Publikationer från Stockholms universitet
Not a member yet
    31237 research outputs found

    En jämförande studie av AI-kodassistenter för automatiserad debugging

    No full text
    LLM:er har alltmer blivit närvarande inom mjukvaruutveckling, vilket också har ökat deras roll inom automatiserad debugging. Debugging kan vara en komplicerad uppgift som kräver en djup förståelse för mjukvaran. Med LLM:ers lovande kapacitet syftar denna studie till att undersöka och jämföra tre ledande AI-modeller: OpenAI:s GPT-4.1, Anthropic:s Claude 3.7 Sonnet och DeepSeek:s DeepSeek-R1, för att fördjupa förståelsen av den optimala LLM:en att använda inom automatiserad debugging. Denna studie syftar till att undersöka dessa modellers kapacitet att lokalisera och fixa logiska buggar och runtime-buggar genom olika Java-klasser. För denna undersökning formulerades tre frågor: Vilka styrkor och svagheter hade respektive modell när det gällde debugging av runtime-buggar och logiska buggar? Hur träffsäker var respektive modell i att lokalisera buggen? Hur effektiv var respektive modell i att generera fungerande lösningar? För att besvara frågorna användes en experimentell metod där ett API-baserat mjukvaruverktyg skapades som kunde skicka en buggig fil till en specifik AI-modell och bland annat ta emot feedback och automatisk korrigering. Alla AI-modeller fick exakt samma buggiga filer som genererades genom Pitest, ett verktyg som planterar mutatorer i fungerande kod. Prestationen hos de olika modellerna mättes genom JUnit-tester som sedan analyserades deskriptivt, där olika medelvärden togs fram för AI-modellernas prestation, och inferentiellt, där signifikanta skillnader identifierades. Resultaten visar att alla modeller har en hög framgångsgrad inom debugging, där OpenAI:s GPT-4.1 presterade bäst på båda buggtyperna och även förblev konsekvent, medan Anthropic:s Claude 3.7 Sonnet presterade näst bäst men med betydande inkonsekvens, då den presterade sämre på runtime-buggar. DeepSeek:s DeepSeek-R1-modell presterade sämst på båda buggtyperna och visade även inkonsekvent prestation, också med sämre prestation på runtime-buggar. När det gällde att lokalisera buggar presterade alla modeller väl utan signifikanta skillnader, men både GPT-4.1 och Claude 3.7 Sonnet var signifikant bättre på att lokalisera runtime-buggar. Vidare visar studien att det finns en signifikant skillnad i prestation för både buggtyper och Java-klasskontext, vilket betonar vikten av att välja den optimala modellen baserat på debugging-scenariot. Dessa fynd kan hjälpa utvecklare att välja en optimal AI-modell för automatiserad debugging. Denna studie var också begränsad på sätt som framtida studier kan ta hänsyn till. Dessa inkluderar att testa flera buggtyper över mer testdata, testa fler AI-modeller, och även utforska andra mjukvaruingenjörsuppgifter utöver bugglokalisering och fixering.LLMs have increasingly become more present in software development, which has also increased their role in automated debugging. Debugging can be a complicated task that requires a deep understanding of the software. With LLMs' promising capabilities, this study aims to research and compare three leading AI models: OpenAI’s GPT-4.1, Anthropic’s Claude 3.7 Sonnet, and DeepSeek’s DeepSeek-R1, to deepen the understanding of the optimal LLM to use in automated debugging. This study aims to investigate these models' capabilities in localising and fixing logical and runtime bugs through different Java classes. For this investigation, three questions were formulated: What strengths and weaknesses did each model have when it came to debugging runtime and logical bugs? How accurate was each model in localising the bug? How effective was each model in generating working solutions? To answer the questions, an experimental method was used where an API-based software tool was created that could send a bugged file to a specific AI model and, among other things, receive feedback and automatic correction. All AI models received exactly the same bugged files that were generated through Pitest, a tool that plants mutators in working code. The performance of the different models was measured through JUnit tests that were then analyzed descriptively, where different averages were produced for the performance of the AI models, and inferentially, where significant differences were identified. The results show that all models have a high success rate in debugging, with OpenAI's GPT-4.1 performing best on both bug types and also remaining consistent, while Anthropic's Claude 3.7 Sonnet performed second best but with significant inconsistency, as it performed worse on runtime bugs. DeepSeek’s DeepSeek-R1 model performed the least well on both bug types and also showed inconsistent performance, also performing worse on runtime bugs. When it came to localising bugs, all models performed well with no significant differences, but both GPT-4.1 and Claude 3.7 Sonnet were significantly better at localising runtime bugs. Furthermore, the study shows that there is a significant difference in performance for both bug types and the Java class context, which emphasizes the importance of choosing the optimal model based on the debugging scenario. These findings can aid developers in choosing an optimal AI model for automated debugging. This study was also restricted in ways that future studies can take into account. These include testing multiple bug types across more test data, testing more AI models, and also exploring other software engineering tasks beyond bug localisation and fixation

