Hoppa till huvudinnehåll

31

May

CS MSc Thesis Presentation Day May 31 2024

Tid: 2024-05-31 09:00 till 16:00 Föreläsning

Seven MSc theses to be presented on Friday May 31, 2024

Friday May 31 is a day for coordinated master thesis presentations in Computer Science at Lund University, Faculty of Engineering. Seven MSc theses will be presented.

You will find information about how to follow along under each presentation. There will be presentations in three different rooms: E:2116, E:2405 (Glasburen) and E:4130 (Lucas). See room for each presentation. A preliminary schedule follows.

Please note that there will also be thesis presentations on Thursday May 30, schedule at: https://cs.lth.se/kalendarium/?evenemang=cs-msc-thesis-presentation-day-may-30-2024

Note to potential opponents: Register as an opponent to the presentation of your choice by sending an email to the examiner for that presentation (firstname.lastname@cs.lth.se). Do not forget to specify the presentation you register for! Note that the number of opponents may be limited (often to two), so you might be forced to choose another presentation if you register too late. Registrations are individual, just as the oppositions are! More instructions are found on this page.


09:15-10:00 in E:2116 N.B. No more opponents for this presentation

Presenters: Nova Svensson, Anja Pettersson
Title: Case study on Open Source Forks: Impact on Knowledge Distribution and Code Quality
Examiner: Alma Orucevic-Alagic
Supervisors: Markus Borg (LTH), Joseph Fahey (CodeScene AB)

High-quality source code and effective knowledge distribution are important factors for software projects to be successful. Sometimes Open Source Software undergoes forking, resulting in the creation of independent projects from copied source code. This thesis conducts a case study analyzing how forking affects Open Source Software projects, focusing on knowledge distribution and code quality using CodeScene. The results showed that three out of four projects experienced prolonged knowledge loss from forking, with no apparent correlation between knowledge loss, code quality changes, bug frequency, and forking. This implies that Open Source communities should be aware of the consequences of forking, that it can lead to extended periods of knowledge loss and challenges in regaining previous knowledge levels. Insights important for communities' decision-making and resource allocation regarding knowledge management and retention. Future studies could investigate how social aspects and governance, leadership and community dynamics affect projects undergoing forking.

Link to popular science summary: https://fileadmin.cs.lth.se/cs/Education/Examensarbete/Popsci/240531_09PetterssonSvensson.pdf


10:15-11:00 in E:2405 (Glasburen)

Presenters: Annie Börjesson, David Jobrant
Title: Exploring Behavior-Driven Development at IKEA using design research
Examiner: Per Runeson
Supervisors: Emelie Engström (LTH), Andreas Trattner (IKEA)

In software development testing plays a crucial role in ensuring fulfilled requirements, yet it is frequently deprioritized. This report investigates how teams within a software department at IKEA work with testing and requirements and how behavior-driven development (BDD) can be applied. We performed a study using design research consisting of three main components: problem conceptualization, solution design and evaluation. The problem conceptualization phase includes interviews, meetings and observations, to gain a thorough understanding of IKEA’s current situation. For the solution design, we undertook a literature review to explore existing instances and aspects of BDD. Subsequently, we developed a process based on the gathered insights. It was evaluated during a workshop. Interview results reveal that teams experience issues with testing and requirements. Respondents were positive towards BDD. The evaluation indicates that our process is beneficial in terms of testing and requirements. However, additional evaluation is suggested to further anchor our approach.

Link to popular science summary: https://fileadmin.cs.lth.se/cs/Education/Examensarbete/Popsci/240531_10BorjessonJobrant.pdf


11:15-12:00 in E:4130 (Lucas) N.B. No more opponents for this presentation

Presenters: Isak Jakobsson, Jonathan Runeke
Title: Point Cloud Based Crowd Counting
Examiner: Elin Anna Topp
Supervisor: Maj Stenmark (LTH)

The goal of this master's thesis was to solve the task of crowd counting in LiDAR point clouds. LiDAR (Light Detection and Ranging) leverages laser technology to scan an environment and produce a point cloud consisting of coordinates in 3D space. By using LiDAR instead of cameras, privacy is preserved while improving performance in different types of lightning. Four approaches have been proposed and implemented, of which three utilised machine learning and one was purely algorithmic. The machine learning models were trained and tested on synthetic data generated in Unreal Engine 5. The evaluation was based on three tests: number of people, crowd density, and distance between the LiDAR and the crowd. The approaches were also loosely validated on real data. The results show that the machine learning approaches have potential, however, the algorithmic approach performed the best.

