Hide menu

AIICS Publications: Student Theses

Hide abstracts BibTeX entries
2024
[117] Niklas Wretblad and Fredrik Gordh Riseby. 2024.
Bridging Language & Data: Optimizing Text-to-SQL Generation in Large Language Models.
Student Thesis. 82 pages.

This thesis explores text-to-SQL generation using Large Language Models within a financial context, aiming to assess the efficacy of current benchmarks and techniques. The central investigation revolves around the accuracy of the BIRD-Bench benchmark and the applicability of text-to-SQL models in real-world scenarios. The research explores why models are not showing significant performance improvements on this benchmark.The methodology adopted involved a thorough manual analysis of inputs and outputs from two distinct text-to-SQL models: a baseline zero-shot model and the DIN-SQL model, within the financial domain of the BIRD-Bench benchmark. The purpose was to identify and understand the limitations inherent in the dataset and the models themselves.Findings revealed that the best performing model on the original data was the DIN-SQL model, achieving an accuracy of 40.57%, a performance that raises questions due to its limited efficacy. Upon manual analysis, various types of noise were identified in the dataset, including string capitalization errors, faulty true SQL queries, grammatical errors, mismatches between questions and database schema, and language ambiguities. This led to the curation of two new datasets: one with cleaned questions and SQL queries, and another with only cleaned SQL queries, correcting a total of 52/106 data points.Subsequent runs of the models on these new datasets showed that data quality significantly impacts model performance. The completely cleaned dataset nearly eliminated the performance gap between the two models, with all models showing a 10%-17% increase in accuracy. Interestingly, on the dataset with only cleaned SQL queries, the performance of the models flipped with the basic zero-shot model now outperformed the DIN-SQL model.Further analysis of BIRD-Bench's development set across different domains indicated the presence of noise in other areas of the benchmark as well. This suggests that while BIRD-Bench closely resembles real-world scenarios, it falls short in offering a detailed understanding of model performance against different types of noise.The thesis introduces the concept of classifying noise in natural language questions, aiming to prevent the entry of noisy questions into text-to-SQL models and annotate noise in existing datasets. Experiments using GPT-3.5 and GPT-4 on a manually annotated dataset demonstrated the viability of this approach, with classifiers achieving up to 0.81 recall and 80% accuracy.Additionally, the thesis explored the use of LLMs for automatically correcting faulty SQL queries. This showed a 100\% success rate for specific query corrections, highlighting the potential for LLMs in improving dataset quality.The implications of these findings are substantial, emphasizing the need for noise-specific benchmarks and enhanced annotations in datasets like BIRD-Bench. This research underscores the importance of addressing specific noise challenges and developing more sophisticated text-to-SQL models.In conclusion, the thesis offers significant insights into the performance and limitations of text-to-SQL models, setting the stage for future research. This includes creating specialized datasets, enhancing annotations, focusing on identified noise types, developing user-input guardrails, and improving text-to-SQL models overall. Such advancements are expected to significantly improve the functionality and practical application of text-to-SQL technologies across various industries.

2023
[116] Anton Gefvert. 2023.
Text Curation for Clustering of Free-text Survey Responses.
Student Thesis. 54 pages. ISRN: LIU-IDA/LITH-EX-A--23/108—SE.

When issuing surveys, having the option for free-text answer fields is only feasible where the number of respondents is small, as the work to summarize the answers becomes unmanageable with a large number of responses. Using NLP techniques to cluster these answers and summarize them would allow a greater range of survey creators to incorporate free-text answers in their survey, without making their workload too large. Academic work in this domain is sparse, especially for smaller languages such as Swedish.The Swedish company iMatrics is regularly hired to do this kind of summarizing, specifically for workplace-related surveys. Their method of clustering has been semiautomatic, where both manual preprocessing and postprocessing have been necessary to accomplish this task.This thesis aims to explore if using more advanced, unsupervised NLP text representation methods, namely SentenceBERT and Sent2Vec, can improve upon these results and reduce the manual work needed for this task. Specifically, three questions are to be answered. Firstly, do the methods show good results? Secondly, can they remove the time-consuming postprocessing step of combining a large number of clusters into a smaller number? Lastly, can a model where unsupervised learning metrics can be shown to correlate to the real-world usability of the model, thus indicating that these metrics can be used to optimize the model for new data?To answer these questions, several models are trained, employed, and then compared using both internal and external metrics: Sent2Vec, SentenceBERT, and traditional baseline models. A manual evaluation procedure is performed to assess the real-world usability of the clusterings looks like, to see how well the models perform as well as to see if there is any correlation between this result and the internal metrics for the clustering.The results indicate that improving the text representation step is not sufficient for fully automating this task. Some of the models show promise in the results of human evaluation, but given the unsupervised nature of the problem and the large variance between models, it is difficult to predict the performance of new data. Thus, the models can serve as an improvement to the workflow, but the need for manual work remains.

[115] Hugo Ekinge. 2023.
How to Estimate Local Performance using Machine learning Engineering (HELP ME): from log files to support guidance.
Student Thesis. 41 pages. ISRN: LIU-IDA/LITH-EX-A--23/112--SE.

As modern systems are becoming increasingly complex, they are also becoming more and more cumbersome to diagnose and fix when things go wrong. One domain where it is very important for machinery and equipment to stay functional is in the world of medical IT, where technology is used to improve healthcare for people all over the world.This thesis aims to help with reducing downtime on critical life-saving equipment by implementing automatic analysis of system logs that without any domain experts involved can give an indication of the state that the system is in.First, a literature study was performed where three potential candidates of suitable neural network architectures was found. Next, the networks were implemented and a data pipeline for collecting and labeling training data was set up. After training the networks and testing them on a separate data set, the best performing model out of the three was based on GRU (Gated Recurrent Unit). Lastly, this model was tested on some real world system logs from two different sites, one without known issues and one with slow image import due to network issues.The results showed that it was feasible to build such a system that can give indications on external parameters such as network speed, latency and packet loss percentage using only raw system logs as input data. GRU, 1D-CNN (1-Dimensional Convolutional Neural Network) and Transformer's Encoder are the three models that were tested, and the best performing model was shown to produce correct patterns even on the real world system logs.

[114] Yohan Ayoub. 2023.
Multi-agent route planning for uncrewed aircraft systems operating in U-space airspace.
Student Thesis. 47 pages. ISRN: LIU-IDA/LITH-EX-A--23/104--SE.

Society today brings a high pace development and demand of Artificial intelligence systems as well as robotics. To further expand and to take one step closer to have Unmanned Aerial Vehicles (UAVs) working in the cities, the European Union Aviation Safety Agency launched a project that introduces U-space airspace, an airspace where UAVs, for instance, are allowed to operate for commercial services.The problems defined for U-space airspace resemble problems defined in the area of multi-agent path finding, such as scaling and traffic etc., resulting an interest to research whether MAPF-solutions can be applied to U-space scenarios. The following thesis extends the state-of-the-art MAPF-algorithm Continuous-time Conflict based search (CCBS) to handle simplified U-space scenarios, as well as extend other A*-based algorithms, such as a version of the Receding Horizon Lattice-based Motion Planning named Extended Multi-agent A* algorithm with Wait-Time (EMAWT) and an extended A* named Extended Multi-agent A* algorithm (EMA) to handle them. Comparisons of the three algorithms resulted in the EMAWT being the most reliable and stable solution throughout all tests, whilst for fewer agents, the CCBS being the clear best solution.

[113] Nisa Andersson. 2023.
Developing High level Behaviours for the Boston Dynamics Spot Using Automated Planning.
Student Thesis. 58 pages. ISRN: LIU-IDA/LITH-EX-A--23/049--SE.

Over the years, the Artificial Intelligence and Integrated Computer Systems (AIICS) Division at Linköping University has developed a high-level architecture for collaborative robotic systems that includes a delegation system capable of defining complex missions to be executed by a team of agents. This architecture has been used as a part of a research arena for developing and improving public safety and security using ground, aerial, surfaceand underwater robotic systems. Recently, the division decided to purchase a Boston Dynamics Spot robot to further progress into the public safety and security research area.The robot has a robotic arm and navigation functionalities such as map building, motion planning, and obstacle avoidance. This thesis investigates how the Boston Dynamics Spot robot can be integrated into the high-level architecture for collaborative robotic systems from the AIICS division. Additionally, how the robot’s functionalities can be extended so that it is capable of determining which actions it should take to achieve high-level behavioursconsidering its capabilities and current state. In this context, higher-level behaviours include picking up and delivering first aid supplies, which can be beneficial in specific emergency situations. The study was divided and done in an iterative approach.The system was tested in various scenarios that represent its intended future use. The result demonstrated the robot’s ability to plan and accomplish the desired high-level behaviours. However, there were instances when achieving the desired behaviours proved challenging due to various limiting factors, including limitations posed by the robot’s internal controller.

[112] Ludvig Widén and Emil Wiman. 2023.
Autonomous 3D exploration with dynamic obstacles: Towards Intelligent Navigation and Collision Avoidance for Autonomous 3D Exploration with dynamic obstacles.
Student Thesis. 99 pages. ISRN: LIU-IDA/LITH-EX-A--23/041--SE.

The advancements within robotics in recent years has increased the demand for sophisticated algorithms that can tackle the challenges associated with building robust and safe autonomous systems. The objective of 3D exploration is to enable a robot to explore an unknown environment with a high degree of accuracy while minimizing time and path length. Planners such as Receding Horizon Next Best View Planner (RH-NBVP) and Autonomous Exploration Planner (AEP) have been widely studied in static environments. However, in dynamic environments where obstacles like pedestrians or vehicles can appear, existing solutions either use dynamic motion planners for obstacle avoidance or simply use reactive behavior to avoid collisions like Dynamic Exploration Planner (DEP). This thesis examines how dynamic obstacles can be included in the planning process while performing 3D exploration to make more advantageous decisions regarding both efficiency and overall safety. Considering the uncertain dynamics in the environment is crucial for a robot to act safely, and hence be deployable in a real-world scenario. The suggested solution, Dynamic Autonomous Exploration Planner (DAEP), has been evaluated with other 3D-exploration planners as benchmarks. Experiments demonstrate that static planners struggle in dynamic environments. However, most collisions can be avoided with a simple reactive motion planner. This thesis presents an extension to the static planner AEP that considers dynamic obstacles. The proposed solution samples safer routes to a goal and takes into account historical observations of dynamic obstacles. Finally, a novel potential volumetric information gain is implemented to predict explored volume in the future. The results demonstrate that the extensions to AEP enhance safety planning and improve coverage compared to regular AEP and DEP.

[111] Robin Edlund and Johannes Kettu. 2023.
Performance of State Distributing Message-Oriented Middleware Systems Using Publish-Subscribe.
Student Thesis. 11 pages. ISRN: LIU-IDA/LITH-EX-G--23/039—SE.

