Loading Events

« All Events

  • This event has passed.

Master Thesis Presentations

2015/06/04 @ 08:00 - 17:00

June 4 is the day for the ninth coordinated master thesis presentations in Computer Science at Lund University, Faculty of engineering. Twenty-one (21) theses will be presented in three parallel tracks. The presentations will take place in E-huset, rooms E:2116, E:2405, E:4130.

Track A (E:2116)

09:15 Gunnar Weibull: Graphical editor for Graphical User Interfaces for an “Internet of Things” system, Examiner: Ulf Asklund

The aim for the graphical editor described in this paper is to simplify the creation of Graphical User Interfaces (GUIs) for PalCom (an “Internet of Things”-solution). Ideally, the people who are to use the GUIs will be able to create them themselves.  Most graphical editors focus on placing graphical components (buttons etc.), then assigning them a function. The graphical editor described in this paper aims to introduce the opposite work flow, i.e. identify functionality and then suggest graphical components to express this functionality in the Graphical User Interface (GUI) to be created.

10:15 Mia Månsson: Dynamic installation and automatic update of Bluetooth low energy devices in the Internet of Things, Examiner: Görel Hedin

Chronic kidney disease is becoming an increasingly common illness.The medical background for this thesis comes from the prediction that frequent monitoring of body parameters may help prevent the progression of the disease. For other chronic diseases other parameters are interesting to monitor and the need is typically dynamic.

This thesis presents a solution for dynamic installation of Bluetooth low energy-enabled health care devices and automatic update of such services in the PalCom middleware. PalCom is aspiring to be a helpful tool in the pursuit of improving the quality of health care and provides functionality for taking measurements at home for remote monitoring. However, as most similar systems, PalCom only allows deployment of predefined  BLE devices. The contribution of  this work is a solution to install BLE-enabled health care devices in PalCom on demand, as well as a solution for automatic update of these services.

11:15 Jens Gustafsson, Alfred Åkesson: System for activity tracking of patients with kidney failure, Examiner: Görel Hedin

Many people suffering from kidney failure are in need of a kidney transplant. A problem in the health care is that patients cannot undergo surgery if they have too much belly fat. Regular exercise and physical activity are therefore crucial for this group of people. A project group at the Skåne University Hospital has recently been established to help patients with kidney failure to lose weight. A question they asked themselves was whether it was possible to use activity tracking devices in the project. The purpose of this thesis is to design and evaluate a system that can be used to track the physical activities of the patients. The system will be built using a Sony SmartBand, a wristband collecting and analyzing data about a user’s daily activities.

—– Lunch —–

13:15 Daniel Forsman: Creating a readable language for checking XML, Examiner: Görel Hedin

Today sharing data is done everywhere. Doctors might want to share patient journal information. Patient journals may contain sensitive information that doctors don’t want to share, the journals needs to be checked before they are shared. In this thesis, data and journals are coded in XML and checking journals and data is the same as validating XML.

Validating XML documents is usually done by following rules from a validator. A validator processes XML documents and checks that the structure follows the rules. The issue with these validators is that they can’t compare arbitrary nodes with each other and there are no mathematical operations to supply these comparisons.

This thesis attempts to solve the issue with existing solutions by creating a readable language for validating XML documents. The solution is done in three steps: investigating similar solutions, implementing a validator, and testing the readability of the validator with a usability test.

14:15 Anders Tilly & Victor Johnsson: Developing a Test Authoring Tool for a Modeling Language, Examiner: Görel Hedin

For the modeling language Modelica, there is no standard for the testing of Modelica models. In this thesis we investigate a new way to create and run tests by developing an external test authoring tool. Our tool was developed by first examining existing testing practices and then trying to improve upon those by iteratively developing the tool while continuously performing user studies with the users and the tool. The result is a functional tool capable of creating and running tests.