Link to popular science summary: https://fileadmin.cs.lth.se/cs/Education/Examensarbete/Popsci/240531_11JakobssonRuneke.pdf


13:15-14:00 in E:2405 (Glasburen)

Presenter: Sofia Wahlmark
Title: Grouping of duplicate Android bug reports through log anomaly detection
Examiner: Emelie Engström
Supervisor: Qunying Song (LTH), Fredrik Henriksson (Volvo Car Corporation)

Bug solving is a critical aspect of modern software development, particularly within Android operating systems. Bug reports, containing extensive diagnostic information, are essential tools for analyzing and resolving issues in these systems. However, one significant challenge in bug resolution is the identification of duplicate bug reports. Existing studies primarily focus on detecting duplicate reports based on e.g. title and description, potentially overlooking cases where bugs manifest differently visually or in log data. In this thesis I utilize abnormal log sequences in the system logs to identify duplicate bug reports. This is done using the log parsing tool Drain followed by the LSTM-based anomaly detector Deeplog to extract abnormal logs which are used to measure similarity between bug reports. The results demonstrate performance comparable to existing duplicate detection methods, with Recall Rate@k of 0.275, 0.471, and 0.588 for k = 1, 3, 5 respectively. These findings are in line with the achieved performance in existing studies about duplicate detection and suggest the potential applicability of this method in industrial settings.

Link to popular science summary: https://fileadmin.cs.lth.se/cs/Education/Examensarbete/Popsci/240531_13Wahlmark.pdf


14:15-15:00 in E:4130 (Lucas) N.B. No more opponents for this presentation

Presenters: Yamen Albdeiwi, Mohammed El-Khalil
Title: Generating Test Cases Using Natural Language Processing
Examiner: Elin Anna Topp
Supervisor: Pierre Nugues (LTH)

Software testing today is a vital factor in maintaining the quality and reliability of software products in an always-advancing technical era.. Nevertheless, making manual high-quality test documents has been historically laborious and lengthy, which results in significant time and money consumption for the whole software creation process. Recently NLP, in particular, the use of transformer models, has become a prospective tool for automatisation of this procedure. This thesis is based on the use of language models to create test case documents from feature specifications written in natural language. We assess different ways to improve our model’s effectiveness, such as fine-tuning, prompt engineering, and agentic workflow methods. We carry out the research with the use of a quantized and optimized model for memory efficiency that demonstrates the possibility of generating good test cases even with the restriction of computational resources. With our approach, we achieved remarkable results in both BLEU and human evaluation scores. Our highest BLEU score achieved with our best approach is 31.60. The findings gathered can be applied in other language model optimization tasks, and they also provide useful hints as to the ways to make models work better with limited computational resources.

Link to popular science summary: To be added


15:00-15:45 in E:4130 (Lucas) N.B. No more opponents for this presentation

Presenters: André Frisk, Amjad Bakir
Title: Exploring Cloud Rendering Techniques for Aerospace Applications
Examiner: Elin Anna Topp
Supervisors: Michael Doggett (LTH), Magnus Wihlborg (Tactel AB)

Tactel's Arc project aims to transform the in-flight experience by integrating dynamic cloud graphics using advanced airplane monitor technology. This thesis explores optimising billboarding techniques for realistic cloud rendering from various spatial perspectives, considering hardware constraints and real-time weather data. The study addresses the neglect of aerial viewpoints in existing cloud rendering methods and proposes novel algorithms tailored for aerospace scenarios.. The research combines theoretical investigations with practical implementations to enhance visual representations and elevate in-flight entertainment standards.

Link to popular science summary: https://fileadmin.cs.lth.se/cs/Education/Examensarbete/Popsci/240531_15FriskBakir.pdf


15:15-16:00 in E:2116 N.B. No more opponents for this presentation

Presenters: Oliver Nederlund Persson, Elias Gustafsson
Title: Study of reactive services vs regular blocking vs virtual threads for a high-load service
Examiner: Flavius Gruian
Supervisors: Jonas Skeppstedt(LTH), Samuel Alberius (Sinch AB)

This study was conducted in order to test the performance of Java’s virtual threads against platform threads and reactive systems in the context of a high load server / client system. The experiments were designed to test performance in several ways, entailing methods with various mixes of simulated IO operations and computational tasks. For computational operations, neither virtual threads nor re-active streams outperformed regular platform threads. However, in IO-heavy operations both high-concurrency technologies performed better than platform threads. Reactive streams utilized less memory and had fewer fluctuations, while virtual threads had lower latencies. For instance, latencies for virtual threads outperformed reactive by up to 44% in the 99th percentile for the IO tests and had the best performance in three out of four benchmarks. Reactive had about half the memory usage of virtual threads in three out of four benchmarks and had the best memory performance in all four benchmarks.

Link to popular science summary: To be added


 

 



Om händelsen
Tid: 2024-05-31 09:00 till 16:00

Plats
See information for each presentation

Kontakt
birger [dot] swahn [at] cs [dot] lth [dot] se