Lunds Tekniska Högskola


CS MSc Thesis Presentation Day October 28


Tid: 2021-10-28 09:15 till 15:00
Plats: Online via zoom (separate link for each presentation)
Kontakt: birger [dot] swahn [at] cs [dot] lth [dot] se
Spara händelsen till din kalender

Four MSc theses to be presented on Thursday October 28, 2021

Thursday October 28 is a day for coordinated master thesis presentations in Computer Science at Lund University, Faculty of Engineering. Four MSc theses will be presented.

The presentations will take place online via Zoom, see separate link for each presentation. A preliminary schedule follows.

Note to potential opponents: Register as an opponent to the presentation of your choice by sending an email to the examiner for that presentation ( 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 (N.B. No more opponents for this thesis)

Presenter: Roland Veiderma Holmberg
Title: The Impact of AI-Based Tools on Software Development Work
Examiner: Per Runeson
Supervisors: Elizabeth Bjarnason (LTH), Björn Granvik (Softhouse Consulting Öresund AB)

What software development tools based on machine learning that there are and how they affect the developer are areas with a lack of previous research. We investigated this by performing a literature review of what such tools that exist followed by a case study with multiple developers at a case company. Our findings from the map suggest that the area is under heavy development but still that several tools are mature enough to be used professionally. Our case study findings suggest that machine learning based development tools are less understood and perceived to be less trustworthy than corresponding conventional tools. These are challenges we think are crucial to overcome in order to successfully introduce such tools in professional software development. Our findings can help future research on how such tools can improve and our map can act as a basis on what tools there are available today for future research.

Link to presentation:

Link to popular science summary: To be updated

10:15-11:00 (N.B. No more opponents for this thesis)

Presenter: Erik Bjäreholt
Title: Classifying brain activity using electroencephalography and automated time tracking
Examiner: Elizabeth Bjarnason
Supervisors: Markus Borg (LTH / RISE)

We investigate the ability of EEG to distinguish between different activities users engage in on their devices, building on previous research which showed a considerable difference in brain activity between code- and prose-comprehension, as well as differences during code- and prose-synthesis. We perform a replication study and improve upon past results using state-of-the-art machine learning classifiers based on Riemannian geometry. Furthermore, we extend the scope of previous work by introducing the automated time tracking application ActivityWatch to track what device activities the user is engaging in. This lets us label EEG data with naturalistic device activity, which we then use to train classifiers of device activity, such as code writing vs prose writing, or work vs media consumption.

Link to presentation:

Link to popular science summary:äreholt.pdf


Presenter: Jonathan Skogeby
Title: Exploring subjectivity in ad hoc assessment of open source software
Examiner: Ulf Asklund​​​​​​​
Supervisors: Martin Höst (LTH), Emil Wåreus (Debricked AB)

Five developers were asked to rank twenty projects on the social coding platform GitHub in terms of two aspects: popularity and quality of contributors. In each aspect, the developers were subsequently instructed to list and order the most important metrics used in their evaluation. This data was used to examine how well their subjective opinion could be represented a using metric-based linear model using data from the GitHub API. By applying Spearman's rank correlation coefficient, the combinations of metrics and their weights with the highest statistically significant correlation to each developer's ranking of each aspect were chosen for comparison against the developers' own narrative about which metrics were important in their evaluation. Although high correlations were achieved, the weighted metrics generated by the program showed very little resemblance to developers' reported metrics and order of importance. It was concluded that the implemented model was not successful in representing the developers' opinion, but the discussion did not rule out the possibility of better results in potential future work.

Link to presentation:

Link to popular science summary:


Presenter: Daniel Jogstad​​​​​​​
Title: Interactive Iterative Patent Search​​​​​​​
Examiner: Jacek Malec​​​​​​​
Supervisors: Pierre Nugues (LTH), Fredrik Edman (LTH), Henrik Benckert (Mindified)

Searching for prior art in databases with millions of patent documents is a very time-consuming process. These days, programs commonly use neural networks as an efficient way of finding similar documents to an input text to solve this. Through natural language processing, they translate the words and sentences into numerical vectors with which they, in some sense, can describe the general meaning of a text. In this Master's thesis, we have investigated the use of the neural network Sentence-BERT together with user input for gradual improvement. The program uses this input for re-ranking the search results. These are then evaluated for patent quality, ordering of the patents and how the method can heighten the rating of lowly rated but good patents. Our results show that it is possible, through simple mathematical operations, to implement an interactive, iterative patent search that improves the initial search results of the neural network.

Link to presentation:

Link to popular science summary: