Blekinge Institute of Technology
Not a member yet
13576 research outputs found
Sort by
The End of Pretraining for Large Language Models : The Future of Agentic and AI Reasoning Beyond Peak Data
No full textArtificial intelligence (AI) has entered a transformative era, as traditional pretraining paradigms face the constraints of “peak data.” This article explores the implications of a “post-pretraining” era, where AI systems transcend static data dependency and evolve into agentic AI entities.
Improving Requirement Traceability in Agile Projects within SMEs
No full textBackground: Requirement traceability is an important but challenging aspect of Agile software de-velopment, especially in small and medium-sized enterprises (SMEs). These teams often rely on informal methods, making it hard to keep track of changing require-ments and their links to development artifacts. Objectives: This thesis aims to explore the specific traceability challenges faced by Agile SMEs,develop a lightweight, practical checklist addressing those challenges, and to evaluatethe checklist’s effectiveness in a real-world web application project in a single casestudy. Methods: Exploratory study with survey and case study. A survey gathered requirement trace-ability challenges experienced by Agile practitioners working in SMEs. The checklist maps the identified traceability challenges to the Gotel et al.’s eight challenges withthe action steps mapped to specific role and experience level. A single case study isconducted to test the checklist’s effectiveness with a real Agile team working on aweb application project. Results: From the survey it is found that frequent requirement changes, tool limitations, andlack of communication are the traceability challenges mostly faced by the people working in SMEs. The developed checklist helps the team to improve documenta-tion process, maintain traceability links, and enhance team communication, while reducing the time spent on traceability process which is found during the case study. Conclusions: The developed checklist found to be useful for Agile teams working in SMEs, whichis more structured and easy to use. It helps teams to handle the changes quicklywithout extra work, making the traceability process easier to follow, as shown by itsperceived usefulness in a single case study
Immersive Analytics Meets Artificial Intelligence : A Systematic Review
No full textIntegrating artificial intelligence (AI) with immersive analytics (IA) represents a promising means of leveraging advanced computational techniques to enhance data visualization and analysis. This study examines the state-of-the-art of AI-IA integration by addressing three key research issues: the significant application domains, the AI techniques used and their combinations, and current challenges and future directions. Results of reviewing 43 relevant studies reveal that AI-IA integration is still in its early stages, as existing research has mainly focused on a limited range of data types and application scenarios. By analyzing the application domains, this systematic literature review supports previous findings of important applications in the fields of education, manufacturing, and healthcare. At the same time, it identifies emerging applications that have progressed from XR and AI domains to AI-IA integration, such as sports events, assistive systems, urban planning, and disaster management. We contribute to extending established visual analytics (VA) pipelines into XR environments with integrated AI techniques. AI techniques are identified as contributing in five ways to this IA pipeline. Our contribution also includes identifying four key challenges and seven opportunities for future exploration. The review concludes that combining AI and IA holds the potential to create innovative applications using advanced AI and immersive visualization techniques. We present an overview of these applications and address key issues for future development.
