lunduniversity.lu.se

Digit@LTH

Faculty of Engineering, LTH

Denna sida på svenska This page in English

Digit@LTH: Events

CS MSc Thesis Presentation Day June 11!

Föredrag

From: 2020-06-11 09:00 to: 17:00
Place: Online Zoom presentations
Contact: birger [dot] swahn [at] cs [dot] lth [dot] se
Save event to your calendar


22 MSc theses to be presented on Thursday June 11, 2020

Thursday June 11 is a day for coordinated master thesis presentations in Computer Science at Lund University, Faculty of Engineering. 22 MSc theses will be presented. Other presentations will take place on June 12 and June 15.

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.

 

Schedule


09:00-10.00

Presenters: Måns Axelsson, Madeleine Jansson
Title: A Federated Learning Method Used to Detect Credit Card Fraud
Examiner: Volker Krüger
Supervisors: Rasmus Ros (LTH), Mehmood Khan (IBM), Peter Forsberg (IBM)

Credit card fraud is a problem for consumers and banks all over the world, thus, an effective Fraud Detection System (FDS) is important. Commonly, FDS are built by using machine learning algorithms and to get a well performing model a large dataset is required. Though, due to the skewness of the credit card transaction dataset as well as the security and privacy attached to it a centralised FDS is not the best suited approach. Instead, we propose an FDS trained with federated learning, a machine learning setting where multiple entities collaborate under the coordination of a central server. With this approach banks can get benefits of a shared model, which has seen more fraud than each bank alone, without sharing datasets. Results in this thesis indicates that Federated Averaging can perform and outperform the Multi Layer Perceptron when detecting credit card fraud.

Link to presentation: https://lu-se.zoom.us/j/7563917023

Code for opponents: 6 VK

Link to popular science summary: https://fileadmin.cs.lth.se/cs/Education/Examensarbete/Popsci/200611/09AxelssonJansson.pdf


09:00-10.00

Presenters: Frida Gunnarsson, Sara Trygve
Title: Smart Personalization for In-flight Entertainment Systems
Examiner: Jörn Janneck
Supervisors: Pierre Nugues (LTH), Sören Just Pedersen (Tactel)

In-flight entertainment (IFE) systems are making a shift towards a more personalized experience for the passengers. A problem is the lack of initial user information, which is often solved with a login or a connection to a user’s own mobile device. However, this solution only covers a small amount of customers. Therefore, this thesis will focus on creating a personalized experience for the rest of the passengers. Different approaches to provide recommendations will be investigated, such as content-based recommendations and market basket analysis. Techniques such as one hot encoding, word embeddings and apriori are utilized. Using data from past passenger interactions with existing IFEs, the experiments are evaluated by a comparison of a given recommendation and previous watched item. The final model gives category recommendations as well as ranks the items after similarity compared to previous watched content.

Link to presentation: https://lu-se.zoom.us/j/4787325598

Code for opponents: 15 JJ

Link to popular science summary: https://fileadmin.cs.lth.se/cs/Education/Examensarbete/Popsci/200611/09GunnarssonTrygve.pdf


09:00-10.00

Presenter: Johanna Hultén
Title: An internal assessment method for the HAVOSS maturity model
Examiner: Emelie Engström
Supervisor: Martin Höst (LTH)

The Handling Vulnerabilities in third party OSS (HAVOSS) maturity model does not have an accompanying assessment methodology. In this thesis an internal assessment methodology for the is developed and evaluated. The resulting process assessment methodology is a four phase lightweight assessment methodology. The four phases for the process assessment are a preparation phase, followed by a digital questionnaire completed by a larger number of participants, called the individual assessment. This is then followed by a workshop that is phase three where the questionnaire is completed by the workshop with the digital tool as a guide. The final phase is generating the report from the assessment. A digital tool prototype was also developed to aid the assessment process and visualize the results. The methodology and digital tool were evaluated by staff at Lund University and two industry representatives.

Link to presentation: https://lu-se.zoom.us/j/64786449668

Code for opponents: 16 EE

Link to popular science summary: https://fileadmin.cs.lth.se/cs/Education/Examensarbete/Popsci/200611/09Hulten.pdf


10:00-11.00

Presenters: Oskar Widmark, Emil Ahlberg
Title: Data-Driven Feature Development
Examiner: Martin Höst
Supervisors: Ulf Asklund (LTH), Markus Andersson (AXIS Communications)

Data-driven decision making in feature development is the continuous utilization of customer feedback data and post-deployment data in the development process, intended to focus development resources efficiently. This thesis aims to do two things: to find evaluation criteria that can be used to evaluate features of a mobile application, and to explore the next step in a data-driven development strategy for an organization, by conducting an online controlled experiment as a proof-of-concept implementation. Development for mobile applications that are not directly monetized can be focused on improving user happiness. Therefore, we try to find a link between user behavior based on post-deployment data and user happiness. The data is processed into evaluation criteria, which show different patterns in user behavior. Four different features are chosen and ordered based on their scoring on the evaluation criteria. The evaluation basis of this ordering is obtained by conducting an online survey, directly asking for the users' opinion of the features. While the ordering produced was the same between the results and the evaluation, the limited amount of responses limits the certainty of the conclusion that usage data can be used to measure happiness, and more evaluation data from a more representative sample of the user base is needed. The proof-of-concept A/B-test was implemented and deployed to live users. Two implementations of a feature were compared to a control group using the aforementioned evaluation criteria. When the results were interpreted, the organization procured additional knowledge of the application and its user base; The two implementations performed better than the control group. We conclude that the A/B-testing framework in many ways is a superior way of experimenting with new ideas in a data-driven manner.

Link to presentation: https://lu-se.zoom.us/j/7512849026

Code for opponents: 30 MHö


10:00-11.00

Presenters: Otto Sörnäs, Erik Gralén
Title: Usage pattern recognition for efficient pre-caching
Examiner: Emelie Engström
Supervisors: Rasmus Ros (LTH), Jens Argentzell (Qlik)

This thesis presents a machine learning-based solution to a precaching-problem, applied in a cloud-based environment, by training two different neural networks with labeled data; namely an FNN and a TCN. This approach relies on analyzing log files in order to concretize a usage pattern on which the neural networks can train. These models are subsequently evaluated and compared to each other. Based on the results we can conclude that both of the machine learning models are potential solutions to this problem, albeit with their own strengths and weaknesses. TCN has a potential higher prediction accuracy, but is more computationally expensive to train. FNN, on the other hand, is a more lightweight approach that has a bit lower accuracy. It is still, however, an adequate solution if applied in a context with strict memory and computational constraints.

Link to presentation: https://lu-se.zoom.us/j/64786449668

Code for opponents: 8 EE

Link to popular science summary: https://fileadmin.cs.lth.se/cs/Education/Examensarbete/Popsci/200611/10SornasGralen.pdf


10:00-11.00

Presenter: Liam Neric
Title: Sim-To-Real: Domain Adaptation of Robot Trajectories with LSTM
Examiner: Volker Krueger
Supervisors: Elin Anna Topp (LTH), Alexander Durr (LTH)

Training intelligent agents in autonomous robotics is a data-intensive process, but gathering data from robotic experiments can be a costly and inefficient process. Robot simulation on the other hand offers an efficient and consistent way to gather data instead, but can often be inaccurate and fail to capture real world complexities. We look into the problem of inadequate accuracy in robot simulators by investigating the discrepancies in trajectories between simulation and reality for a Franka Emika Panda robot.. In our experiment we create an extensive data set with free movements of our robot, repeat it in simulation, and subsequently use this data to develop a Long-Short Term Memory architecture that can transfer simulated sensor readings of position, velocity and torque into more realistic ones. Our architecture was able to compensate for the simulator's shortcomings, the estimation of torque, and improved the root mean square error between simulated and real torque with at least 70%.

Link to presentation: https://lu-se.zoom.us/j/61094214487

Code for opponents: 20 VK

Link to popular science summary: https://fileadmin.cs.lth.se/cs/Education/Examensarbete/Popsci/200611/10Neric.pdf


10:00-11.00

Presenters: Kevin Johansson, Olof Nordengren
Title: Neural Network Model Evaluation on Satellite Imagery Classification
Examiner: Jörn Janneck
Supervisors: Jacek Malec (LTH) , Frank Camara (ÅF Pöyry)

Digital maps are built manually, making mistakes inevitable. Various functional city zones make up cities, i.e. residential areas, industrial areas, park areas. These real-world areas are also subject to change, rendering the map inaccurate, creating a need for maintenance. Eight neural network architectures are evaluated on a classification problem, with the goal of finding the best performing architecture, to be used to automatically aid with the maintenance problem presented above. The data consists of 5-band satellite imagery covering cities and the classes consist of six different functional city zones. We do coarse-grained image segmentation and make use of OpenStreetMap to create a process for automatic annotation. We evaluate each architecture with and without transfer learning, fed with three (RGB) and five bands, respectively. The best-performing architecture was DenseNet (89.5% test accuracy with all bands, no transfer-learning), which we argue makes sense because of the high detail-level that satellite imagery contains.