15:15 Johan Fänge: A Tracing JIT Compiler for Erlang using LLVM, Examiner: Görel Hedin

We have modified the Erlang runtime to add support for a tracing Just-In-Time (JIT) Compiler, similar to Mozilla’s TraceMonkey. Tracing is a technique to augment an existing interpreter with a JIT simply by recording the instructions executed during a loop iteration, and then generate optimized native code from this. Tracing compilers are particularly suited to optimize number crunching tight loops, an area where Erlang traditionally has been lacking. We make use of the LLVM compiler library to optimize and emit native code. In micro benchmarks we show some major improvements, reducing execution time by up to 75%. However, from an engineering point of view, we conclude that the effort of an industrial strength implementation would be substantial – essentially reimplementing large parts of Erlang’s interpreter – and discuss a potential solution based on recent research in the area.

16:15 Anton Fagerberg: Optimising clients with API gateways, Examiner: Görel Hedin

The thesis investigates the benefits and complications around working with API gateways. When we say API gateway, we mean to proxy and potentially enhance the communication between servers and clients, such as browsers, by transforming the data. We do this by examining the underlying protocol HTTP/1.1 and the general theory regarding API gateways.

An API gateway framework was developed to help further understand some of the common problems and provide a way to rapidly develop prototype solutions to them. This framework was then applied in three case studies in order to determine potential problematic areas and solve them in real world production systems. We could from these results see that the benefits of an API gateway varied from case to case, and with results in hand, predict in which scenarios API gateways are beneficial.

Track B (E:2405)

09:15 Felix Glas: Machine Learning Techniques for Customer Recommendations, Examiner: Thore Husfeldt

10:15 Edward Dunn Ekelund: Test Selection Based on Historical Test Data, Examiner: Emelie Engström

This thesis presents a method to enable test selection based on historical test data in order to minimize the time spent running unnecessary tests. Evaluation results show that the prototype program implemented for finding correlations between tests and code packages is robust, and running only the recommended tests is more effective than running the full regression test suite by a significant factor. On average it successfully identifies nearly all relevant tests (median success rate is 100%), and it discards 14 times more irrelevant tests compared to the default selection strategy of running all tests. Furthermore, the average size of the recommended subset is only 5% of the size of the full regression test suite, meaning that the time spent running tests can be decreased by a factor of 20.

11:15 Ture Karlsson: Automated build service to facilitate Continuous Delivery, Examiner: Martin Höst

Continuous Delivery can be seen as an evolution from agile software development methods and high demands to deliver working software quickly. It aims to always be able to deliver working and reliable software in short iterations by continuously integrate, build and test the software. This puts high demands on automation and the focus of this thesis will be to automate the pipeline between source code and deployable software artifacts. The problem definition of this thesis is to improve and unify the deployment pipeline of software running on Linux at IKEA IT. The thesis resulted in a service that supports Continuous Delivery by providing automated building, testing, signing and deployment of software. It runs in production environment at IKEA IT and provides a high level of automation. It was evaluated with help from end users and the evaluation showed that the service is useful and satisfies the goals.

—– Lunch —–

13:15 Nicklas Johansson & Christian Tenggren: Social Network Analysis of Open Source Projects, Examiner: Martin Höst

A large amount of widespread software used today is either open source or includes open-source projects. Much open-source software has proved to be of very high quality despite being developed through unconventional methods. The success of open-source products has sparked an interest in the software industry in why these projects are so successful and how this seemingly unstructured development process can yield such great results.

This thesis presents a study done on the projects hosted by one of the largest and most well-known open-software communities that exists. The study involves gathering developer collaboration data and then using social network analysis to find trends in the data that eventually might be used to create benchmarks for open-source software development. The results show that several interesting trends can be found.