Distributed simulations require efficient communication to represent complex scenarios, which presents a great challenge. This paper investigates the use of message-oriented middleware (MOM) to address this challenge by integrating the flight simulator X-Plane with the tactical simulator TACSI and evaluating the performance of different data transfer approaches. The study assesses performance by measuring the maximum sustainable throughput (MST) and the latency of a publish-subscribe-based MOM system. Two data distribution methods are compared: single-topic publishing and publishing to multiple subtopics. The results show that single-topic publishing achieves higher MST and lower latency when transmitting the same data volume. These findings provide valuable insights for deciding the state distribution method for publish-subscribe MOM systems. Additionally, this study highlights the limitations of manual determination of MST and underlines the need for accurate performance measurement techniques.

[110] Rachel Homssi and Jacob Möller. 2023.
Load Balancing In The Edge Cloud With Service Degradation: Combining Application-level Service Degradation With Load Balancing in a Kubernetes-based Edge Cloud.
Student Thesis. 79 pages. ISRN: LIU-IDA/LITH-EX-A--23/007--SE.

Edge cloud is a distributed computing architecture that is growing in popularity. It aims to bring the cloud closer to the edge of a network, reducing latency and improving performance through the use of distributed servers (edge nodes) spread out geographically. However, in the case of sudden increases in user requests, a node may run short of resources and need to implement a strategy that allows the node's service to degrade its service quality to a level that requires fewer resources so that the service can still be delivered. One such strategy is brownout, a control theory-based algorithm that dynamically controls the node's service quality in order to meet e.g., a latency goal. This thesis explores the use of brownout, previously used in combination with load balancing in the cloud, in conjunction with load balancing in an edge-cloud environment.In this thesis, four load-balancing strategies are evaluated in a Kubernetes-based edge-cloud environment, along with an application that implements the brownout feature. Two of the strategies are originally designed to be used with brownout but made for the regular cloud, one is a recently introduced strategy that performs well in the edge cloud but is brownout unaware, and the last is a random load balancer used as a baseline (also brownout unaware). The goal of the evaluation is to determine the efficiency of these strategies in different edge-cloud scenarios, with regard to service quality-weighted throughput, average latency, adherence to a set latency goal, and outsourcing (requests load balanced to another edge node). The results show that the first two strategies perform worse than the random load balancer in many regards. Their performance is also less predictable and tends to get worse with increasing network delays. The edge cloud strategy, however, shows an improvement in performance when the brownout is introduced in the majority of the test scenarios. Furthermore, the thesis introduced three possible modifications to make one of the cloud-based strategies perform better in the edge cloud. These modifications were tested in the same environment as the other load-balancing strategies and compared against each other. The first modification consisted of making the load-balancing logic treat its own node differently from other edge nodes. The second version was devised to only outsource when a certain resource threshold is exceeded and the last version was designed to prioritize its own node when below a certain resource threshold. The last version improved on the others and performed better than the base version in all measured metrics. Compared to the edge cloud strategy with brownout, it performed better with regard to service quality-weighted throughput but was outperformed in all other metrics.

2022
[109] Edvin Bergström. 2022.
Towards Combinatorial Assignment in a Euclidean Environmentwith many Agents: applied in StarCraft II.
Student Thesis. 35 pages. ISRN: LIU-IDA/LITH-EX-A--22/075--SE.

This thesis investigates coordinating units through simultaneous coalition structuregeneration and task assignment in a complex Euclidean environment. The environmentused is StarCraft II, and the problem modeled and solved in the game is the distribution ofcombat units over the game’s map. The map was split into regions, and every region wasmodeled as a task to which the combat units were assigned.In a number of experiments, we compare the performance of our approach with thegame’s built-in bots. Against most of the non cheating options, our agent wins 20% of thegames played on a large map, against the Hard built-in bot. On a smaller and simpler mapit wins 22% of games played against the hardest non-cheating difficulty.One of the main limitations of the method used to solve the assignment was the utility function. Which should describe the quality of a coalition and the task assignment.However, as the utility function described the state’s utility better, the win rate increased.Therefore the result indicates that the simultaneous coalition structure generation and taskassignment work for unit distribution in a complex environment like StarCraft II if a sufficient utility function is provided.

[108] Teodor Ganestćl. 2022.
Learning Multi-Agent Combat Scenarios in StarCraft II with League Training: an exploration of advanced learning techniques on a smaller scale.
Student Thesis. 25 pages. ISRN: LIU-IDA/LITH-EX-A--22/013--SE.

Google DeepMind trained their state-of-the-art StarCraft II agent AlphaStar using leaguetraining with massive computational power. In this thesis we explore league training onsmall-scale combat scenarios in StarCraft II, using limited computational resources, to an-swer whether this approach is suitable for smaller problems. We present two types ofagents: one trained using league training, and one trained against StarCraft II’s built-in AIusing traditional reinforcement learning. These agents are evaluated against each other, aswell as versus the built-in AI and human players. The results of these evaluations show thatthe agents trained with traditional reinforcement learning outperform the league-trainedagents, winning 520 out of 900 games played between them, with 21 ties. They also showbetter performance against the built-in AI and human players. The league agents still per-form well, with their skill being rated at the level of a Diamond or low Master player by ahuman Grandmaster ranked player. This shows that, while producing good results, leaguetraining does not reach the performance of the less computationally dependent traditionalreinforcement learning for problems of this scale. This might be due to the problem’s lowcomplexity in comparison to the full game of StarCraft II. A core aspect of league trainingis to find new strategies that target agents’ weaknesses in order to let agents learn to dealwith them, thus decreasing their exploitability. In smaller scenarios, there are not manydiverse strategies to apply, so this aspect cannot be fully utilised.

[107] Benjamin Danielsson. 2022.
Exploring Patient Classification Based on Medical Records: The case of implant bearing patients.
Student Thesis. 36 pages. ISRN: LIU-IDA/KOGVET-A--22/014--SE.

In this thesis, the application of transformer-based models on the real-world task of identifying patients as implant bearing is investigated. The task is approached as a classification task and five transformer-based models relying on the BERT architecture are implemented, along with a Support Vector Machine (SVM) as a baseline for comparison. The models are fine-tuned with Swedish medical texts, i.e. patients’ medical histories.The five transformer-based models in question makes use of two pre-trained BERT models, one released by the National Library of Sweden and a second one using the same pre-trained model but which has also been further pre-trained on domain specific language. These are in turn fine-tuned using five different types of architectures. These are: (1) a typical BERT model, (2) GAN-BERT, (3) RoBERT, (4) chunkBERT, (5) a frequency based optimized BERT. The final classifier, an SVM baseline, is trained using TF-IDF as the feature space.The data used in the thesis comes from a subset of an unreleased corpus from four Swedish clinics that cover a span of five years. The subset contains electronic medical records of patients belonging to the radiology, and cardiology clinics. Four training sets were created, respectively containing 100, 200, 300, and 903 labelled records. The test set, containing 300 labelled samples, was also created from said subset. The labels upon which the models are trained are created by labelling the patients as implant bearing based on the amount of implant terms each patient history contain.The results are promising, and show favourable performance when classifying the patient histories. Models trained on 903 and 300 samples are able to outperform the baseline, and at their peak, BERT, chunkBERT and the frequency based optimized BERT achieves an F1-measure of 0.97. When trained using 100 and 200 labelled records all of the transformerbased models are outperformed by the baseline, except for the semi-supervised GAN-BERT which is able to achieve competitive scores with 200 records.There is not a clear delineation between using the pre-trained BERT or the BERT model that has additional pre-training on domain specific language. However, it is believed that further research could shed additional light on the subject since the results are inconclusive.

[106] Oscar Bergman. 2022.
Generating fishing boats behaviour based on historic AIS data: A method to generate maritime trajectories based on historicpositional data.
Student Thesis. 54 pages. ISRN: LIU-IDA/LITH-EX-A--22/052--SE.

This thesis describes a method to generate new trajectories based on historic positiondata for a given geographical area. The thesis uses AIS-data from fishing boats to first describe a method that uses DBSCAN and OPTICS algorithms to cluster the data into clustersbased on routes where the boats travel and areas where the boats fish.Here bayesian optimization has been utilized to search for parameters for the clusteringalgorithms. In this given scenario it was shown DBSCAN is better in all fields, but it hasmany points where OPTICS has the potential to become better if it was modified a bit.This is followed by a method describing how to take the clusters and build a nodenetwork that then can be traversed using a path finding algorithm combined with internalrules to generate new routes that can be used in simulations to give a realistic enoughsituation picture. Finally a method to evaluate these generated routes are described andused to compare the routes to each other

[105] Oscar Amcoff. 2022.
I don?t know because I?m not a robot: I don?t know because I?m not a robot:A qualitative study exploring moral questions as a way to investigate the reasoning behind preschoolers? mental state attribution to robots.
Student Thesis. 37 pages. ISRN: LIU-IDA/KOGVET-G--22/026--SE.

Portrayals of artificially intelligent robots are becoming increasingly prevalent in children’s culture. This affects how children perceive robots, which have been found to affect the way children in school understand subjects like technology and programming. Since teachers need to know what influences their pupils' understanding of these subjects, we need to know how children’s preconceptions about robots affect the way they attribute mental states to them. We still know relatively little about how children do this. Based on the above, a qualitative approach was deemed fit. This study aimed to (1) investigate the reasoning and preconceptions underlying children’s mental state attribution to robots, and (2) explore the effectiveness of moral questions as a way to do this. 16 children aged 5- and 6 years old were asked to rate the mental states of four different robots while subsequently being asked to explain their answers. Half of the children were interviewed alone and half in small groups. A thematic analysis was conducted to analyze the qualitative data. Children’s mental state attribution was found to be influenced by preconceptions about robots as a group of entities lacking mental states. Children were found to perceive two robots, Atlas, and Nao, differently in various respects. This was argued to be because the children perceived these robots through archetypal frameworks. Moral questions were found successful as a way to spark reflective reasoning about the mental state attribution in the children.

[104] Ulrika Ćkerström. 2022.
Lekmannabedömning av ett självkörande fordons körförmćga: betydelsen av att erfara fordonet i trafiken.
Student Thesis. 31 pages. ISRN: LIU-IDA/KOGVET-G--22/025--SE.

Datorstyrda maskiner som bÄde kan styra sina egna aktiviteter och som har ett stort rörelseomfÄng kommer snart att dela vÄr fysiska miljö vilket kommer innebÀra en drastisk förÀndring för vÄr nuvarande mÀnskliga kontext. Tidigare olyckor som skett mellan mÀnskliga förare och automatiserade fordon kan förklaras genom en bristande förstÄelse för de automatiserade fordonets beteende. Det Àr dÀrför viktigt att ta reda pÄ hur mÀnniskor förstÄr automatiserade fordons förmÄgor och begrÀnsningar. SAE International, en global yrkeskÄr fÄr ingenjörer verksamma inom fordonsindustrin, har definierat ett ramverk som beskriver funktionaliteten hos automatiserade fordon i 6 olika nivÄer. Den rapporterade studien undersökte med utgÄngspunkt i detta ramverk vilken automationsgrad deltagarna antar att en sjÀlvkörande buss har genom deltagarnas upplevelse av fordonet. Inom ramarna för studien fÀrdades deltagarna en kort strÀcka pÄ en sjÀlvkörande buss och besvarade en enkÀt om hur de ser pÄ bussens förmÄgor och begrÀnsningar bÄde före och efter fÀrden. Studieresultatet visade att hÀlften av deltagarna överskattade bussens automationsgrad. Efter att ha fÀrdats med bussen justerade deltagarna ner sina förvÀntningar pÄ fordonets körförmÄga vilket stÀmde bÀttre överens med bussens förmÄgor och begrÀnsningar. Deltagarna rapporterade Àven att de var mer sÀkra i sina bedömningar efter erfarenhet av fordonet. Sammanfattningsvis tyder resultatet pÄ att (1) mÀnniskor tenderar att överskatta automatiserade fordons körförmÄga, men att (2) deras uppfattning justeras i samband med att de kommer i kontakt med det automatiserade fordonet i verkligheten och att (3) de dÄ Àven blir mer sÀkra i sina bedömningar. Detta borde tas i beaktning vid utveckling av sjÀlvkörande fordon för att minska risken för olyckor i trafiken.

[103] Erik Lundin. 2022.
Generating Directed & Weighted Synthetic Graphs using Low-Rank Approximations.
Student Thesis. 65 pages. ISRN: LIU-IDA/LITH-EX-A--2022/042--SE.

Generative models for creating realistic synthetic graphs constitute a research area that is increasing in popularity, especially as the use of graph data is becoming increasingly common. Generating realistic synthetic graphs enables sharing of the information embedded in graphs without directly sharing the original graphs themselves. This can in turn contribute to an increase of knowledge within several domains where access to data is normally restricted, including the financial system and social networks. In this study, it is examined how existing generative models can be extended to be compatible with directed and weighted graphs, without limiting the models to generating graphs of a specific domain. Several models are evaluated, and all use low-rank approximations to learn structural properties of directed graphs. Additionally, it is evaluated how node embeddings can be used with a regression model to add realistic edge weights to directed graphs.The results show that the evaluated methods are capable of reproducing global statistics from the original directed graphs to a promising degree, without having more than 52% overlap in terms of edges. The results also indicate that realistic directed and weighted graphs can be generated from directed graphs by predicting edge weights using pairs of node embeddings. However, the results vary depending on which node embedding technique is used.

[102] Emil Brćkenhielm and Kastrati Drinas. 2022.
Thermal Imaging-Based Instance Segmentation for Automated Health Monitoring of Steel Ladle Refractory Lining.
Student Thesis. 45 pages. ISRN: LIU-IDA/LITH-EX-A--22/039--SE.

Equipment and machines can be exposed to very high temperatures in the steel mill industry. One particularly critical part is the ladles used to hold and pour molten iron into mouldings. A refractory lining is used as an insulation layer between the outer steel shell and the molten iron to protect the ladle from the hot iron. Over time, or if the lining is not completely cured, the lining wears out or can potentially fail. Such a scenario can lead to a breakout of molten iron, which can cause damage to equipment and, in the worst case, workers. Previous work analyses how critical areas can be identified in a proactive matter. Using thermal imaging, the failing spots on the lining could show as high-temperature areas on the outside steel shell. The idea is that the outside temperature corresponds to the thickness of the insulating lining. The detection of these spots is identified when temperatures over a given threshold are registered within the thermal camera's field of view. The images must then be manually analyzed over time, to follow the progression of a detected spot. The existing solution is also prone to the background noise of other hot objects. This thesis proposes an initial step to automate monitoring the health of refractory lining in steel ladles. The report will investigate the usage of Instance Segmentation to isolate the ladle from its background. Thus, reducing false alarms and background noise in an autonomous monitoring setup. The model training is based on Mask R-CNN on our own thermal images, with pre-trained weights from visual images. Detection is done on two classes: open or closed ladle. The model proved reasonably successful on a small dataset of 1000 thermal images. Different models were trained with and without augmentation, pre-trained weights as well multi-phase fine-tuning. The highest mAP of 87.5\% was achieved on a pre-trained model with image augmentation without fine-tuning. Though it was not tested in production, temperature readings could lastly be extracted on the segmented ladle, decreasing the risk of false alarms from background noise.

[101] Evelina Holmgren and Simon Wijk Stranius. 2022.
A Multi-Agent Pickup and Delivery System for Automated Stores with Batched Tasks.
Student Thesis. 101 pages. ISRN: LIU-IDA/LITH-EX-A--22/038--SE.

Throughout today’s society, increasingly more areas are being automated. Grocery stores however have been the same for years. Only recently, self-checkout counters and online shopping have been utilised in this business area. This thesis aims to take it to the next step by introducing automated grocery stores using a multi-agent system. Orders will be given to the system, and on a small area, multiple agents will pick the products in a time-efficient way and deliver them to the customer. This can both increase the throughput but also decrease the food waste and energy consumption of grocery stores. This thesis investigates already existing solutions for the multi-agent pickup and delivery problem. It extends these to the important case of batched tasks in order to improve the customer experience. Batches of tasks represent shopping carts, where fast completion of whole batches gives greater customer satisfaction. This notion is not mentioned in related work, where completion of single tasks is the main goal. Because of this, the existing solution does not accommodate the need of batches or the importance of completing whole batches fast and in somewhat linear order. For this purpose, a new metric called batch ordering weighted error (BOWE) was created that takes these factors into consideration. Using BOWE, one existing algorithm has been extended into prioritizing completing whole batches and is now called B-PIBT. This new algorithm has significantly improved BOWE and even batch service time for the algorithm in key cases and is now superior in comparison to the other state-of-the-art algorithms.

[100] Aksel Holmgren. 2022.
Out of sight, out of mind?: Assessing human attribution of object permanence capabilities to self-driving cars.
Student Thesis. 34 pages. ISRN: LIU-IDA/KOGVET-G--22/016--SE.

Autonomous vehicles are regularly predicted to be on the verge of broad integration into regular traffic. A crucial aspect of successful traffic interactions is one agent’s ability to adequately understand other agents’ capabilities and limitations. Within the current state of the art concerning self-driving cars, there is a discrepancy between what people tend to believe the capabilities of self-driving cars are, and what those capabilities actually are. The aim of this study was to investigate whether people attribute the capacity of object permanence to self-driving cars roughly in the same manner as they would to a human driver. The study was conducted with online participants (N = 105).The results showed that the participants did not attribute object permanence differently between a self-driven car and a human driver. This indicates that people attribute object permanence similarly to self-driving cars as they do toward human drivers. Furthermore, the results indicate no connection between participants’ tendency to anthropomorphize and whether they attributed object permanence or not. The findings provide evidence for the issues connected to the perceptual belief problem in human-robot interaction, where people attribute capabilities to autonomous vehicles that are not there. The results highlight the importance of understanding which mechanisms underlie these attributions as well as when they happen, in order to mitigate unrealistic expectations.

[99] David Ćngström. 2022.
Genetic Algorithms for optimizing behavior trees in air combat.
Student Thesis. 40 pages.

Modelling and simulating entities in virtual environments are tools commonly used by companies to test, validate and verify their products in close to real scenarios; effectivelyreducing the cost, time and effort compared to real life testing. This is especially the case in the area of air combat where realistic behaviors are not only a necessity, but paramount to replace the costs of fuel and operation time. The behavior tree framework is a behavior model whichrepresents entity actions with regards to its perception of the world whilst being easy to manuallyvalidate through its intuitively structured nature. However, as different simulated scenarios require different behaviors, operators commonly has to manually craft new behavior trees at the cost of time and effort.In this thesis, the AI technique Genetic Algorithms (GA) is used to improve a previously crafted general behavior tree with regards to a given 4v4 beyond-visual-range air combat scenario. To this end, a select number of parameters within the behavior tree are optimized in two experiments where a) all parameters are optimized globally and b) the parameters are divided into blocks of sub-behaviors (Engage, Fire missile, etc.) which are then optimized individuallyand combined at a later stage. The agents in the GA are put against the base tree where the baseline is referred to as the base tree vs itself. As the problem proved too easy and resulted in an over-optimized behavior when a single scenario was used, the decision was made to increase the number of the scenarios to three; differing in positions and orientations. The former experimentresulted in a behavior capable of defeating all entities in the other team without any casualties in all three scenarios while the behavior in the latter experiment failed to find the cross-blockrelations, and thus, only achieved a slightly better result than that of the baseline. However, the parameters of highest importance are found to be highly correlated in both experiments and GA is concluded to be a satisfactory technique for the problem of generating improved behaviors with regards to given scenarios.

[98] Sindre Jonsson Wold. 2022.
Evaluation of different runner set-ups for CI/CD pipelines.
Student Thesis. 98 pages. ISRN: LIU-IDA/LITH-EX-A--22/030--SE.

DevOps and continuous practices are increasingly popular development practices aiming at bridging the gap between software development and IT operations with the indented outcome of shorter development life cycles while maintaining a high software quality. A fundamental part of many DevOps systems is a CI/CD (continuous integration/deployment) pipeline allowing for automatic building, testing and deployment of software. The use of continuous practices have been shown to achieve the desired outcomes, whereas the adopting of such practices has been attributed with the challenges of lacking expertise and skill as well as lacking available tools and technology.Execution of commands in a CI/CD pipeline are handled by a runner application, which can be configured in different ways allowing for different levels of the quality attributes performance, response time, throughput, robustness, stability, resource constraints, cost and maintainability. Five different types of runner infrastructure were implemented and evaluated on the quality attributes. These were: one single-machine implementation, one serverless implementation and three autoscaling implementations.For robustness and stability autoscaling implementations achieved the best results. Performance and throughput were affected by resource constraints which in turn affected the cost. Similar results were found for response time for all but one of the three autoscaling implementations, and for the serverless implementation. Finally, all implementations had similar results for reliability.

[97] William Bergekrans. 2022.
Automatic Man Overboard Detection with an RGB Camera: Using convolutional neural networks.
Student Thesis. 50 pages. ISRN: LIU-IDA/LITH-EX-A--22/036--SE.

Man overboard is one of the most common and dangerous accidents that can occur whentraveling on a boat. Available research on man overboard systems with cameras have focusedon man overboard taking place from larger ships, which involves a fall from a height.Recreational boat manufacturers often use cord-based kill switches that turns of the engineif the wearer falls overboard. The aim of this thesis is to create a man overboard warningsystem based on state-of-the-art object detection models that can detect man overboard situationthrough inputs from a camera. Awell performing warning system would allow boatmanufactures to comply with safety regulations and expand the kill-switch coverage to allpassengers on the boat. Furthermore, the aim is also to create two new datasets: one dedicatedto human detection and one with man overboard fall sequences. YOLOv5 achievedthe highest performance on a new human detection dataset, with an average precision of97%. A Mobilenet-SSD-v1 network based on weights from training on the PASCAL VOCdataset and additional training on the new man overboard dataset is used as the detectionmodel in final warning system. The man overboard warning system achieves an accuracyof 50% at best, with a precision of 58% and recall of 78%.