Assessment Practices to Support Sustainable Product Development : Analyzing Approaches in Literature and Practice Through a Strategic Sustainable Development Lens
No full textThe aim of this thesis was to understand and explore how a strategic sustainable development perspective can be integrated in product sustainability performance assessment practices. Based on an interactive qualitative research approach, three studies were conducted. The first study reviewed early-phase assessment approaches available in literature and found that although many approaches exist, a systemic and strategic perspective was often not completely addressed. Six criteria were therefore developed to support integration of such a perspective in assessment approaches. The second study focused on two EU policy instruments and while both instruments presented opportunities to incentivize and support product sustainability performance assessments, they provide limited support for considering the full socio-ecological systems perspective and long-term strategic decision-making. The third study used a multi-case study to deepen the understanding of how product developing companies work with and use product sustainability performance assessments. The findings showed that product sustainability performance was assessed, but companies face challenges and their practices were not fully aligned with a strategic sustainable development perspective. Ten propositions were formulated for what must be considered to integrate such a perspective in company practices, so that assessments can function as strategic decision-support for developing products that can contribute to society’s transition towards sustainability in ways that benefit the own organization. Together, the three studies highlight the need to develop both individual assessment approaches and the surrounding practices in which they are embedded to support effective integration of a strategic sustainable development perspective into assessment practices in the product development process
An Empirical Evaluation of LLM:s for Software Architecture Recovery
No full textBackground. Software architecture documentation is often outdated or missing in long-lived software systems, making maintenance difficult. Software Architecture Recovery (SAR) aims to reconstruct architectural representations from source code, but existing techniques are often difficult to apply in practice. Recent advances in Large Language Models (LLMs) offer new possibilities for architecture recovery. Objectives. This thesis investigates how accurately terminal-integrated LLM-based agents can recover high-level architectural representations from source code, and how self-reflective prompting affects the accuracy and robustness of such recovery. The focus is on recovering container diagrams according to the C4 model, which is a hierarchical architectural modeling framework, from large-scale software systems. Method. A controlled experiment was conducted using three contemporary terminal-integrated LLM-based agents on two large open-source systems. Agent outputs were evaluated against manually constructed ground truth C4 container diagrams using an established error taxonomy. Statistical analysis was applied to compare agent performance before and after self-reflective prompting. In addition, a qualitative industrial replication was performed at Ericsson. Results. All evaluated agents were able to produce architecturally meaningful container-level representations, but with moderate and variable accuracy. No statistically significant differences were found between agents after self-reflection. The dominant source of errors was misalignment in abstraction level rather than semantic misunderstanding. Self-reflective prompting significantly reduced hallucination errors and improved output robustness for agents prone to over-generation, while having limited effect on more conservative agents. Conclusions. Terminal-integrated LLM agents constitute a viable approach for exploratory SAR, particularly when combined with self-reflective prompting. The main challenge is not architectural reasoning itself, but consistent control of abstraction level. These tools are best suited as interactive aids for architectural understanding rather than fully automatic reconstruction mechanisms.Bakgrund. Dokumentation kring arkitekturen för programvara är ofta föråldrad eller helt saknad i långlivade system, vilket försvårar underhåll och vidareutveckling. Software Architecture Recovery (SAR) syftar till att återskapa arkitektoniska representationer från källkod, men befintliga metoder är ofta svåra att tillämpa i praktiken. Nya framsteg inom stora språkmodeller (LLM:er) skapar nya möjligheter för återskapande av arkitekturdokumentation. Syfte. Syftet med denna studie är att undersöka hur väl terminalintegrerade LLM-baserade agenter kan återskapa abstrakta arkitektoniska representationer från källkod, samt hur självreflekterande promptning påverkar noggrannhet och robusthet i denna process. Fokus ligger på återskapande av C4-containermodeller för stora mjukvarusystem. Metod. En kontrollerad experimentstudie genomfördes med tre moderna terminalintegrerade LLM-baserade agenter på två stora open-source-system. Agenternas resultat utvärderades mot manuellt framtagna referensarkitekturer enligt C4-modellen, med hjälp av en etablerad taxonomi för felklassificering. Statistiska analyser användes för att jämföra resultat före och efter självreflekterande prompting. Därutöver genomfördes en kvalitativ industriell replikation hos Ericsson. Resultat. Samtliga agenter kunde generera arkitektonisk meningsfulla containerrepresentationer, men med varierande och måttlig noggrannhet. Inga statistiskt signifikanta skillnader mellan agenterna observerades efter självreflektion. Den dominerande felkällan var felaktig abstraktionsnivå snarare än semantiska missförstånd. Självreflekterande prompting minskade hallucinationsfel avsevärt och ökade robustheten för agenter med tendens till övergenerering. Slutsatser. Terminalintegrerade LLM-baserade agenter är ett lovande verktyg för explorativ SAR, särskilt i kombination med självreflekterande prompting. Den främsta utmaningen är inte arkitektonisk förståelse, utan konsekvent kontroll av abstraktionsnivå. Verktygen lämpar sig bäst som interaktiva stöd för arkitekturell förståelse snarare än helt automatiserade rekonstruktionslösningar
Inflated Modified Kumaraswamy Regression Model for Invasive Plants Detection in NDVI Imagery
No full textThis study proposes the inflated modified Kumaraswamy (iMK) distribution, a flexible probability model defined on the unit interval [0,1]. It captures asymmetric behaviors while accommodating inflation at zero, one, or both boundaries, as commonly observed in normalized difference vegetation index (NDVI) data. Based on the iMK distribution, we develop a new regression model (iMKreg) suitable for double-bounded responses. From this model, we derive a detection tool for invasive plant species, particularly applicable to NDVI imagery. Model performance was evaluated using synthetic NDVI data, with further assessment of predictive accuracy and detection efficacy conducted on real-world measured NDVI image. The application to detecting black-grass (Alopecurus myosuroides) in wheat crops in southern Sweden shows that the iMKreg model outperforms both standard Gaussian-based linear regression and existing inflated Kumaraswamy regression models.
Ore extensions of abelian groups with operators
No full textGiven a set A and an abelian group B with operators in A, in the sense of Krull and Noether, we introduce the Ore group extension B[x;δ_B,σ_B] as the additive group B[x], with A[x] as a set of operators. Here, the action of A[x] on B[x] is defined by mimicking the multiplication used in the classical case where A and B are the same ring. We derive generalizations of Vandermonde's and Leibniz's identities for this construction, and they are then used to establish associativity criteria. Additionally, we prove a version of Hilbert's basis theorem for this structure, under the assumption that the action of A on B is what we call weakly s-unital. Finally, we apply these results to the case where B is a left module over a ring A, and specifically to the case where A and B coincide with a non-associative ring which is left distributive but not necessarily right distributive
Smart cities : the future with Artificial Intelligence
No full textArtificial Intelligence (AI) has become the latest technology disrupter that has the potential to dramatically change the nature of economy and sustainability in the decades to come. It is still in the nascent stage where wide spread adoption of AI methods including Machine Learning (ML), federated learning, unsupervised automation is still under the scrutiny for public sector adoption. As a foresight exercise, this chapter discusses the role of AI in cities. It provides an outlook on the use of AI in cities for city intelligence and their impact on Sustainable Development Goals (SDGs) as determined by the United Nations (UN). Use-cases, frameworks, and best practices for AI use in cities proposed by nations and international bodies are presented along with technical components that can enable AI adoption. Additionally, a detailed description of AI based techniques is provided to give a technical understanding of AI algorithms from academic and industry perspectives. The focus on data and federated components for discovery and interoperability enabling AI is presented to enable readers to build a vision for future connected urban AI ecosystems. Finally, concerns with the adoption of AI in the public sector are presented along with the conclusions of the chapter.
A dynamical regression model for double-bounded time series based on the reflected unit Burr XII distribution
No full textThis paper introduces a new time series model based on the reflected unit Burr XII (RUBXII) distribution that is an alternative to the Kumaraswamy autoregressive moving average and Beta autoregressive moving average models for time series analysis taking values in the standard unit interval. The proposed model describes the conditional median of RUBXII-distributed discrete-time series by a dynamic structure that includes autoregressive and moving average (ARMA) terms, a set of regressors, and a link function. We perform the model's parameter estimation using the conditional maximum likelihood method. Closed-form expressions for the score vector and observed information matrix are presented. We propose and discuss techniques of diagnostic and forecasting for the new model. A Monte Carlo simulation study is carried out to evaluate the finite sample performance of the conditional maximum likelihood estimator. Finally, the proportion of stored hydroelectric energy in Northern Brazil is analyzed through the proposed model. The results evidence that the introduced RUBXII-ARMA model is suitable for describing the dynamics of the data and provides more accurate forecasts for the proportion of stored energy in Northern Brazil than those from competitors' models
Source Data Selection for Brain–Computer Interfaces Based on Simple Features
No full textCarefully selecting the source data is crucial to achieve high performance of transfer learning methods for brain–computer interfaces (BCIs). Especially so in settings where a large amount of source data is available, and finding the optimal source is not computationally feasible. This paper presents a novel method for source selection, the so-called Transfer Performance Predictor (TPP) method. The TPP method is based on computationally simple features, a choice made to enable real-time implementation and reduce calibration time. The presented method outperforms other comparable source selection methods in BCI settings where a large amount of source data is available. By using the TPP method, source selection can be performed quickly with good results for transfer learning performance, which means that the BCI calibration time can be reduced and a new target user can more quickly start using the BCI.