CS Extra MSc Thesis Presentation Day June 15!
Place: Online Zoom presentations
Contact: birger [dot] swahn [at] cs [dot] lth [dot] se
Save event to your calendar
Five MSc theses to be presented on Monday June 15, 2020
Monday June 15 is an extra day for coordinated master thesis presentations in Computer Science at Lund University, Faculty of Engineering. Five MSc theses will be presented. Other presentations will take place on June 11 and June 12.
The presentations will take place online via Zoom, see separate link for every presentation. A preliminary schedule follows.
Note to potential opponents: Register as an opponent to the presentation of your choice by following the instructions on Doodle at https://doodle.com/poll/cpfr4bh5npvfstra. 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.
Presenters: Emil Bengtsson, Fredrik Nyberg
Title: Memory Fault and Leak Detection in a Constrained Embedded Linux System
Examiner: Jonas Skeppstedt
Supervisors: Flavius Gruian (LTH), Paul Asterland (Volvo Cars)
Memory errors are the cause of many bugs in software written in C and C++. Embedded systems often run software written in these languages, and often have very limited resources. They are also susceptible to errors, since they are often supposed to run for long times without crashing. This makes analysing the software rewarding, but the low resource availability can make dynamic analysis tools challenging to run. We have evaluated different dynamic memory analysis tools, intending to provide a recommendation of tools which are useful for debugging software running on embedded Linux. We have evaluated both the tools' error finding capabilities, and how much of an impact they have on performance. We have also used the most promising tools to analyse some of Volvo's programs in a real setting. Based on our findings, we recommend a few tools that we consider useful for analysing software on embedded Linux.
Link to presentation: https://lu-se.zoom.us/j/62635030935
Code for opponents: 31 JS
Presenter: Michael Hansen
Title: Domain Adapting English Speech Recognition to Swedish
Examiner: Jacek Malec
Supervisor: Pierre Nugues (LTH)
Automatic speech recognition (ASR) is a technology that allows the computer to transcribe human speech into text. Keyword spotting is a subfield of this, which restricts the problem to the identification of a few selected words. Most current recognition algorithms require huge corpora of annotated spoken data. In this project, I developed a system to use an English dataset from Google to train a speech recognition model, and then adapting that model to a smaller, Swedish dataset, which I collected myself. I could reach an accuracy of 86% with a dataset 100 times smaller than the original one. This is an improvement of 29% lower false rejection rate compared to a model not previously trained on English. With this system, specified applications can be feasibly constructed without the need for a large dataset, making it easier to make mobile systems utilise verbal keyword spotting.
Link to presentation: https://lu-se.zoom.us/j/67814419237
Code for opponents: 7 JM
Link to popular science summary: https://fileadmin.cs.lth.se/cs/Education/Examensarbete/Popsci/200615/11Hansen.pdf
Presenter: Mattias Jonsson
Title: SLAM in Low Speed Scenarios Using Ultrasonic Sensors
Examiner: Jacek Malec
Supervisors: Elin Topp (LTH), Magnus Wendt (Volvo Cars)
This master's thesis investigates how ultrasonic sensors on cars can be used to assist a driver in low speed parking scenarios. A simple odometry model based on rear wheel rotation sensors is implemented. Using this model to approximate the car's position, a Bayesian map of the environment is built from ultrasonic sensor data. Based on the results from this, the possibility of using SLAM with data from ultrasonic sensors is investigated. The simple odometry model is sufficiently accurate for the use cases in this thesis. The map built from ultrasonic sensor data is accurate enough for aiding the driver in some not too tight low speed parking scenarios. SLAM using ultrasonic sensor data is found to be feasible given a known simple structure of the environment according to which the sensor readings can be interpreted before being used as input to the SLAM algorithm.
Link to presentation: https://lu-se.zoom.us/j/68799394374
Code for opponents: 27 JM
Link to popular science summary: https://fileadmin.cs.lth.se/cs/Education/Examensarbete/Popsci/200615/13Jonsson.pdf
Presenters: Fredrik Olsson, Gustav Handmark
Title: Sentiment Analysis and Aspect Extraction for Business Intelligence
Examiner: Jacek Malec
Supervisors: Pierre Nugues (LTH), Oskar Handmark (Backtick Technologies AB)
Data mining and predictive analysis are important instruments in business intelligence. This should lead to insights essential for identifying strategic business opportunities. The amount of publicly available data can be a daunting task to process manually, which is why automated approaches have become popular. In this thesis we explore current state-of-the-art NLP techniques for processing company targeted customer reviews to provide meaningful and actionable insights. Our approach is two-fold. First we train, fine-tune and evaluate multiple different models for sentiment analysis of review texts. Secondly, we conduct aspect-based opinion mining to extract fine-grained information in every review text. The results are aggregated and displayed in multiple graphs and informative tables allowing easy interpretation of the data and the captured trends. This is done for two languages, English and Swedish. We found that with current technology we are able to train models that are effective and achieves good results on benchmarks.
Link to presentation: https://lu-se.zoom.us/j/67791016123
Code for opponents: 19 JM
Link to popular science summary: https://fileadmin.cs.lth.se/cs/Education/Examensarbete/Popsci/200615/14OlssonHandmark.pdf
Presenter: Anton Engström
Title: Towards agile data engineering for small scale teams
Examiner: Per Runesson
Supervisor: Emma Söderberg
Enabling production-level machine learning is hard. In producing machine learning models for industry, a majority of the time is spent on working on the data and the infrastructure that will support the model in its production environment. In this thesis, we investigate the challenges related to data engineering for machine learning purposes. The purpose is not to tackle the challenges related to big data. Instead, it focuses on a small scale context with few data sources and small datasets. The context is studied through three iterations; a literature study, a case study of a small scale machine learning company and a mapping of agile principles from the software engineering domain to the data engineering domain.. We note that there is a gap between literature and the needs in a small context. We note that the potential of a knowledge transfer from agile principles to the data engineering domain is promising.
Link to presentation: https://lu-se.zoom.us/j/61081936812
Code for opponents: 26 PR