[96] Christoffer Sjöbergsson. 2022.
Comparison of Distance Metrics for Trace Clustering in Process Mining: An Effort to Simplify Analysis of Usage Patterns in PACS.
Student Thesis. 43 pages. ISRN: LIU-IDA/LITH-EX-A--2022/003--SE.

This study intended to validate if clustering could be used to simplify models generated with process mining. The intention was also to see if these clusters could suggest anything about user efficiency. To that end a new metric where devised, average mean duration deviation. This metric aimed to show if a trace was more or less efficient than a comparative trace. Since the intent was to find traces with similar characteristics the clustering was done with characteristic features instead of time efficiency features. The aim was to find a correlation between efficiency after the fact. A correlation with efficiency could not be found.

[95] Oskar Skoglund. 2022.
Finding co-workers with similar competencies through data clustering.
Student Thesis. 40 pages. ISRN: LIU-IDA/LITH-EX-A--21/081--SE.

In this thesis, data clustering techniques are applied to a competence database from the company Combitech. The goal of the clustering is to connect co-workers with similar competencies and competence areas in order to enable more skill sharing. This is accomplished by implementing and evaluating three clustering algorithms, k-modes, DBSCAN, and ROCK. The clustering algorithms are fine-tuned with the use of three internal validity indices, the Dunn, Silhouette, and Davies-Bouldin score. Finally, a form regarding the clustering of the three algorithms is sent out to the co-workers, which the clustering is based on, in order to obtain external validation by calculating the clustering accuracy. The results from the internal validity indices show that ROCK and DBSCAN create the most separated and dense clusters. The results from the form show that ROCK is the most accurate of the three algorithms, with an accuracy of 94%, followed by k-modes at 58% and DBSCAN at 40% accuracy. However, the visualization of the clusters shows that both ROCK and DBSCAN create one very big cluster, which is not desirable. This was not the case for k-modes, where the clusters are more evenly sized while still being fairly well-separated. In general, the results show that it is possible to use data clustering techniques to connect people with similar competencies and that the predicted clusters agree fairly well with the gold-standard data from the co-workers. However, the results are very dependent on the choice of algorithm and parametric values, and thus have to be chosen carefully.

[94] Alfred Hagberg and Mustaf Abdullahi Musse. 2022.
Instance Segmentation on depth images using Swin Transformer for improved accuracy on indoor images.
Student Thesis. In series: arXiv.org #??. 43 pages. ISRN: LIU-IDA/LITH-EX-A--2022/004--SE.

The Simultaneous Localisation And Mapping (SLAM) problem is an open fundamental problem in autonomous mobile robotics. One of the latest most researched techniques used to enhance the SLAM methods is instance segmentation. In this thesis, we implement an instance segmentation system using Swin Transformer combined with two of the state of the art methods of instance segmentation namely Cascade Mask RCNN and Mask RCNN. Instance segmentation is a technique that simultaneously solves the problem of object detection and semantic segmentation. We show that depth information enhances the average precision (AP) by approximately 7%. We also show that the Swin Transformer backbone model can work well with depth images. Our results also show that Cascade Mask RCNN outperforms Mask RCNN. However, the results are to be considered due to the small size of the NYU-depth v2 dataset. Most of the instance segmentation researches use the COCO dataset which has a hundred times more images than the NYU-depth v2 dataset but it does not have the depth information of the image.

[93] Anton Hansson and Hugo Cedervall. 2022.
Insurance Fraud Detection using Unsupervised Sequential Anomaly Detection.
Student Thesis. 63 pages. ISRN: LIU-IDA/LITH-EX-A--21/084--SE.

Note: Gjordes digitalt via Zoom. 

Fraud is a common crime within the insurance industry, and insurance companies want to quickly identify fraudulent claimants as they often result in higher premiums for honest customers. Due to the digital transformation where the sheer volume and complexity of available data has grown, manual fraud detection is no longer suitable. This work aims to automate the detection of fraudulent claimants and gain practical insights into fraudulent behavior using unsupervised anomaly detection, which, compared to supervised methods, allows for a more cost-efficient and practical application in the insurance industry. To obtain interpretable results and benefit from the temporal dependencies in human behavior, we propose two variations of LSTM based autoencoders to classify sequences of insurance claims. Autoencoders can provide feature importances that give insight into the models' predictions, which is essential when models are put to practice. This approach relies on the assumption that outliers in the data are fraudulent. The models were trained and evaluated on a dataset we engineered using data from a Swedish insurance company, where the few labeled frauds that existed were solely used for validation and testing. Experimental results show state-of-the-art performance, and further evaluation shows that the combination of autoencoders and LSTMs are efficient but have similar performance to the employed baselines. This thesis provides an entry point for interested practitioners to learn key aspects of anomaly detection within fraud detection by thoroughly discussing the subject at hand and the details of our work.

2021
[92] David Grönberg. 2021.
Extracting Salient Named Entities from Financial News Articles.
Student Thesis. ISRN: LIU-IDA/LITH-EX-A--21/040--SE.

This thesis explores approaches for extracting company mentions from financial newsarticles that carry a central role in the news. The thesis introduces the task of salient named entity extraction (SNEE): extract all salient named entity mentions in a text document. Moreover, a neural sequence labeling approach is explored to address the SNEE task in an end-to-end fashion, both using a single-task and a multi-task learning setup. In order to train the models, a new procedure for automatically creating SNEE annotations for an existing news article corpus is explored. The neural sequence labeling approaches are compared against a two-stage approach utilizing NLP parsers, a knowledge base and a salience classifier. Textual features inspired from related work in salient entity detection are evaluated to determine what combination of features results in the highest performance on the SNEE task when used by a salience classifier. The experiments show that the difference in performance between the two-stage approach and the best performing sequence labeling approach is marginal, demonstrating the potential of the end-to-end sequence labeling approach on the SNEE task.

[91] Axel Wickman. 2021.
Exploring feasibility of reinforcement learning flight route planning.
Student Thesis. 36 pages. ISRN: LIU-IDA/KOGVET-G–21/031—SE.

Alternativ nerladdning: https://public.axelwickman.com/published...

This thesis explores and compares traditional and reinforcement learning (RL) methods of performing 2D flight path planning in 3D space. A wide overview of natural, classic, and learning approaches to planning s done in conjunction with a review of some general recurring problems and tradeoffs that appear within planning. This general background then serves as a basis for motivating different possible solutions for this specific problem. These solutions are implemented, together with a testbed inform of a parallelizable simulation environment. This environment makes use of random world generation and physics combined with an aerodynamical model. An A* planner, a local RL planner, and a global RL planner are developed and compared against each other in terms of performance, speed, and general behavior. An autopilot model is also trained and used both to measure flight feasibility and to constrain the planners to followable paths. All planners were partially successful, with the global planner exhibiting the highest overall performance. The RL planners were also found to be more reliable in terms of both speed and followability because of their ability to leave difficult decisions to the autopilot. From this it is concluded that machine learning in general, and reinforcement learning in particular, is a promising future avenue for solving the problem of flight route planning in dangerous environments.

[90] Adam Lager. 2021.
Improving Solr search with Natural Language Processing: An NLP implementation for information retrieval in Solr.
Student Thesis. 24 pages. ISRN: LIU-IDA/LITH-EX-G–21/030—SE.

The field of AI is emerging fast and institutions and companies are pushing the limits of impossibility. Natural Language Processing is a branch of AI where the goal is to understand human speech and/or text. This technology is used to improve an inverted index,the full text search engine Solr. Solr is open source and has integrated OpenNLP makingit a suitable choice for these kinds of operations. NLP-enabled Solr showed great results compared to the Solr that’s currently running on the systems, where NLP-Solr was slightly worse in terms of precision, it excelled at recall and returning the correct documents.

[89] Oskar Hidén and David Björelind. 2021.
Clustering and Summarization of Chat Dialogues: To understand a company?s customer base.
Student Thesis. 61 pages. ISRN: LIU-IDA/LITH-EX-A--21/037--SE.

The Customer Success department at Visma handles about 200 000 customer chats each year, the chat dialogues are stored and contain both questions and answers. In order to get an idea of what customers ask about, the Customer Success department has to read a random sample of the chat dialogues manually. This thesis develops and investigates an analysis tool for the chat data, using the approach of clustering and summarization. The approach aims to decrease the time spent and increase the quality of the analysis. Models for clustering (K-means, DBSCAN and HDBSCAN) and extractive summarization (K-means, LSA and TextRank) are compared. Each algorithm is combined with three different text representations (TFIDF, S-BERT and FastText) to create models for evaluation. These models are evaluated against a test set, created for the purpose of this thesis. Silhouette Index and Adjusted Rand Index are used to evaluate the clustering models. ROUGE measure together with a qualitative evaluation are used to evaluate the extractive summarization models. In addition to this, the best clustering model is further evaluated to understand how different data sizes impact performance. TFIDF Unigram together with HDBSCAN or K-means obtained the best results for clustering, whereas FastText together with TextRank obtained the best results for extractive summarization. This thesis applies known models on a textual domain of customer chat dialogues, something that, to our knowledge, has previously not been done in literature.

[88] Agaton Sjöberg. 2021.
Extracting Transaction Information from Financial Press Releases.
Student Thesis. 38 pages. ISRN: LIU-IDA/LITH-EX-A--21/039--SE.

DOI: 21/039.

The use cases of Information Extraction (IE) are more or less endless, often consisting of a combination of Named Entity Recognition (NER) and Relation Extraction (RE). One use case of IE is the extraction of transaction information from Norwegian insider transaction Press Releases (PRs), where a transaction consists of at most four entities: the name of the owner performing the transaction, the number of shares transferred, the transaction date, and the price of the shares bought or sold. The relationships between the entities define which entity belongs to which transaction, and whether shares were bought or sold. This report has investigated how a pair of supervised NER and RE models extract this information. Since these Norwegian PRs were not labeled, two different approaches to annotating the transaction entities and their associated relations were investigated, and it was found that it is better to annotate only entities that occur in a relation than annotating all occurrences. Furthermore, the number of PRs needed to achieve a satisfactory result in the IE pipeline was investigated. The study shows that training with about 400 PRs is sufficient for the results to converge, at around 0.85 in F1-score. Finally, the report shows that there is not much difference between a complex RE model and a simple rule-based approach, when applied on the studied corpus.

[87] Felix Nodelijk and Arun Uppugunduri. 2021.
Estimating lighting from unconstrained RGB images using Deep Learning in real-time for superimposed objects in an augmented reality application.
Student Thesis. 62 pages. ISRN: LIU-IDA/LITH-EX-A--21/042—SE.

Modern deep learning enables many new possibilities for automation. Within augmented reality, deep learning can be used to infer the lighting to accurately render superimposed objects with correct lighting to mix seamlessly with the environment. This study aims to find a method of light estimation from RGB images by investigating Spherical Harmonic coefficients and how said coefficients could be inferred for use in an AR application in real-time. The pre-existing method employed by the application estimates the light by comparing two points cheek-to-cheek on a face. This fails to accurately represent the lighting in many situations, causing users to stop using the application. This study investigates a deep learning model that shows significant improvements in regards to the lighting estimation while also achieving fast inference time. The model results were presented to respondents in a survey and was found to be the better method of the two in terms of light estimation. The final model achieved 19 ms in inference time and 0.10 in RMS error.