    Migranters upplevelse av digital legitimering i Sverige

    No full text
    Introduktion Denna uppsats undersöker migranters upplevelse av att skapa och aktivera BankID, ett centralt verktyg i samhällets digitala infrastruktur. Studien belyser hur digitala identifieringssystem, såsom BankID, kan möjliggöra och begränsa tillgången till viktiga samhällsfunktioner. Forskningsfråga Den primära forskningsfrågan är: "Hur upplever migranter processen att skapa och aktivera BankID?" Metod En kvalitativ fallstudie med nio semi-strukturerade intervjuer genomfördes med migranter för att samla in data om deras upplevelser av processen att skapa och aktivera BankID. Det insamlade materialet analyserades med hjälp av tematisk analys i enlighet med Braun och Clarke (2006) med en induktiv ansats. Resultat Analysen resulterade i tre centrala teman: osäkerhet och brist på vägledning i mötet med svenska myndigheter, bankens avgörande roll i processen samt BankID:s symboliska betydelse för tillhörighet och kontroll över vardagen. Migranter upplevde processen som komplex, ofta beroende av stöd från andra, och att tillgång till BankID var avgörande för att kunna delta i samhället. Diskussion Dessa resultat tyder på att tillgången till BankID inte bara är en teknisk fråga, utan också djupt kopplad till frågor om inkludering och självständighet. Brist på tydlig information, språkbarriärer och godtyckliga bankrutiner bidrar till att förstärka digitalt utanförskap.This thesis explores migrants’ experiences of creating and activating BankID - a central tool in Sweden’s digital infrastructure. The study highlights how digital identification systems can both enable and limit access to essential public services, particularly for groups such as migrants. To investigate this, a qualitative case study was conducted, based on nine semi-structured interviews with migrants. The collected material was analyzed using thematic analysis in accordance with Braun and Clarke (2006) with an inductive approach that allowed themes to emerge from the participants’ own narratives. The analysis resulted in three central themes: uncertainty and lack of guidance in interactions with Swedish authorities, the decisive role of banks as gatekeepers in the process and the symbolic meaning of BankID for social belonging and personal agency. The results show that migrants often experience the process as complex and fragmented and that access to BankID is crucial for participating in society on equal terms. The study demonstrates that this is not merely a technical or administrative matter, but also reflects broader issues of inclusion, autonomy and digital participation. A lack of clear information, language barriers and arbitrary assessments by banks contribute to reinforcing digital exclusion for already vulnerable groups

    The interplay of standard essential patents (SEPs) and EU competition law in the 5G era

