolithic architectures to microservices based on ad hoc data, expert opinions, or industry trends without assessing their specific context and needs. Such transitions tend to coincide with increased architectural complexity and technical debt (TD), making it crucial to understand how TD evolves over time in industrial settings to manage it effectively. This observational study explores the evolution of technical debt density (TDD) in a single software product consisting of both monolithic and microservice architectures at a Swedish fintech company, without aiming to establish causality between architectural styles and TDD trends. We further investigate TDD trends across various microservice size categories, team types, and the relationship between size and TDD. We analyzed SonarQube TD data collected from one monolith and 78 microservices from August 2022 to December 2024, and conducted semi-structured interviews with practitioners (a development manager, a product owner, and a lead developer) to validate and contextualize the quantitative findings. Our results show that, in this case, the monolithic system exhibits a decreasing TDD trend over time despite continued growth in size, while a gradual increase in TDD is observed across microservices. Furthermore, TDD trends appear inconsistent among small microservices, more consistently growing in medium-sized microservices, and comparatively stable in larger services. Differences in TDD trends are observed across services owned by platform teams and product teams. Overall, the findings from this specific case suggest that TDD evolves differently in monolith and microservices, highlighting the importance of continuous monitoring and context-aware interpretation of TDD trends in practice.