[86] Daniel Roos. 2021.
Evaluation of BERT-like models for small scale ad-hoc information retrieval.
Student Thesis. 34 pages. ISRN: LIU-IDA/LITH-EX-A--21/051—SE.

Measuring semantic similarity between two sentences is an ongoing research field with big leaps being taken every year. This thesis looks at using modern methods of semantic similarity measurement for an ad-hoc information retrieval (IR) system. The main challenge tackled was answering the question \"What happens when you don’t have situation-specific data?\". Using encoder-based transformer architectures pioneered by Devlin et al., which excel at fine-tuning to situationally specific domains, this thesis shows just how well the presented methodology can work and makes recommendations for future attempts at similar domain-specific tasks. It also shows an example of how a web application can be created to make use of these fast-learning architectures.

[85] Philip Palapelas Kantola. 2021.
Extreme Quantile Estimation of Downlink Radio Channel Quality.
Student Thesis. 46 pages. ISRN: LIU-IDA/LITH-EX-A--21/048--SE.

The application area of Fifth Generation New Radio (5G-NR) called Ultra-Reliable and Low-Latency Communication (URLLC) requires a reliability, the probability of receiving and decoding a data packet correctly, of 1 - 10^5. For this requirement to be fulfilled in a resource-efficient manner, it is necessary to have a good estimation of extremely low quan- tiles of the channel quality distribution, so that appropriate resources can be distributed to users of the network system. This study proposes and evaluates two methods for estimating extreme quantiles of the downlink channel quality distribution, linear quantile regression and Quantile Regression Neural Network (QRNN). The models were trained on data from Ericsson’s system-level radio network simulator, and evaluated on goodness of fit and resourcefulness. The focus of this study was to estimate the quantiles 10^2, 10^3 and 10^4 of the distribution. The results show that QRNN generally performs better than linear quantile regression in terms of pseudoR2, which indicates goodness of fit, when the sample size is larger. How- ever, linear quantile regression was more effective for smaller sample sizes. Both models showed difficulty estimating the most extreme quantiles. The less extreme quantile to esti- mate, the better was the resulting pseudoR2-score. For the largest sample size, the resulting pseudoR2-scores of the QRNN was 0.20, 0.12 and 0.07, and the scores of linear quantile regression was 0.16, 0.10 and 0.07 for the respective quantiles 10^2, 10^3 and 10^4. It was shown that both evaluated models were significantly more resourceful than us- ing the average of the 50 last measures of channel quality subtracted with a fixed back-off value as a predictor. QRNN had the most optimistic predictions. If using the QRNN, theo- retically, on average 43% more data could be transmitted while fulfilling the same reliability requirement than by using the fixed back-off value.

[84] Axel Holmberg and Wilhelm Hansson. 2021.
Kombinatorisk Optimering med Pointer Networks och Reinforcement Learning.
Student Thesis. ISRN: LIU-IDA/LITH-EX-A--21/035—SE.

Given the complexity and range of combinatorial optimization problems, solving them can be computationally easy or hard. There are many ways to solve them, but all available methods share a problem: they take a long time to run and have to be rerun when new cases are introduced. Machine learning could prove a viable solution to solving combinatorial optimization problems due to the possibility for models to learn and generalize, eliminating the need to run a complex algorithm every time a new instance is presented. Uniter is a management consulting firm that provides services within product modularization. Product modularization results in the possibility for many different product variations to be created based on customer needs. Finding the best combination given a specific customer's need will require solving a combinatorial optimization problem. Based on Uniter's need, this thesis sought to develop and evaluate a machine learning model consisting of a Pointer Network architecture and trained using Reinforcement Learning. The task was to find the combination of parts yielding the lowest cost, given a use case. Each use case had different attributes that specified the need for the final product. For each use case, the model was tasked with selecting the most appropriate combination from a set of 4000 distinct combinations. Three experiments were conducted: examining if the model could suggest an optimal solution after being trained on one use case, if the model could suggest an optimal solution of a previously seen use case, and if the model could suggest an optimal solution of an unseen use case. For all experiments, a single data set was used. The suggested model was compared to three baselines: a weighted random selection, a naive model implementing a feed-forward network, and an exhaustive search.The results showed that the proposed model could not suggest an optimal solution in any of the experiments. In most tests conducted, the proposed model was significantly slower at suggesting a solution than any baseline. The proposed model had high accuracy in all experiments, meaning it suggested almost only feasible solutions in the allowed solution space. However, when the model converged, it suggested only one combination for every use case, with the feed-forward baseline showing the same behavior. This behavior could suggest that the model misinterpreted the task and identified a solution that would work in most cases instead of suggesting the optimal solution for each use case. The discussion concludes that an exhaustive search is preferable for the studied data set and that an alternate approach using supervised learning may be a better solution.

[83] David Nyberg. 2021.
Exploring the Capabilities of Generative Adversarial Networks in Remote Sensing Applications.
Student Thesis. ISRN: LIU-IDA/LITH-EX-A--2021/043--SE.

attachment: http://liu.diva-portal.org/smash/get/div...

The field of remote sensing uses imagery captured from satellites, aircrafts, and UAVs in order to observe and analyze the Earth. Many remote sensing applications that are used today employ deep learning models that require large amounts of data or specific types of data. The lack of data can hinder model performance. A generative adversarial network (GAN) is a deep learning model that can generate synthetic data and can be used as a method for data augmentation to increase performance of data reliant deep learning models. GANs are also capable of image-to-image translation such as transforming a satellite image containing cloud coverage into one without clouds. These possibilities have led to many new and exciting GAN applications.This thesis explores ways generative adversarial networks (GANs) can be applied in a variety of remote sensing applications. To evaluate this, four experiments using GANs are implemented. The tasks are: generating synthetic aerial forestry imagery, translating a satellite segmentation mask into a real satellite image, removal of thin cloud cover from a satellite image, and super resolution to increase the resolution of a satellite image. In all experiments the tasks were deemed successful and prove the potential for further use of GANs in the field of remote sensing.

[82] Lukas Borggren. 2021.
Automatic Categorization of News Articles With Contextualized Language Models.
Student Thesis. 63 pages. ISRN: LIU-IDA/LITH-EX-A--21/038--SE.

This thesis investigates how pre-trained contextualized language models can be adapted for multi-label text classification of Swedish news articles. Various classifiers are built on pre-trained BERT and ELECTRA models, exploring global and local classifier approaches. Furthermore, the effects of domain specialization, using additional metadata features and model compression are investigated. Several hundred thousand news articles are gathered to create unlabeled and labeled datasets for pre-training and fine-tuning, respectively. The findings show that a local classifier approach is superior to a global classifier approach and that BERT outperforms ELECTRA significantly. Notably, a baseline classifier built on SVMs yields competitive performance. The effect of further in-domain pre-training varies; ELECTRA’s performance improves while BERT’s is largely unaffected. It is found that utilizing metadata features in combination with text representations improves performance. Both BERT and ELECTRA exhibit robustness to quantization and pruning, allowing model sizes to be cut in half without any performance loss.

[81] Andreas Magnvall and Alexander Henne. 2021.
Real-time Aerial Photograph Alignment using Feature Matching.
Student Thesis. 21 pages. ISRN: LIU-IDA/LITH-EX-G--21/035--SE.

With increased mobile hardware capabilities, improved UAVs and modern algorithms, accurate maps can be created in real-time by capturing overlapping photographs of the ground. A method for mapping that can be used is to position photos by relying purely on the GPS position and altitude. However, GPS inaccuracies will be visible in the created map. In this paper, we will instead present a method for aligning the photos correctly with the help of feature matching. Feature matching is a well-known method which analyses two photos to find similar parts. If an overlap exists, feature matching can be used to find and localise those parts, which can be used for positioning one image over the other at the overlap. When repeating the process, a whole map can be created. For this purpose, we have also evaluated a selection of feature detection and matching algorithms. The algorithm found to be the best was SIFT with FLANN, which was then used in a prototype for creating a complete map of a forest. Feature matching is in many cases superior to GPS positioning, although it cannot be fully depended on as failed or incorrect matching is a common occurrence.

[80] Agnes Hallberg. 2021.
Using Low-Code Platforms to Collect Patient-Generated Health Data: A Software Developer?s Perspective.
Student Thesis. ISRN: LIU-IDA/LITH-EX-G--21/034--SE.

The act of people collecting their health data through health apps on their smartphones is becoming increasingly popular. Still, it is difficult for healthcare providers to use this patient-generated health data since health apps cannot easily share its data with the health care providers’ Electronic Health Records (EHR). Simultaneously, it is becoming increasingly popular to use low-code platforms for software development. This thesis explored using low-code platforms to create applications intended to collect patient-generated health data and send it to EHRs by creating a web application prototype with the low-code platforms Mendix and Better EHR Studio. During the web application prototype development, the developer conducted a diary to capture their impressions of Mendix to show how a developer experiences developing in a low-code platform compared to traditional programming. The result shows that it is impractical to create applications intended to collect patient-generated health data with the two low-code platforms chosen. The analysis of the conducted diary showed that using a low-code platform is straightforward but also challenging for an experienced software developer.

[79] Anna Lindqvist. 2021.
Threats to smart buildings: Securing devices in a SCADA network.
Student Thesis. ISRN: LIU-IDA/LITH-EX-G--21/038--SE.

This paper examines the possibilities of performing tests with the aim to ensure that devices in a SCADA network can be deemed secure before deployment. SCADA systems are found in most industries and have recently seen an increased use in building automation, most importantly the healthcare sector, which means that a successful attack toward such a system could endanger lives of patients and healthcare professionals.The method of testing was created to examine whether devices conflicted with the security flaws identified by OWASP IoT Top 10 list, meaning that OWASP IoT Top 10 was the foundation for the methodology used in this paper.Results of the tests show that the devices used in testing are not in conflict with the OWASP IoT Top 10 list when using the default settings. However, some settings that can be enabled on the devices would constitute a security risk if enabled.

[78] Patrick Lundberg. 2021.
Verktyg för hyperparameteroptimering.
Student Thesis. 8 pages. ISRN: LIU-IDA/LITH-EX-G--21/042--SE.

Hyperparameteroptimering Àr ett viktigt uppdrag för att effektivt kunna anvÀnda en modell för maskininlÀrning. Att utföra detta manuellt kan vara tidskrÀvande, utan garanti för god kvalitet pÄ resulterande hyperparametrar. Att anvÀnda verktyg för detta ÀndamÄl Àr att föredra, men det finns ett stort antal verktyg som anvÀnder olika algoritmer. Hur effektiva dessa olika verktyg Àr relativt varandra Àr ett mindre utforskat omrÄde. Denna studie bidrar med en enkel analys av hur tvÄ verktyg för sökning av hyperparametrar, Scikit och Ray Tune, fungerar i jÀmförelse med varandra.