    No full text
    This thesis examines the relationship between standard essential patents (SEPs) and competition law within the context of the 5G era in the European Union (EU). SEPs play an important role in the development and implementation of 5G by ensuring interoperability and standardisation, which helps facilitate innovation and seamless communication between devices and networks. However, licensing and enforcing SEPs introduce competition law challenges, especially in balancing the incentives for innovation with fair competition practices. The thesis addresses two primary research questions: 1) How regulatory frameworks and legal precedents shape the interplay between SEPs and competition law in the 5G industry within the EU, and what specific gaps or inconsistencies hinder effective market regulation. It also examines 2) how SEPs influence market dynamics, identifying potential barriers to innovation and competition arising from their implementation, and how key competition law issues impact the balance between incentivising innovation and preventing market abuse. The methodology combines a literature review, legal dogmatic analysis, and case studies to offer an understanding of the theoretical frameworks, key studies, and legal norms related to SEPs and competition law. The findings indicate that SEPs are integral to the advancement and implementation of 5G technology, fostering innovation and ensuring interoperability. Nevertheless, the licensing and enforcement of SEPs pose substantial challenges under competition law, particularly within the stringent regulatory framework of the EU. This thesis examines the economic impact of SEP licensing practices, emphasising the need to balance the encouragement of innovation with the prevention of market abuse. Policy recommendations suggest strengthening FRAND (fair, reasonable, and non-discriminatory) commitments, improving transparency in SEP licensing, and encouraging innovation and R&D investment. The study finds that a balanced approach to SEP enforcement is important for maintaining a competitive and innovative telecommunications industry in the EU

    Värna och skydda de viktiga funktionerna i djupa mjukbottnar

    No full text
    Djupa mjukbottnar är viktiga för hela det marina ekosystemets funktion och för långsiktig kolinlagring. Dessa funktioner riskerar att försämras när bottnarna trålas vid kommersiellt fiske och vid gruvdrift på havsbotten. Ändå förbises dessa livsmiljöer ofta när man fastställer marina skyddsområden och begränsar fiske och exploatering. EU:s medlemsländer måste skydda djupa mjukbottnar i Östersjön och Nordsjön bättre

    Urban segregation or residential integration in France? : A comparative study of Algerian, Moroccan and Portuguese Populations across Paris and Marseille (2006-2020).

    No full text
    Research on ethnic and urban segregation has been particularly prominent in the United States and parts of Europe, yet they remain relatively limited in the French context. Given France immigration history, and current political debates surrounding integration, this study investigates the residential integration patterns of immigrants in France (2006-2020). It focuses on the three largest immigrant groups, Algerians, Moroccans, and Portuguese, across the country’s two largest urban areas: Paris and Marseille. The study is structured in three parts. First, it maps the spatial distribution of these groups in relation to the native French population and measures segregation levels using dissimilarity indices. Second, it examines the role of socio-economic factors (income and education) in shaping residential patterns, using statistical analyses to assess their influence. Third, the study complements quantitative findings with a qualitative exploration of how immigrant residential concentrations manifest in the physical and social features of urban spaces. The findings reveal different patterns of integration and segregation between European (Portuguese) and non-European (Algerian and Moroccan) immigrant groups. They also challenge the assumption that immigrant residential integration in France follows a linear trajectory as predicted by the spatial assimilation theory, suggesting the need to consider alternative explanatory frameworks such as the place stratification and the ethnic preference theories

    Women's homelessness in the Swedish welfare state : Critical discourse analysis of the state's understanding and positioning of women's homelessness in SOUs between 1922-2024 

    No full text
    Denna studie undersöker hur kvinnors hemlöshet har diskursivt konstruerats i Sverige. Genom att undersöka Statens offentliga utredningar, SOU, mellan 1922 och 2024, är syftet att bidra till en förståelse för hur denna diskurs har förändrats över tid särskilt vad gäller den hemlösa kvinnans kropp och sexualitet i välfärdsstaten Sverige. Genom en kritisk diskursanalys analyseras hur kvinnors hemlöshet definieras, vilken plats deras kroppar och erfarenheter ges i det offentliga rummet, samt vilka politiska och sociala åtgärder som föreslås. Studiens analys vilar på ett feministiskt teoretiskt ramverk där frågor om kunskapsproduktion, kroppslighet och sexualitet, samt kvinnors plats i välfärdssamhället står i centrum. Resultaten visar en tydlig diskursiv förskjutning: från att hemlösa kvinnor framställdes som moraliskt förfallna till senare beskrivas som utsatta offer för strukturell och individuell marginalisering. Samtidigt sker denna omsvängning inom snäva, normativa ramar, där endast vissa typer av kvinnlig utsatthet ges erkännande. Studien visar att även om språkbruket har förändrats, kvarstår många av de underliggande exkluderande strukturerna. Hemlöshet bör därför inte enbart förstås som ett bostadspolitiskt eller socialt problem, utan också som ett geografiskt och rumsligt förhållande, där kvinnors rätt till plats är villkorad av kön, respektabilitet och samhällelig kontroll. Studien avslutas med en reflektion kring föreställningen om kvinnan som det "ideala offret" och ett kritiskt upprop till planeringspraktiker: att inte enbart se till tak över huvudet, utan att omförhandla vilka kroppar, röster och erfarenheter som ges plats i stadens framtid och vilka värden som tillskrivs dem.