14:15 Andreas Olsson & Thomas Mattsson: Cross-Platform Video Management Solutions, Examiner: Mathias Haage

With a multitude of platform and operating system combinations available to- day, ranging from laptops and workstations to tablets and smartphones, users want to use their favorite applications regardless of device. Cross-platform development has thus become more important in recent years. When devel- oping a new application the developers must decide what platforms to support and what strategy to use to reach out to them. By developing both native and cross-platform prototypes we try to find advantages and disadvantages of using a cross-platform strategy for video management applications. We show that it indeed is possible to develop cross-platform video management applications for both Windows and OS X and find both advantages and disadvantages of this strategy. The result of this thesis state that the choice of cross-platform or not depends much on the situation and the preferences of the developers.

15:15 William Martinsson & Robert Bagge: Energy Efficient Geo-Localization for a Wearable Device, Examiner: Jacek Malec

During the last decade there have been a surge of smart devices on markets around the world. The latest trend is devices that can be weared, so called wearable devices. These devices are small and usually set a high demand on energy efficiency, which makes traditional localization techniques unfeasible for them to use.

In this thesis we investigate and succeed in providing a localization solution for a wearable camera that is both accurate and energy efficient. Localization is done through a combination of Wi-Fi- and GPS positioning with a mean accuracy of 28 m. Furthermore we utilize an activity recognition algorithm with data from an accelerometer to decide when a new position estimate should be obtained or not. Our evaluation of the algorithm shows that by applying this method, 83.2% of the position estimates can be avoided with an insignificant loss in accuracy.

16:15 Jonathan Bratel: The Business Insight Ratio – Evaluating Customer Insights through Hybrid Models, Examiner: Jacek Malec

Customer segmentation and target analysis are two essential tasks when identifying a company’s customers. To perform these tasks, the author applies hybrid data-mining models, integrating clustering and decision trees. From previous research, we find that these models lack measures for evaluating the amount of insights revealed to decision-makers.

For this reason, we propose, test, and validate a new evaluation measure – the insight ratio. Through experiments on five separate datasets, we conclude that the measure enables an evaluation of the insights gained through the hybrid model. In each case, the measure generates the best result for the expected number of segments.

We then integrate the insight ratio with existing evaluation models to form the Business Insight Ratio. This measure evaluates customer segmentation and target anal

Track C (E:4130)

09:15 Max Åberg & Alexander Söderberg: Optimizing business intelligence extraction speed from an ERP-system’s database, Examiner: Per Andersson

The popularity of cloud-based applications has increased the importance of performance optimized databases. In this thesis the database for a cloud-based ERP-system called BeX®Online is investigated with the goal to optimize the generation speed of reports in the business intelligence functionality in the case system. The result is a set of recommended solutions to bottlenecks and inefficiencies found during the analysis, as well as suggestions for architectural improvements. Improvements to the generation speed of the business intelligence reports could be made in query tuning, index fragmentation, maintenance planning and table partitioning. Although database optimization is a widely researched subject it remains a complex and time consuming process, and the thesis concludes with a recommendation of areas that need to be further investigated.

10:15 Eric Ottosson & Suraj Sinha: Database Paradigms for Recordings Management, Examiner: Per Andersson

The relational database has long been considered the de facto standard for managing data in software applications. Today, the need for more scalable, flexible and distributed software solutions has led to the development of NoSQL (“Not only SQL”) database technologies that aim to replace the relational database in applications where such features are needed.

In this thesis we have investigated the potential benefits when replacing SQLite, the database used by Axis Communications to manage recordings in their camera products, with a NoSQL database in an embedded camera system. To evaluate their performance test cases were designed, based on important functionality in Axis’ storage solution, to measure execution times and resource consumption for database operations.

In the end the document database EJDB was identified, and was found to be more efficient than SQLite at creating, updating and removing records. It was, however, less efficient when performing queries based on conditional operators.

11:15 Erik Henriksson: Source Based Non-Minimal Adaptive Routing in Low Diameter Network Topologies, Examiner: Jonas Skeppstedt

The latency of networks depends closely on the network diameter. Unfortunately, obtaining good performance in low-diameter networks is often difficult and requires a thorough understanding of both the characteristics of the network topology and the applications.