[77] Johan Lind. 2021.
Evaluating CNN-based models for unsupervised image denoising.
Student Thesis. 43 pages. ISRN: LIU-IDA/LITH-EX-A--21/011--SE.

Images are often corrupted by noise which reduces their visual quality and interferes with analysis. Convolutional Neural Networks (CNNs) have become a popular method for denoising images, but their training typically relies on access to thousands of pairs of noisy and clean versions of the same underlying picture. Unsupervised methods lack this requirement and can instead be trained purely using noisy images.This thesis evaluated two different unsupervised denoising algorithms: Noise2Self (N2S) and Parametric Probabilistic Noise2Void (PPN2V), both of which train an internal CNN to denoise images. Four different CNNs were tested in order to investigate how the performance of these algorithms would be affected by different network architectures. The testing used two different datasets: one containing clean images corrupted by synthetic noise, and one containing images damaged by real noise originating from the camera used to capture them.Two of the networks, UNet and a CBAM-augmented UNet resulted in high performance competitive with the strong classical denoisers BM3D and NLM. The other two networks - GRDN and MultiResUNet - on the other hand generally caused poor performance.

[76] Oskar Holmström. 2021.
Exploring Transformer-Based Contextual Knowledge Graph Embeddings: How the Design of the Attention Mask and the Input Structure Affect Learning in Transformer Models.
Student Thesis. 41 pages. ISRN: LIU-IDA/LITH-EX-A--21/002--SE.

The availability and use of knowledge graphs have become commonplace as a compact storage of information and for lookup of facts. However, the discrete representation makes the knowledge graph unavailable for tasks that need a continuous representation, such as predicting relationships between entities, where the most probable relationship needs to be found. The need for a continuous representation has spurred the development of knowledge graph embeddings. The idea is to position the entities of the graph relative to each other in a continuous low-dimensional vector space, so that their relationships are preserved, and ideally leading to clusters of entities with similar characteristics. Several methods to produce knowledge graph embeddings have been created, from simple models that minimize the distance between related entities to complex neural models. Almost all of these embedding methods attempt to create an accurate static representation of each entity and relation. However, as with words in natural language, both entities and relations in a knowledge graph hold different meanings in different local contexts. With the recent development of Transformer models, and their success in creating contextual representations of natural language, work has been done to apply them to graphs. Initial results show great promise, but there are significant differences in archi- tecture design across papers. There is no clear direction on how Transformer models can be best applied to create contextual knowledge graph embeddings. Two of the main differences in previous work is how the attention mask is applied in the model and what input graph structures the model is trained on. This report explores how different attention masking methods and graph inputs affect a Transformer model (in this report, BERT) on a link prediction task for triples. Models are trained with five different attention masking methods, which to varying degrees restrict attention, and on three different input graph structures (triples, paths, and interconnected triples). The results indicate that a Transformer model trained with a masked language model objective has the strongest performance on the link prediction task when there are no restrictions on how attention is directed, and when it is trained on graph structures that are sequential. This is similar to how models like BERT learn sentence structure after being exposed to a large number of training samples. For more complex graph structures it is beneficial to encode information of the graph structure through how the attention mask is applied. There also seems to be some indications that the input graph structure affects the models’ capabilities to learn underlying characteristics in the knowledge graph that is trained upon.

[75] Rasmus Larsson. 2021.
Creating Digital Twin Distributed Networks Using Switches With Programmable Data Plane.
Student Thesis. 70 pages. ISRN: LIU-IDA/LITH-EX-A--2021/009--SE.

The domain specific language P4 is a novel initiative which extends the Software-Defined Networking (SDN) paradigm by allowing for data plane programmability. Network virtualisation is a class of network technologies which can be used to abstract the addressing in a network, allowing multiple tenants to utilise the network resources while being agnostic to the underlying network and the other tenants. In other words, <em>twins</em> of tenants using the same addresses can co-exist on the same underlying network. If a twin is a distributed network, it may even be spread out across multiple sites which are connected to a common backbone.In this study, network virtualisation using P4 is evaluated with emphasis on scalability in terms of number of twins and sites. A set of potential network virtualisation technologies are identified and categorised. Based on this categorisation, two variations of network virtualisation are implemented on the P4 capable software switch BMv2 and the performance of both variations are evaluated against the non-P4 solution Linux bridge. Linux bridge was found to yield 451 times more useful bandwidth than the best performing P4 implementation on BMv2, while also learning MAC addresses faster and generating less traffic on the backbone. It is concluded that the performance of network virtualisation implemented and running on BMv2 is worse compared to the non-P4 solution Linux bridge.

[74] Carl Brage. 2021.
Synchronizing 3D data between software: Driving 3D collaboration forward using direct links.
Student Thesis. 43 pages. ISRN: LIU-IDA/LITH-EX-A--21/010—SE.

In the area of 3D visualization there are often several stages in the design process. These stages can involve creating a model, applying a texture to the model and creating a rendered image from the model. Some software can handle all stages of the process while some are focused on a single stage to try to perfect and narrow down the service provided. In this case there needs to be a way to transfer 3D data between software in an efficient way where the user experience isn’t lacking. This thesis explores the area of 3D data synchronization by first getting foundation from the prestudy and literature study. The findings from these studies are used in a shared file-based implementation and a design of a network-based system. The work presented in this thesis forms a comprehensive overview which can be used for future work.

[73] Berggren Mathias and Sonesson Daniel. 2021.
Design Optimization in Gas Turbines using Machine Learning: A study performed for Siemens Energy AB.
Student Thesis. 56 pages. ISRN: LIU-IDA/LITH-EX-A-21/007--SE.

In this thesis, the authors investigate how machine learning can be utilized for speeding up the design optimization process of gas turbines. The Finite Element Analysis (FEA) steps of the design process are examined if they can be replaced with machine learning algorithms. The study is done using a component with given constraints that are provided by Siemens Energy AB. With this component, two approaches to using machine learning are tested. One utilizes design parameters, i.e. raw floating-point numbers, such as the height and width. The other technique uses a high dimensional mesh as input. It is concluded that using design parameters with surrogate models is a viable way of performing design optimization while mesh input is currently not. Results from using different amount of data samples are presented and evaluated.

2020
[72] André Willquist. 2020.
Uncertainty Discretization for Motion Planning Under Uncertainty.
Student Thesis. 55 pages. ISRN: LIU-IDA/LITH-EX-A--20/060--SE.

In this thesis, the problem of motion planning under uncertainty is explored. Motion planning under uncertainty is important since even with noise during the execution of the plan, it is desirable to keep the collision risk low. However, for the motion planning to be useful it needs to be possible to perform it in a reasonable time. The introduction of state uncertainty leads to a substantial increase in search time due to the additional dimensions it adds to the search space. In order to alleviate this problem, different approaches to pruning of the search space are explored. The initial approach is to prune states based on having strictly worse uncertainty and path cost than other found states. Having performed this initial pruning, an alternate approach to comparing uncertainties is examined in order to explore if it is possible to achieve a lower search time. The approach taken in order to lower the search time further is to discretize the covariance of a state by using a number of buckets. However, this discretization results in giving up the completeness and optimality of the algorithm. Having implemented these different ways of pruning, their performance is tested on a number of different scenarios. This is done by evaluating the planner using the pruning in several different scenarios including uncertainty and one without uncertainty. It is found that all of the pruning approaches reduce the overall search time compared to when no additional pruning based on the uncertainty is done. Additionally, it is indicated that the bucket-based approach reduce the search time to a greater extent than the strict pruning approach. Furthermore, the extensions made results in no increase in cost or a very small increase in cost for the explored scenarios. Based on these results, it is likely that the bucket pruning approach has some potential. However more studies, particularly with additional scenarios, needs to be made before any definitive conclusions can be made.

[71] André Willquist. 2020.
Uncertainty Discretization for Motion Planning Under Uncertainty.
Student Thesis. 55 pages. ISRN: LIU-IDA/LITH-EX-A--20/060--SE.

In this thesis, the problem of motion planning under uncertainty is explored.Motion planning under uncertainty is important since even with noise during the execution of the plan, it is desierable to keep the collision risk low.However, for the motion planning to be useful it needs to be possible to perform it in a reasonable time.The introduction of state uncertainty leads to a substantial increase in search time due to the additional dimensions it adds to the search space.In order to alleviate this problem, different approaches to pruning of the search space are explored.The initial approach is to prune states based on having strictly worse uncertainty and path cost than other found states.Having performed this initial pruning, an alternate approach to comparing uncertainties is examined in order to explore if it is possible to achieve a lower search time. The approach taken in order to lower the search time further is to discretize the covariance of a state by using a number of buckets.However, this discretization results in giving up the completeness and optimality of the algorithm.Having implemented these different ways of pruning, their performance is tested on a number of different scenarios.This is done by evaluating the planner using the pruning in several different scenarios including uncertainty and one without uncertainty.It is found that all of the pruning approaches reduce the overall search time compared to when no additional pruning based on the uncertainty is done.Additionally, it is indicated that the bucket based approach reduce the search time to a greater extent than the strict pruning approach.Furthermore, the extensions made results in no increase in cost or a very small increase in cost for the explored scenarios.Based on these results, it is likely that the bucket pruning approach has some potential.However more studies, perticularly with additional scenarios, needs to be made before any definitive conclusions can be made.

[70] Zacharias Nordström. 2020.
Extracting Behaviour Trees from Deep Q-Networks: Using learning from demostration to transfer knowledge between models.
Student Thesis. 54 pages. ISRN: LIU-IDA/LITH-EX-A--20/059—SE.

In recent years the advancement in machine learning have solved more and more complex problems. But still these techniques are not commonly used in the industry. One problem is that many of the techniques are black boxes, it is hard to analyse them to make sure that their behaviour is safe. This property makes them unsuitable for safety critical systems. The goal of this thesis is to examine if the deep learning technique Deep Q-network could be used to create a behaviour tree that can solve the same problem. A behaviour tree is a tree representation of a flow structure that is used for representing behaviours, often used in video games or robotics. To solve the problem two simulators are used, one models a cart that shall balance a pole called cart pole, the other is a static world which needs to be navigated called grid world. Inspiration is taken from the learning from demonstration field to use the Deep Q-network as a teacher and then create a decision tree. During the creation of the decision tree two attributes are used for pruning; to look at the trees accuracy or performance. The thesis then compare three techniques, called Naive, BT Espresso, and BT Espresso Simplified. The techniques are used to transform the extracted decision tree into a behaviour tree. When it comes to the performance of the created behaviour trees they all manage to complete the simulator scenarios in the same, or close to, capacity as the trained Deep Q-network. The trees created from the performance pruned decision tree are generally smaller and less complex, but they have worse accuracy. For cart pole the trees created from the accuracy pruned tree has around 10 000 nodes but the performance pruned trees have around 10-20 nodes. The difference in grid world is smaller going from 35-45 nodes to 40-50 nodes. To get the smallest tree with the best performance then the performance pruned tree should be used with the BT Espresso Simplified algorithm. This thesis have shown that it is possible to use knowledge from a trained Deep Q-network model to create a Behaviour tree that can complete the same task.