    Between Silence and Strolling : The case of the Woodland Cemetery

    No full text
    This thesis explores how the Woodland Cemetery functions as a multifunctional urban green space, serving as both a site for mourning and a space for recreation. Urban densification has drawn attention to cemeteries as overlooked urban green infrastructure. Although previous research has recognized the recreational potential of cemeteries, the utilization and experience of these spaces in the Swedish context remain underexplored. This research combines policy document analysis, observations, and semi-structured interviews with visitors and cemetery stakeholders. The aim is to investigate the potential of cemeteries, specifically the Woodland Cemetery, as multifunctional urban green spaces that serve both as places of mourning and recreational areas. The analysis draws on theories of deathscapes, sense of place, and multifunctional urban space. Results indicate that visitors often describe the site as peaceful and beautiful, using it for walking, reflection, or daily routines while adapting their behavior to the cemetery’s symbolic atmosphere. Institutional actors acknowledge this everyday use but emphasize its primary role as a site for mourning. The study demonstrates how cemeteries can provide ecological, social, and symbolic value in urban life. It contributes to urban planning by highlighting how meaning is negotiated in emotionally charged, multifunctional spaces within the city

    Reframing Parking Planning in a Million Housing Programme Neighbourhood : Navigating Justice amid Socio-Technical Constraints and Opportunities in Visättra’s Car-Centric Legacy

    No full text
    In car-centric modernist neighbourhoods, such as Visättra and other Million Housing Programme areas, ample surface parking was established at the edges of residential areas. Parking is an inherent component of car use; however, it offers no benefits to non-car owners and instead hampers density and safety while privileging private car use over sustainable transport. Therefore, this study explores how parking planning has evolved over time, investigating how the boundaries of the socio-technical system of automobility constrain the pursuit of a more just urban environment and mobility. Drawing on socio-technical systems theory and complementing it with the concept of prioritarian distributive justice, the study thematically analyses archival and present-day planning documents, together with interviews with practitioners. It concludes that incremental changes in parking planning are achievable, whereas radical ones are not. The socio-technical limits of automobility constrain the potential of alternative sustainable transport modes. Lacking viable alternatives to compensate for car-induced accessibility, some groups risk enduring excessive burdens, effectively making justice a limiting factor in changing parking planning. By combining socio-technical systems theory with a prioritarian justice lens, this thesis offers a novel framework for analysing the evolution of parking planning and for pragmatically contextualising justice within automobility