This thesis investigates the characteristics of the topologies DragonFly and SlimFly. These have a low diameter, which means that packets need to traverse a low number of switches in the network.

This thesis presents an addressing scheme for the DragonFly topology. We describe how to perform non-minimal routing using commodity hardware. We provide a flow control for the transport protocol MPTCP.

We determine (based on experimental analysis) that the latency in a network depends in a large degree on the flow control. Our results show that it is possible to reduce the latency during load with minimal affect to the throughput. The results show that our flow control have 20%-75% lower latency than TCP.

—– Lunch —–

13:15 David Larsson & Antonina Tokarchuk: Randomized word parallel algorithms for detection of small induced subgraphs, Examiner: Jonas Skeppstedt

Induced subgraph detection is a widely studied set of problems in theoretical computer science, with applications in e.g. social networks, molecular biology and other domains that use graph representations. Our focus lies on practical comparison of some well-known deterministic algorithms to recent Monte Carlo algorithms for detecting subgraphs on three and four vertices. For algorithms that involve operations with adjacency matrices, we study the gain of applying word parallelism, i.e. exploiting the parallel nature of common processor operations such as bitwise conjunction and disjunction. We present results of empirical running times for our implementations of the algorithms. Our results reveal insights as to when the Monte Carlo algorithms trump their deterministic counterparts and also include statistically significant improvements of several algorithms when applying word parallelism.

14:15 Jerry Lindström & Stefan Nannesson: Implementing a streaming application on a processor array; A case study on the Epiphany architecture, Examiner: Krzysztof Kuchcinski

A modern way of processing information is to do it in parallel. This Master Thesis conducts a case study of how to parallelize a streaming application on a highly parallel platform.

This involves porting a real-world application, written in a stream processing language and compiled by tools, developed by the Embedded System Design research group at Lund University, onto a platform including an embedded processor array (the Adapteva’s Epiphany), an ARM processor, and programmable logic.

The driver application that we have used is a video decoder. The host platform is a Parallella board, with a dual core ARM host processor and a 16-core Epiphany coprocessor.

Our Master Thesis covers the creation of some library elements to support complex applications on that platform, such as FIFOs between Epiphany cores and the ARM host, some components that handle access to external RAM and a component that draws pixels onto a screen.

15:15 Karl Hylén: Processor Models for Instruction Scheduling using Constraint Programming, Examiner: Krzysztof Kuchcinski

Instruction scheduling is one of the most important optimisations performed when producing code in a compiler. The problem consists of finding a minimum length schedule subject to latency and different resource constraints. This is a hard problem, classically approached by heuristic algorithms. In the last decade, research interest has shifted from heuristic to potentially optimal methods. When using optimal methods, a lot of compilation time is spent searching for an optimal solution. This makes it important that the problem definition reflects the reality of the processor.

In this work, a constraint programming approach was used to study the impact that the model detail has on performance. Several models of a superscalar processor were embedded in LLVM and evaluated using SPEC CPU2000. The result shows that there is substantial performance to be gained, over 5% for some programs. The stability of the improvement is heavily dependent on the detail of the model.

16:15 Johan Ullén: Recommender System Validation Platform, Examiner: Jonas Skeppstedt

With most applications where recommender systems are used it is important that they produce a better result than a system with no recommender or one with a previous recommender. Deploying an untested system, even to a smaller user sample can be very costly if the system produces negative results. It is often in a developer’s interest to create several candidate systems and will thus need some way of comparing them before selecting one or a few to launch. While the methods of testing have been explored and their statistical soundness motivated in other work. This paper will focus on how to set up a generic and modular environment for offline experimentation, and discuss what parts can be generalized over different problems, the needs for application specific implementations, and how to test models with different types of data.

Programvaruportalen (Software Portal) is an organizational network and a portal for the information and knowledge available within the areas of computer science, software and software engineering at the Faculty of Engineering (LTH) and Lund University (LU): http://www.lth.se/programvaruportalen


08:00 - 17:00
Event Category: