Ten MSc theses to be presented on August 25, 2016.
August 25 is the day for coordinated master thesis presentations in Computer Science at Lund University, Faculty of Engineering. Twenty MSc theses will be presented.
The presentations will take place in the E-house, rooms E:2405 (Glasburen) and E:4130 (LUCAS). A preliminary schedule follows.
Note to potential opponents: Register as opponent to the presentation of your choice by sending an email to the examiner for that presentation (firstname.lastname@example.org). 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.
|PRESENTERS||Billy Johansson, Martin Lichstam|
|TITLE||Requirements Engineering in Startups with Open Source Software related business strategies|
|ABSTRACT||As startups are typically very resource constrained and open source potentially offers low cost labour and innovation sources; the intersection of startups and OSS is worth studying.|
In this thesis we perform a literature study on the current state of Requirements Engineering (RE) in startups and OSS, showing that the RE process is comparatively more informal in both startups and OSS than what would be the norm in classical RE. We also present interviews with four different startups and examine how they manage their OSS projects through an RE perspective, confirming the results of the literature study. RE related challenges faced by the startups are presented and a tool is produced; a four step iterative process to help startups leverage their OSS communities.
|PRESENTERS||Fredrik Folkesson, Johan Nyholm|
|TITLE||Clustering of back-end failures in automated testing|
|SUPERVISORS||Emelie Engström (LTH), Lars Andersson (Qlik)|
|ABSTRACT||Automated Software testing is becoming increasingly popular, which in turn creates more information that has to be analyzed. In order to decrease the amount of information it is desired that failed test cases are grouped together based on their failure origin. In this thesis we have looked into different types of back-end information and different methods for using this information in order to create clusters of failed test case executions originating from the same code defect. We created a prototype that clusters failed test case executions by analyzing methods names used in requests sent to the server. We did this using the vector space model in which we evaluated multiple approaches for weighting terms. The best approach seemed to be weighting the methods using a suspiciousness rating. The prototype shows great promise of working well at Qlik but further work and research has to be done to be conclusive.|
|TITLE||A Framework for Selecting an ERP Open Source System: A Case Study|
|ABSTRACT||An ERP system is a company spanning program that handles everything from production and logistics to sales and customer support but they are extremely costly to implement and thus excludes smaller businesses from usage. By us- ing an open source solution the total cost can be reduced to modification and support costs, eliminating the acquisition and license costs. In this thesis we construct a framework from previous related work and then use it to determine what criteria can be useful in an e-commerce scenario. This was done by per- forming a case study at a Swedish business-to-business e-commerce company that was looking to bundle their proprietary web shop offer with an ERP sys- tem to their clients. The criteria that were found to be of great importance were functional fitness, internationalization, security, programming language, community activity and license.|
|PRESENTERS||Axel Ahlbeck, Anton Jakobsson|
|TITLE||Distributing neural networks|
|SUPERVISORS||Patrik Persson (LTH), Mikael Lindberg (Axis), Niclas Danielsson (Axis)|
|ABSTRACT||A central problem with deep neural networks is that they contain a large number of parameters and heavy computations. The goal of the thesis is to investigate whether it is feasible to distribute and run a deep neural network on a network of cameras with tight constraints such as bandwidth and memory capacity.|
|PRESENTERS||Nils Fagerberg, Daniel Odenbrand|
|TITLE||0-1 Integer Programming in Associative Data Models|
|SUPERVISORS||Thore Husfeldt (LTH), José Díaz López (Qlik)|
Qlik is a company based in Lund that makes products for business intelligence (BI), that is products that help the customers visualize their data. As opposed to many other commercial BI products Qlik uses an associative model of data in their engine. This thesis has investigated the merits of implementing a feature which allows the solving of optimization problems in the data model in said engine.
We implement an implicit enumeration algorithm and compare it to the commercial optimizer Gurobi as well as the add on for Visual Studio: Microsoft Solver Foundation. It turns out that our algorithm is much better than the commercial products in some special cases (specifically easy problems with cheap optimal objective functions) even without exploiting the associativity of the engine in an optimal way. Future work may give our algorithm an even greater edge toward the commercial products.
|TITLE||Navigation Planning using Active Logic|
|EXAMINER||Elin Anna Topp|
With active logic, programs can adjust to the unreliability and incompleteness of real-world information and change beliefs accordingly. In a planner, this can be used to make on-the-fly changes to plans. In this thesis active logic is used to make navigation plans which are run on a wheeled robot. It is tested how well the logic system holds up compared to a more commonly used PDDL planner when controlling a physical robot with real-time requirements.
|TITLE||Construction of a PEG-based compiler|
|SUPERVISORS||Christopher Gautier (ARM), Jonas Skeppstedt (LTH)|
Compiler optimizations are typically implemented as a sequence of transformations, applied to some representation of source code. However, no guarantee is given that the ordering of these transformations is in any way optimal. This problem is commonly referred to as the phase-ordering problem. A Program Expressions Graphs (PEGs) are a code representation designed to makie it possible to perform optimizations without destroying information, by building several versions of the same program. We investigate this approach to optimization, and in doing so have developed a prototype extension to an existing ARM compiler. While time constraints prevented us from exploring the possibility of equality saturation, we identify several issues when translating a PEG back to imperative code. We present the issues we found and give our solutions to these. In addition, we have extended the Program Expression Graph construction and reversion algorithms to handle representations of programs on SSA form.
|TITLE||Patterns in live performance data|
|SUPERVISORS||Pierre Nugues (LTH) , Jens Gulin (Sony Mobile)|
|ABSTRACT||This thesis explores methods to find features that affect the start time of mobile apps. To help app-developers improve performance over patch cycles, an alarm pipeline is implemented in Apache Spark and tested. The implementation is able to detect notable changes in the start time distribution and alert the user. Spark is a scalable open source cluster computing framework, that proved to be well suited for the given problem. |
The program consists of five steps, preprocessing the data, fitting a Gaussian Mixture Model (GMM) to the data and finding differences in the distributions. Then a linear regression model is fit to map the available features to the GMM parametrization. The Linear regression enables the users to find relationships between the available features and the parameterization, by analyzing the regression weights.
The alarm pipeline is tested on two data sets and is shown to identify statistically significant changes in the start-time distribution.
|PRESENTERS||Mikael Gråborg, Oskar Handmark|
|TITLE||Summarizing Product Reviews Using Dynamic Relation Extraction|
The accumulated review data for a single product on Amazon.com could potentially take several weeks to examine manually. Computationally extracting the essence of a document is a substantial task, which has been explored previously through many different approaches. We explore how statistical prediction can be used to perform dynamic relation extraction. Using patterns in the syntactic structure of a sentence, each word is classified as either product feature or descriptor, and then linked together by association. The classifiers are trained with a manually annotated training set and features from dependency parse trees produced by the Stanford CoreNLP library.
In this thesis we compare the most widely used machine learning algorithms to find the one most suitable for our scenario. We ultimately found that the approach was very successful, reaching an overall FScore well over 80 percent. Following an aggregation procedure, the results are presented in a graphical interface displaying the key points regarding a product’s features together with a visualization of the end-to-end accuracy of the extraction.
|TITLE||Eventtjänst för diagnos av tunga fordon|
|SUPERVISORS||Sven Gestegård Robertz (LTH), Andreas Jonasson (Scania)|
The integration of computer units in vehicles has made more advanced in-vehicle functionality possible. To diagnose these more advanced functions, diagnostic services are used to extract data from the integrated computer units. These services give rise to an increased network load which today threatens to overload the network. There is therefore a need for new services which can contribute to reduce the load during vehicle diagnostics. One potential solution is to use an event based service. In the thesis a case study has been carried out at Scania in order to examine the pros and cons of such a service and to examine if it can be of any value. Different requirements on an event based service were investigated and a prototype service was implemented and evaluated. The thesis shows on different problems related to an event based service and how these could be solved.