[69] Karol Wojtulewicz and Viktor Agbrink. 2020.
Evaluating DCNN architecturesfor multinomial area classicationusing satellite data.
Student Thesis. ISRN: LIU-IDA/LITH-EX-A--20/031--SE.

The most common approach to analysing satellite imagery is building or object segmentation,which expects an algorithm to find and segment objects with specific boundaries thatare present in the satellite imagery. The company Vricon takes satellite imagery analysisfurther with the goal of reproducing the entire world into a 3D mesh. This 3D reconstructionis performed by a set of complex algorithms excelling in different object reconstructionswhich need sufficient labeling in the original 2D satellite imagery to ensure validtransformations. Vricon believes that the labeling of areas can be used to improve the algorithmselection process further. Therefore, the company wants to investigate if multinomiallarge area classification can be performed successfully using the satellite image data availableat the company. To enable this type of classification, the company’s gold-standarddataset containing labeled objects such as individual buildings, single trees, roads amongothers, has been transformed into an large area gold-standard dataset in an unsupervisedmanner. This dataset was later used to evaluate large area classification using several stateof-the-art Deep Convolutional Neural Network (DCNN) semantic segmentation architectureson both RGB as well as RGB and Digital Surface Model (DSM) height data. Theresults yield close to 63% mIoU and close to 80% pixel accuracy on validation data withoutusing the DSM height data in the process. This thesis additionally contributes with a novelapproach for large area gold-standard creation from existing object labeled datasets.

[68] Marc Pàmies Massip. 2020.
Multilingual identification of offensive content in social media.
Student Thesis. 50 pages. ISRN: LIU-IDA/LITH-EX-A--20/053--SE.

In today’s society there is a large number of social media users that are free to express their opinion on shared platforms. The socio-cultural differences between the people behind those accounts (in terms of ethnicity, gender, sexual orientation, religion, politics, . . . ) give rise to an important percentage of online discussions that make use of offensive language, which often affects in a negative way the psychological well-being of the victims. In order to address the problem, the endless stream of user-generated content engenders a need to find an accurate and scalable solution to detect offensive language using automated methods. This thesis explores different approaches to the offensiveness detection task focusing on five different languages: Arabic, Danish, English, Greek and Turkish. The results obtained using Support Vector Machines (SVM), Convolutional Neural Networks (CNN) and the Bidirectional Encoder Representations from Transformers (BERT) are compared, achieving state-of-the-art results with some of the methods tested. The effect of the embeddings used, the dataset size, the class imbalance percentage and the addition of sentiment features are studied and analysed, as well as the cross-lingual capabilities of pre-trained multilingual models.

[67] Samuel Blomqvist and Björn Detterfelt. 2020.
Real Time Integrated Tools for Video Game Development: a usability study.
Student Thesis. 78 pages. ISRN: LIU-IDA/LITH-EX-A--20/050--SE.

The video game industry can be ruthless. As a developer, you usually find yourself working in the popular third-party development tools of the time. These tools however might not provide the best usability and quality of life one desires. This can lead to a lot of frustration for the developer, especially when the development enters a crunch period of long and hard work. We believe some of the frustration can be avoided, and we believe this can be done by creating effective, functional and user-friendly integrated development tools specialized for the development environment. In this master's thesis we investigated just that, how integrated game development tools can be designed to be usable in terms of effectiveness and learnability. The investigation was performed by designing and implementing an integrated game development tool. The development of the tool was performed iteratively with user testing between every iteration to find usability defects, allowing the tool to be refined and improved throughout the development process. To finish off the development process, there was a final user test where professional video game developers tried out the tool and then answered a System Usability Scale questionnaire. The System Usability Scale score and task completion rate showed that the final state of the tool can be considered highly usable in terms of effectiveness and averagely usable in terms of learnability. This suggests that involving user testing in the development process is vital for ensuring good usability in the end product.

[66] Pia Lűtvedt. 2020.
Implementation of visualizations using a server-client architecture: Effects on performance measurements.
Student Thesis. 10 pages. ISRN: LIU-IDA/LITH-EX-G--20/052--SE.

Visualizing large datasets poses challenges in terms of how to create visualization applications with good performance. Due to the amount of data, transfer speed and processing speed may lead to waiting times that cause users to abandon the application. It is therefore important to select methods and techniques that can handle the data in as efficient a way as possible. The aim of this study was to investigate if a server-client architecture had better performance in a visualization web application than a purely client-side architecture in terms of selected performance metrics and network load, and whether the selection of implementation language and tools affected the performance of the server-client architecture implementation. To answer these questions, a visualization application was implemented in three different ways: a purely client-side implementation, a server-client implementation using Node.js for the server, and a server-client implementation using Flask for the server. The results showed that the purely client-side architecture suffered from a very long page loading time and high network load but was able to process data quickly in response to user actions in the application. The server-client architecture implementations could load the page faster, but responding to requests took longer, whereas the amount of data transferred was much lower. Furthermore, the server-client architecture implemented with a Node.js server performed better on all metrics than the application implemented with a Flask server. Overall, when taking all measurements into consideration, the Node.js server architecture may be the best choice among the three when working with a large dataset, although the longer response time compared to the purely client-side architecture may cause the application to seem less responsive.

[65] Sebastian Lundqvist and Oliver Ekstrand. 2020.
Evaluating an ARCore application to get an image of the state of AR technology today.
Student Thesis. 8 pages. ISRN: LIU-IDA/LITH-EX-G--20/051--SE.

Augmented reality is an old technology that is still far away from being perfect. It is also quickly being improved upon and the state of AR today has come a long way from AR just a couple of years ago. New big players have recently introduced their tools and have made it easier than ever to develop AR applications. In this study we look at what established methods (if any) there are for AR evaluation, develop AR evaluation methods that fit our needs, carry out the evaluation and analyze the collected data. We also note some important things to think about when working with AR to increase tracking and recognition stability. The recommendations are: try to have reference images with high scores, have reference objects that are distinct enough from one another to not be mixed up and make sure that the visual for the reference image matches the visual for the reference object in its intended viewing environment.

[64] Per Olin. 2020.
Evaluation of text classification techniques for log file classification.
Student Thesis. 48 pages. ISRN: LIU-IDA/LITH-EX-A--20/048--SE.

System log files are filled with logged events, status codes, and other messages. By analyzing the log files, the systems current state can be determined, and find out if something during its execution went wrong. Log file analysis has been studied for some time now, where recent studies have shown state-of-the-art performance using machine learning techniques. In this thesis, document classification solutions were tested on log files in order to classify regular system runs versus abnormal system runs. To solve this task, supervised and unsupervised learning methods were combined. Doc2Vec was used to extract document features, and Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) based architectures on the classification task. With the use of the machine learning models and preprocessing techniques the tested models yielded an f1-score and accuracy above 95% when classifying log files.

[63] Oscar Lundblad. 2020.
The autonomous crewmate: A sociotechnical perspective to implementation of autonomous vehicles in sea rescue.
Student Thesis. 72 pages. ISRN: LIU-IDA/KOGVET-A--20/009--SE.

The usage of autonomous vehicles is starting to appear in several different domains and the domain of public safety is no exception. Wallenberg Artificial Intelligence, Autonomous Systems and Software Program (WASP) has created a research arena for public safety (WARA-PS) to explore experimental features, usages, and implementation of autonomous vehicles within the domain of public safety. Collaborating in the arena are several companies, universities, and researchers. This thesis examines, in collaboration with Combitech, a company partnered in WARA-PS, how the implementation of autonomous vehicles affects the sociotechnical system of a search and rescue operation during a drifting boat with potential castaways. This is done by creating a case together with domain experts, analyzing the sociotechnical system within the case using cognitive work analysis and then complementing the analyses with the unmanned autonomous vehicles of WARA-PS. This thesis has shown how the WARA-PS vehicles can be implemented in the case of a drifting boat with potential castaways and how the implementation affects the sociotechnical system. Based on the analyses and opinions of domain experts’ future guidelines has been derived to further the work with sociotechnical aspects in WARA-PS.

[62] Viktor Brandt and Jesper Olofsson. 2020.
Undersökning av flexibel implementation för hantering av multipla rösttjänster.
Student Thesis. 12 pages. ISRN: LIU-IDA/LITH-EX-G--20/021--SE.

Att vÀlja vilken eller vilka röststyrningstjÀnster man som företag vill stödja kan i dagens lÀge vara ett svÄrt val att göra. Det kan Àven var sÄ att man inte har resurser att göra tvÄ olika implementationer. I den hÀr undersökningen tittar vi pÄ om det finns ett bra sÀtt att göra en implementation som kan hantera fler Àn en röststyrningstjÀnst. TjÀnsterna vi har fokuserat pÄ i undersökningen Àr Amazon Alexa och Google Assistant.

[61] Arvid Edenheim. 2020.
Using Primary Dynamic Factor Analysis on repeated cross-sectional surveys with binary responses.
Student Thesis. 54 pages. ISRN: LIU-IDA/LITH-EX-A--20/007--SE.

With the growing popularity of business analytics, companies experience an increasing need of reliable data. Although the availability of behavioural data showing what the consumers do has increased, the access to data showing consumer mentality, what the con- sumers actually think, remain heavily dependent on tracking surveys. This thesis inves- tigates the performance of a Dynamic Factor Model using respondent-level data gathered through repeated cross-sectional surveys. Through Monte Carlo simulations, the model was shown to improve the accuracy of brand tracking estimates by double digit percent- ages, or equivalently reducing the required amount of data by more than a factor 2, while maintaining the same level of accuracy. Furthermore, the study showed clear indications that even greater performance benefits are possible.

2019
[60] Anton Silfver. 2019.
Short-Term Forecasting of Taxi Demand using a two Channelled Convolutional LSTM network.
Student Thesis. 39 pages. ISRN: LIU-IDA/LITH-A--19/097—SE.

In this thesis a model capable of predicting taxidemand with high accuracy across five different real world single company datasets is presented. The model uses historical drop off and arrival information to make accurate shortterm predictions about future taxi demand. The model is compared to and outperforms both LSTM and statistical baselines. This thesis uniquely uses a different tessellation strategy which makes the results directly applicable to smaller taxi companies.This paper shows that accurate short term predictions of taxi demand can be made using real world data available to taxi companies. MSE is also shown to be a more robust to uneven demand distributions across cities than MAE. Adding drop offs to the input had provided only marginal improvements in the performance of the model.

[59] Christoffer Fors Johansson. 2019.
Arrival Time Predictions for Buses using Recurrent Neural Networks.
Student Thesis. 65 pages. ISRN: LIU-IDA/LITH-EX-A--19/098--SE.