Link to presentation: https://lu-se.zoom.us/j/65758417761

Code for opponents: 1 JJ

Link to popular science summary: https://fileadmin.cs.lth.se/cs/Education/Examensarbete/Popsci/200611/10JohanssonNordengren.pdf


10:15-11.00

Presenters: Pontus Olsson, Samuel Klingström
Title: Type Inference in PHP using Deep Learning
Examiner: Görel Hedin
Supervisors: Niklas Fors (LTH), Johannes Claesson (Axis), Simon D Månsson (Axis)

Dynamically typed programming languages such as PHP, JavaScript and Python have recently started supporting gradual typing, where type annotations can be added to part of the code. Tools that can perform type inference are therefore becoming increasingly helpful as they could ease the labor intensive task of updating legacy code for developers. However, for PHP, most static code analysis tools have lacking or unsatisfactory type inference functionality. In this thesis, we use deep learning to predict type annotations for parameters in PHP. The neural network can, given a function or method, predict the type annotations for the parameters based on their usage. The predictions are then presented in the code comment. This approach builds upon the previous work, code2vec, and is based on the idea of representing code as paths in its abstract syntax tree. After training the model with the 10,000 most popular PHP repositories from Github, it was able to correctly predict type annotations with a top-1 accuracy of 75.6 % and a top-3 accuracy of 83.8 %. These results are significantly better than any current code analysis tool for PHP. We conclude that deep learning can successfully be used for type inference with great results.

Link to presentation: https://lu-se.zoom.us/j/63963644167

Code for opponents: 9 GH

Link to popular science summary: https://fileadmin.cs.lth.se/cs/Education/Examensarbete/Popsci/200611/10OlssonKlingstrom.pdf


11:00-12.00

Presenters: Tony Liu, Arvid Mildner
Title: Training Deep Neural Networks on Synthetic Data (Analysis of the Effect on Object Detection in Cityscapes)
Examiner: Jörn Janneck
Supervisors: Pierre Nugues (LTH), Martin Ljungqvist (Axis C. AB), Otto Nordander (Axis C. AB)

To train well-behaved generalizing neural networks, sufficiently large and diverse datasets are needed. Collecting and annotating these datasets are both resource heavy and time consuming. An approach to overcome these issues is to use computer-generated data for training.. In this thesis, we investigate this idea by training the object detector YOLOv3 on synthetic images. We present a performance comparison between leveraging synthetic data and training on only real data. Our results show that performance is increased using synthetic data when small amounts of real data is available. However, this performance increase is not as noticeable as more real data is introduced. To analyze how models trained on synthetic and real data differ, different similarity and sensitivity metrics are used. The results of this analysis indicate that the models transfer learned on synthetic data makes new and different predictions and generalize better compared to a model trained only on real data.

Link to presentation: https://lu-se.zoom.us/j/4787325598

Code for opponents: 2 JJ

Link to popular science summary: https://fileadmin.cs.lth.se/cs/Education/Examensarbete/Popsci/200611/11LiuMildner.pdf


11:00-12.00

Presenters: Vilhelm Åkerström, Anders Klint
Title: Continuous Delivery: Challenges, Best Practices, and Important Metrics
Examiner: Elizabeth Bjarnason
Supervisors: Lars Bendix (LTH), Axel Franke (Robert Bosch AB), Peter Walls (Robert Bosch AB)

Continuous Delivery (CD) is a practice that can help reduce long release times, and give companies the ability to quickly react to customer demands. Although this practice has been well documented and adopted by more and more companies, there is still some confusion in the software industry about how to utilize CD for the best effect on the productivity of teams. This thesis will seek to answer what the challenges and best practices of CD are, as well as what metrics are useful to look at when looking to improve the CD workflow. In order to answer these questions a literature study, combined with interviews and the creation of a proof of concept for displaying metrics all done at Robert Bosch AB will be used. This gave a list of challenges, best practices, and some useful developer feedback that could be used for future studies.

Link to presentation: https://lu-se.zoom.us/j/4608071745

Code for opponents: 11 EB


11:00-12.00

Presenter: Ola Westerlund
Title: Representing and Grouping Technical Issues for Business Insights
Examiner: Jacek Malec
Supervisors: Pierre Nugues (LTH), Astrid Nielsen (Tetra Pak AB)

In any product, errors are inevitable. In a large corporation with many products, prioritizing the correct errors is crucial but often non-trivial. Given an issue data base consisting of hundreds of thousands of data points, all containing a mixture of data, including free text, this thesis presents an automatic solution for representing and grouping these issues to aid business analysts in their prioritization decisions. Previous studies have shown the value of applying natural language processing and machine learning techniques in finding meaningful relationships between sentences and documents. In this thesis, these earlier findings were applied to the domain of machine errors. Embedding techniques and clustering algorithms were applied to error data. The results show that with sufficient data and state of the art sentence embeddings, meaningful clusterings can be constructed, keywords can be extracted, and new issues can be successfully linked to existing clusters.

Link to presentation: https://lu-se.zoom.us/j/69586685638

Code for opponents: 3 JM

Link to popular science summary: https://fileadmin.cs.lth.se/cs/Education/Examensarbete/Popsci/200611/11Westerlund.pdf


11:00-12.00

Presenters: Andreas Warvsten, Karl Lundberg
Title: Automated Fuzzy Logic Risk Assessment and its Role in Continuous Deployment
Examiner: Martin Höst
Supervisors: Ulf Asklund (LTH), Jonas Schultz (Sinch Sweden AB)

Risk management is a key component of the DevOps philosophy. In contrary tothe general paradigm of making data driven decisions, it is an area that largely is based on subjective predictions and gut feel of involved competence. One of the benefits of a DevOps work ow that includes versioning tools, is the metadata stored during the development process. This paper studies if this metadata can be used in order to automate some parts of the risk assessment process, and to what extent subjectivity can be removed from the process.

Link to presentation: https://lu-se.zoom.us/j/7512849026

Code for opponents: 24 MHö

Link to popular science summary: https://fileadmin.cs.lth.se/cs/Education/Examensarbete/Popsci/200611/11WarvstenLundberg.pdf


13:00-14.00

Presenters: Axel Jensen, Adam Koch
Title: Detecting Stitching Errors in Panoramic Images - A Deep Learning Based Approach
Examiner: Volker Krüger
Supervisors: Pierre Nugues (LTH), Tim Borglund (Axis Communications AB)

Panoramic image stitching is a powerful tool for creating video surveillance systems with a very broad view. Cameras utilizing such panoramic features are useful when monitoring big spaces such as airports, train stations and malls with a single unit. Convolutional neural networks have proven very effective in image stitching tasks: Nonetheless, they have not been applied to detect stitching errors. In this Master's thesis, we explore the detection of stitching errors, both in direction and magnitude, for Axis camera surveillance systems. The results show that neural networks with smaller architectures are good at finding clear cut stitching errors, but have difficulties detecting errors on alpha blended images. Our recommendation for future research is to experiment with larger architectures and see if they can detect such errors.

Link to presentation: https://lu-se.zoom.us/j/7563917023

Code for opponents: 13 VK

Link to popular science summary: https://fileadmin.cs.lth.se/cs/Education/Examensarbete/Popsci/200611/13JensenKoch.pdf


13:00-14.00

Presenters: Franz Lang, Alexander Mjöberg
Title: Prototyping - a Technique in Requirements Engineering
Examiner: Björn Regnell
Supervisors: Elizabeth Bjarnason (LTH), Maria Blomberg (Telavox AB)

Requirements Engineering is an important part of Software development. Software development has largely moved towards agile practices and the field of Requirements Engineering is no exception. Prototyping has been identified by one study as a technique that can solve challenges surrounding Agile Requirements Engineering. To find out what prototyping can offer a case-study is conducted with a company seeking to develop a new product using agile software development methodology. Information about prototyping is gathered through various means and leads to the creation of a set of agile prototyping guidelines that can help companies establish how prototyping can support their Requirements Engineering.

Link to presentation: https://lu-se.zoom.us/j/4608071745

Code for opponents: 22 BR

Link to popular science summary: https://fileadmin.cs.lth.se/cs/Education/Examensarbete/Popsci/200611/13LangMjoberg.pdf


14:00-15.00 N.B. Change of time

Presenter: Gustaf Backman
Title: Forecasting Financial Indices from Financial News
Examiner: Marcus Klang
Supervisors: Pierre Nugues (LTH), Edvard Sjögren (Kidbrooke Advisory)

I have evaluated the predictive power of financial news headlines on movement of financial indices. The tested indices are S&P 500 and US treasury rate with 1 and 3 years maturity. Text representation of different complexity was used, from TF-IDF to recent transformer models such as BERT and Sentence-BERT. My experiments show that no model has significant predictive power on future movements. Some models does however estimate the movement of S&P 500 from day k-1 to day k given the news from day k better than a random classifier.

Link to presentation: https://lu-se.zoom.us/j/67651926005

Code for opponents: 18 MK

Link to popular science summary: https://fileadmin.cs.lth.se/cs/Education/Examensarbete/Popsci/200611/14Backman.pdf


14:00-15.00

Presenters: Rasmus Berggren, Dennis Londögård
Title: Benchmarking and comparison of a relational and a graph database in a CMDB context
Examiner: Per Andersson
Supervisors: Lars Bendix (LTH), Guido Guidos (Axis)

Axis has grown a lot in the past decade and they are now having trouble keeping track of everything efficiently and hypothesize that they are in need of a CMDB, while also wondering what type of database it would be best implemented as.The goal of this thesis was to create a CMDB requirement specification and compare its validity as a graph and relational database. The work was done by first creating a requirement specification for Axis based on a literature study as well as interviews, from which a minimum viable product was formed to benchmark the performance and maintainability. We were able to create a requirement specification specifically for Axis. The results from the benchmarking suggests that relational databases are better from a performance perspective, while graph databases are easier to maintain. The conclusion is that a CMDB will benefit Axis and is slightly better as a graph database.

Link to presentation: https://lu-se.zoom.us/j/63202856027

Code for opponents: 14 PA

Link to popular science summary: https://fileadmin.cs.lth.se/cs/Education/Examensarbete/Popsci/200611/14BerggrenLondogard.pdf


14:00-15.00

Presenters: Anton Gudjonsson, Felicia Sucurovic Hedström
Title: Drones in the Cloud: A Study of IoT Architectures and Simulation in AWS
Examiner: Flavius Gruian
Supervisors: Ulf Asklund (LTH), Christian Eriksson (Dewire by Knightec)

With IoT devices becoming both smaller and more computationally powerful, new cloud computing architectures have evolved. These new architectures has presented an opportunity to use IoT for more purposes, such as autonomous drones which is what will be focused on in this thesis. As a result of this, many cloud computing platforms now offer support for hosting IoT networks. Amazon Web Services will be used for developing and hosting the IoT networks. The three IoT architectures which will be investigated in this thesis are Cloud, Fog and Edge computing. These architectures will be evaluated using metrics relevant for a real-time IoT system. Since IoT networks can consist of many devices it might not be feasible to create large networks of physical devices for testing. Therefore there is a need to investigate whether simulation of devices is something that can be used for testing, especially smart devices such as drones. As a solution to this, this thesis will explore the possibilities of simulation using AWS.

Link to presentation: https://lu-se.zoom.us/j/61795474317

Code for opponents: 29 FG


15:00-16.00

Presenters: Simon Åkesson, John Helbrink
Title: Data anonymization using machine learning and natural language processing
Examiner: Jacek Malec
Supervisors: Pierre Nugues(LTH), Michael Truong och Jianhua Cao (Sinch AB)

With the introduction of the General Data Protection Regulation (GDPR),the demand to protect personal identifiable information from unnecessary exposure is growing. The objective of this thesis is to explore and constructa system that will use natural language processing and machine learning toanonymize data. This will provide further analysis opportunities and additional protection for the single person possible. We implemented and com-pared different neural network architectures and frameworks. Using a biLSTM network together with contextualized embeddings and conditional random fields, state-of-the-art results were reached with the bestF1score at 95.58%. This shows the importance of contextual awareness as well as the impactof pre-trained word embeddings. However, human supervision will still beneeded for it to completely guarantee the anonymization of the messages. Ultimately, the resulting system fulfills its purpose of masking texts, removingthe risk of unnecessary information exposure.

Link to presentation: https://lu-se.zoom.us/j/65328829910

Code for opponents: 4 JM


15:15-16.00

Presenters: Jakob Hök, Martin Lindström
Title: Quantization Profiler for Artificial Neural Networks
Examiner: Flavius Gruian
Supervisors: Jörn Janneck (LTH), Axel Berg (ARM Sweden AB), Kevin Wohnrade (ARM Sweden AB)

We develop a software framework that is able to modify implementations of operators within artificial neural networks (ANNs). The framework is able to import a trained TensorFlow model and target a subset of its network layers, to provide them with custom operator implementations. Furthermore, the framework uses signal-to-quantization-noise ratio (SQNR) as a metric to identify potential layer implementations that are bottlenecks for prediction accuracy. With the use of the framework, we test various settings of operator implementations for the MobileNet V2 architecture. Specifically, we carry out experiments that benchmark operators that are well adapted for low memory usage and execution time, e.g. 8-bit quantization, but have a potential cost in prediction accuracy. With our results, we prove that this tool can be useful for industries where running ANNs on devices with limited hardware, like mobile phones, are of interest.

Link to presentation: https://lu-se.zoom.us/j/61795474317

Code for opponents: 17 FG

Link to popular science summary: https://fileadmin.cs.lth.se/cs/Education/Examensarbete/Popsci/200611/15HokLindstrom.pdf


16:00-17.00

Presenters: Erik Rosengren, Jonathan Strandberg
Title: Motion detection alarm verification using deep learning in surveillance systems
Examiner: Mathias Haage
Supervisors: Jörn Janneck (LTH), Jon Lindeheim (Axis Communications)

AXIS Companion is a cloud video management software which features configurable push notifications when something triggers a motion alarm. Motion alarms are however prone to false alarms which can be annoying for the end user. This thesis proposes using object detection based on deep convolutional neural networks to filter the alarms in order to lower the number of false alarms the end user receives. The object detection is run on a recording unit present in Axis surveillance systems. Several networks such as MobileNetV1 through MobileNetV3, YOLOv3 and InceptionNetV2 were compared against each other on manually annotated video with different lighting conditions to see which network performed best. Different optimizations were also compared to find the most optimal combination of networks and optimizations. The thesis found that MobileNetV3 was the most effective network, it produced around 3% false alarms and around 60% true alarms and did the best on night scenes.

Link to presentation: https://lu-se.zoom.us/s/68837513845

Code for opponents: 21 MHa

Link to popular science summary: https://fileadmin.cs.lth.se/cs/Education/Examensarbete/Popsci/200611/16RosengrenStrandberg.pdf


16:00-17.00

Presenter: Malte Kauranen
Title: Cross-lingual comment toxicity classification
Examiner: Jacek Malec
Supervisors: Pierre Nugues (LTH), Gustav Hjärn (Ifrågasätt Media AB)

The goal of the thesis is to explore the possibility for cross-lingual comment toxicity classification with a focus on newspaper comments. For this purpose a dataset from Ifrågasätt Media AB is used to investigate if these comments are different in nature to other types of texts online. For the evaluation of the cross-lingual task the OffensEval 2020 dataset is used. The results are that while zero-shot toxicity classification is possible, even small amounts of data in the target language augmented with more data from another language achieves results similar to having all the data in the target language. However, due to the differences in annotation standard and language in the Ifrågasätt Media AB dataset and the OffensEval 2020 dataset they are not useful to combine in cross-lingual training. This problem is potentially made larger due to unclear moderation guidelines or missing context in the Ifrågasätt Media AB dataset.

Link to presentation: https://lu-se.zoom.us/j/67969698809

Code for opponents: 5 JM


16:15-17.00

Presenters: Michaela Karlsson, Emmy Dahl
Title: Sharing is caring: Communicating recipient tailored OSS vulnerability information on an online platform
Examiner: Martin Höst
Supervisor: Martin Hell (LTH)

With IoT and digitisation in general comes initiatives for malicious actors to exploit possible vulnerabilities in the used product software. Both the usage of open source software and detected vulnerabilities have risen in the last decade and it is becoming increasingly important for companies to know what weaknesses they have in their software systems, to ensure safe products. The total cybersecurity of a product often depends on cooperation between several actors and sharing of sensitive information can improve the overall efficiency of each actor. This insinuates an industrial need for structured communication between organisations regarding software vulnerability management. The purpose of this thesis is thus to investigate how vulnerability information could be tailored to different recipients. More specifically how vulnerability information could be grouped and presented on an online platform, where different views could be accessed by the targeted recipients. The study is conducted as a qualitative case study.

Link to presentation: https://lu-se.zoom.us/j/66054321490

Code for opponents: 25 MHö

Link to popular science summary: https://fileadmin.cs.lth.se/cs/Education/Examensarbete/Popsci/200611/16KarlssonDahl.pdf