    Delight och tillit hos seriösa tjänster

    No full text
    Denna uppsats jämför användares tillit till gränssnitt för seriösa ändamål med och utan designval för överraskande delight, som syftar till att positivt överraska användaren genom animationer, illustrationer, personliga interaktioner och andra mikrointeraktioner. Digitala system för seriösa ändamål, som sjukvård och ekonomi, är mycket vanliga och kan ses som en förutsättning i det moderna samhället. Det är inte fastställt huruvida tjänster med seriösa ändamål kan och bör främja överraskande delight utan att påverka tillit, som är en avgörande faktor för hur villig man är att använda ett system. Frågeställning i denna uppsats lyder: Påverkar gränssnittsdesign för överraskande delight användares tillit till seriösa tjänster? A/B tester utfördes där deltagarna fick testa en prototyp av CSN:s digitala tjänst. Experimentversionen inkluderade designval för att väcka överraskande delight medan kontrollversionen inte gjorde det. Efter testet svarade deltagarna på en enkät för att mäta deras upplevda tillit. 45 personer deltog och datan analyserades med Mann-Whitney U-test för att undersöka skillnader mellan grupperna. Resultatet visar att experimentgruppens svar generellt har något högre medelvärde på samtliga frågor jämfört med kontrollgruppens. Endast en av frågorna resulterade i ett signifikant resultat mellan grupperna. Effektstorleksanalyser indikerar att designvalen för överraskande delight har en positiv effekt på användarnas tillit. Studien visar att tidigare tillit till CSN kan ha påverkat de generellt höga tillitsvärdena, men trots små och subtila designförändringar i experimentversionen uppmättes en hög effektstorlek. Endast deltagarnas villighet att ange betaluppgifter visade en statistiskt signifikant skillnad, där experimentversionen ökade denna tillit. Resultaten tyder på att överraskande delight kan främja tillit i seriösa tjänster, men vidare forskning behövs för att undersöka effekter vid längre och upprepad användning.In modern society, interacting with digital services is not only common but often a necessity in daily life. This study seeks to examine whether surprise-based delight, i.e. design choices intended to positively surprise the user through animations, illustrations, personal interactions and other microinteractions, can be appropriately integrated in systems with serious goals, such as banking or healthcare services. The prevailing view seems to be that surprise-based delight should not be integrated within systems with serious goals, but this has not been clearly confirmed or refuted. A randomized controlled field experiment was conducted. Participants (n = 45) tested one of two interactive prototypes based on the student loan agency Centrala studiestödnämnden’s website. One of the prototypes included design elements meant to evoke surprise-based delight, including animations and illustrations, while the other did not. After completing tasks in the prototype, participants answered a survey measuring their perceived trust in the system. The results showed that the version with surprise-based delight elements was perceived as more trustworthy. One of the five survey questions demonstrated a statistically significant difference and all questions indicated a high effect size in favor of the delightful version. The findings suggest that surprise-based delight may have a positive impact on trust in services with serious goals. However, further research with larger participant samples and long-term testing is recommended to validate and expand on these results

    Distributed Machine Learning for Cyber Defense in IoT Environments - A Systematic Literature Review

    No full text
    The Internet of Things (IoT) has become a critical technology across diverse industries, but its rapid expansion has brought significant security challenges. Centralized security mechanisms often struggle to address the evolving threat landscape. Distributed machine learning (DML) has emerged as an effective approach to strengthen IoT cyber defense, but a comprehensive review of its application and effectiveness is still lacking. Therefore, this thesis investigates how DML can be effectively designed and implemented to enhance IoT security and addresses how such designs can improve privacy, communication efficiency, and robustness against adversarial attacks. A systematic literature review (SLR) was conducted to explore existing DML applications for IoT security. Thematic analysis of 34 selected studies identified six key themes, including IoT and DML attacks, security requirements, technologies and approaches, privacy protection, communication efficiency, and robustness. Based on these findings, a four-layer conceptual framework was developed, comprising Threat Landscape, Core and Functional Requirements, Technologies and Approaches, and IoT Security. Supported by nine key evaluation metrics for assessing DML system performance, this framework provides structured guidance for designing DML-based security solutions in IoT environments. In addition, three open issues were identified in the literature: the trustworthiness of DML predictions, the heterogeneity management in DML, and the trade-offs between privacy, scalability, and performance. By highlighting the critical challenges and future research directions, this thesis contributes to a deeper understanding of the current state and potential of DML for IoT cyber defense

    0

    full texts

    31,237

    metadata records
    Updated in last 30 days.
    Publikationer från Stockholms universitet
    Access Repository Dashboard
    Do you manage Open Research Online? Become a CORE Member to access insider analytics, issue reports and manage access to outputs from your repository in the CORE Repository Dashboard! 👇