In this thesis, two different types of bus passengers are identified. These two types, namely current passengers and passengers-to-be have different needs in terms of arrival time predictions. A set of machine learning models based on recurrent neural networks and long short-term memory units were developed to meet these needs. Furthermore, bus data from the public transport in Östergötland county, Sweden, were collected and used for training new machine learning models. These new models are compared with the current prediction system that is used today to provide passengers with arrival time information.The models proposed in this thesis uses a sequence of time steps as input and the observed arrival time as output. Each input time step contains information about the current state such as the time of arrival, the departure time from thevery first stop and the current position in Cartesian coordinates. The targeted value for each input is the arrival time at the next time step. To predict the rest of the trip, the prediction for the next step is simply used as input in the next time step.The result shows that the proposed models can improve the mean absolute error per stop between 7.2% to 40.9% compared to the system used today on all eight routes tested. Furthermore, the choice of loss function introduces models thatcan meet the identified passengers need by trading average prediction accuracy for a certainty that predictions do not overestimate or underestimate the target time in approximately 95% of the cases.

[58] Magnus Selin. 2019.
Efficient Autonomous Exploration Planning of Large-Scale 3D-Environments: A tool for autonomous 3D exploration indoor.
Student Thesis. 50 pages. ISRN: LIU-IDA/LITH-EX-A--19/017--SE.

Exploration is of interest for autonomous mapping and rescue applications using unmanned vehicles. The objective is to, without any prior information, explore all initially unmapped space.We present a system that can perform fast and efficient exploration of large scale arbitrary 3D environments. We combine frontier exploration planning (FEP) as a global planning strategy, together with receding horizon planning (RH-NBVP) for local planning. This leads to plans that incorporate information gain along the way, but do not get stuck in already explored regions. Furthermore, we make the potential information gain estimation more efficient, through sparse ray-tracing, and caching of already estimated gains. The worked carried out in this thesis has been published as a paper in Robotand Automation letters and presented at the International Conference on Robotics and Automation in Montreal 2019.

[57] Viktor Holmgren. 2019.
General-purpose maintenance planning using deep reinforcement learning and Monte Carlo tree search.
Student Thesis. 42 pages. ISRN: LIU-IDA/LITH-EX-A--19/096--SE.

Maintenance planning and execution is increasingly important for the modern industrial sector. Maintenance costs can amount to a major part of industrial spending. However, it is not as simple as just reducing maintenance budgets. A balance must be struck between risking unplanned downtime and the costs of maintenance efforts, in order to keep the profit margins needed to compete in the global markets of today. One approach to improve the effectiveness of industries is to apply intelligent maintenance planners. In this thesis, a general-purpose maintenance planner based on Monte-Carlotree search and deep reinforcement learning is presented. This planner was evaluated and compared against two different periodic planners as well as the oracle lower bound on four different maintenance scenarios. These four scenarios are all based on servicing wind turbines. All scenarios include imperfect maintenance actions, as well as uncertainty in terms of the outcomes of maintenance actions. Furthermore, the four scenarios include both single and multi-component variants. The evaluation showed that the proposed method is outperforming both periodic planners in three of the four scenarios, with the forth being inconclusive. These results indicate that the maintenance planner introduced in this paper is a viable method, at least for these types of maintenance problems. However, further research is needed on this topic of maintenance planning under uncertainty. More specifically, the viability of the proposed method on a more diverse set of maintenance problems is needed to draw any clear general conclusions. Finally, possible improvements to the training process that are discussed in this thesis should be investigated.

[56] Fredrik Bengtsson and Adam Combler. 2019.
Automatic Dispatching of Issues using Machine Learning.
Student Thesis. 90 pages. ISRN: LIU-IDA/LITH-EX-A--19/043--SE.

Many software companies use issue tracking systems to organize their work. However, when working on large projects, across multiple teams, a problem of finding the correctteam to solve a certain issue arises. One team might detect a problem, which must be solved by another team. This can take time from employees tasked with finding the correct team and automating the dispatching of these issues can have large benefits for the company. In this thesis, the use of machine learning methods, mainly convolutional neural networks (CNN) for text classification, has been applied to this problem. For natural language processing both word- and character-level representations are commonly used. The results in this thesis suggests that the CNN learns different information based on whether word- or character-level representation is used. Furthermore, it was concluded that the CNN models performed on similar levels as the classical Support Vector Machine for this task. When compared to a human expert, working with dispatching issues, the best CNN model performed on a similar level when given the same information. The high throughput of a computer model, therefore, suggests automation of this task is very much possible.

[55] Sebastian Sibelius Parmbäck. 2019.
HMMs and LSTMs for On-line Gesture Recognition on the Stylaero Board: Evaluating and Comparing Two Methods.
Student Thesis. 36 pages. ISRN: LIU-IDA/LITH-EX-A--2019/091--SE.

In this thesis, methods of implementing an online gesture recognition system for the novel Stylaero Board device are investigated. Two methods are evaluated - one based on LSTMs and one based on HMMs - on three kinds of gestures: Tap, circle, and flick motions. A method’s performance was measured in its accuracy in determining both whether any of the above listed gestures were performed and, if so, which gesture, in an online single-pass scenario. Insight was acquired regarding the technical challenges and possible solutions to the online aspect of the problem. Poor performance was, however, observed in both methods, with a likely culprit identified as low quality of training data, due to an arduous and complex gesture performance capturing process. Further research improving on the process of gathering data is suggested.

[54] Richard Wigren and Filip Cornell. 2019.
Marketing Mix Modelling: A comparative study of statistical models.
Student Thesis. 113 pages. ISRN: LIU-IDA/LITH-EX-A--19/054--SE.

Deciding the optimal media advertisement spending is a complex issue that many companies today are facing. With the rise of new ways to market products, the choices can appear infinite. One methodical way to do this is to use Marketing Mix Modelling (MMM), in which statistical modelling is used to attribute sales to media spendings. However, many problems arise during the modelling. Modelling and mitigation of uncertainty, time-dependencies of sales, incorporation of expert information and interpretation of models are all issues that need to be addressed. This thesis aims to investigate the effectiveness of eight different statistical and machine learning methods in terms of prediction accuracy and certainty, each one addressing one of the previously mentioned issues. It is concluded that while Shapley Value Regression has the highest certainty in terms of coefficient estimation, it sacrifices some prediction accuracy. The overall highest performing model is the Bayesian hierarchical model, achieving both high prediction accuracy and high certainty.

[53] David Hilm and David Rahim. 2019.
Two-factor Authentication and Digital Signing for an Enterprise System utilizing Yubikey.
Student Thesis. 10 pages. ISRN: LIU-IDA/LITH-EX-G--19/040--SE.

The use of a second factor to increase the security of systems is growing and has continued to do so for a long time. This thesis explores options for implementation to use a YubiKey as an authentication method (OTP) as well as for signing digital transactions through a web browser client. Measures of network overhead that occurs in conjunction with Digital Signing of transactions are also disclosed. Our findings show that YubiKey provides flexible and readily available solutions that can be used with only small implementations for OTP authentication. It is also shown that the major concern for implementing a solution for a web browser is to intuitively use certificates stored on a USB-device without installing any plugins or with the use of a third-party application running on the client machine.

[52] Martin Lundberg. 2019.
Automatic parameter tuning in localization algorithms.
Student Thesis. 57 pages. ISRN: LIU-IDA/LITH-EX-A--19/052--SE.

Many algorithms today require a number of parameters to be set in order to perform well in a given application. The tuning of these parameters is often difficult and tedious to do manually, especially when the number of parameters is large. It is also unlikely that a human can find the best possible solution for difficult problems. To be able to automatically find good sets of parameters could both provide better results and save a lot of time.The prominent methods Bayesian optimization and Covariance Matrix Adaptation Evolution Strategy (CMA-ES) are evaluated for automatic parameter tuning in localization algorithms in this work. Both methods are evaluated using a localization algorithm on different datasets and compared in terms of computational time and the precision and recall of the final solutions. This study shows that it is feasible to automatically tune the parameters of localization algorithms using the evaluated methods. In all experiments performed in this work, Bayesian optimization was shown to make the biggest improvements early in the optimization but CMA-ES always passed it and proceeded to reach the best final solutions after some time. This study also shows that automatic parameter tuning is feasible even when using noisy real-world data collected from 3D cameras.

[51] Elina Lundberg and Erica Gavefalk. 2019.
Investigating the impact on subjective satisfaction and learnability when adopting cloud in an SME.
Student Thesis. 74 pages. ISRN: LIU-IDA/LITH-EX-A--19/030--SE.

Cloud services and solutions have served as a shift in the computer industry and create new opportunities for users. Clouds have been described as easily usable and fluid in terms of expansion and contraction depending on the real-time needs. Although the cloud is promoted with several benefits, it is not always apparent for the users that this is the case. Understanding both the benefits and challenges that exist is substantial for a successful adoption to cloud. This master’s thesis is conducted in collaboration with Exsitec ABand aims to investigate how the adoption of the cloud service Microsoft Azure will affect the development process. Also, it aims to provide a best practice for potentially needed updated working procedures, in terms of satisfaction and learnability. The investigation was performed through interviews and the System Usability Scale, to assess how the end users experienced development in a cloud environment. The thesis revealed that the Azure portal has low overall usability, but that there also exists an inconsistency of that perception. Two major factors that contributed to the satisfaction and learnability was the lack of documentation and that the Azure portal was considered hard to master. The SUS score revealed that the mean value was below an acceptable level, and thus changes in the company’s working procedures need to be implemented. Internal documentation regarding how the company should use both cloud in general, as well as the portal in particular, are required in order to increase the learnability and subjective satisfaction.

[50] Full text  David Bergström. 2019.
Bayesian optimization for selecting training and validation data for supervised machine learning: using Gaussian processes both to learn the relationship between sets of training data and model performance, and to estimate model performance over the entire problem domain.
Student Thesis. 39 pages. ISRN: LIU-IDA/LITH-EX-A--19/016--SE.

Validation and verification in machine learning is an open problem which becomes increasingly important as its applications becomes more critical. Amongst the applications are autonomous vehicles and medical diagnostics. These systems all needs to be validated before being put into use or else the consequences might be fatal.This master’s thesis focuses on improving both learning and validating machine learning models in cases where data can either be generated or collected based on a chosen position. This can for example be taking and labeling photos at the position or running some simulation which generates data from the chosen positions.The approach is twofold. The first part concerns modeling the relationship between any fixed-size set of positions and some real valued performance measure. The second part involves calculating such a performance measure by estimating the performance over a region of positions.The result is two different algorithms, both variations of Bayesian optimization. The first algorithm models the relationship between a set of points and some performance measure while also optimizing the function and thus finding the set of points which yields the highest performance. The second algorithm uses Bayesian optimization to approximate the integral of performance over the region of interest. The resulting algorithms are validated in two different simulated environments.The resulting algorithms are applicable not only to machine learning but can also be used to optimize any function which takes a set of positions and returns a value, but are more suitable when the function is expensive to evaluate.

[49] Pernilla Eilert. 2019.
Learning behaviour trees for simulated fighter pilots in airborne reconnaissance missions: A grammatical evolution approach.
Student Thesis. 94 pages. ISRN: LIU-IDA/LITH-EX-A--19/015--SE.

Fighter pilots often find themselves in situations where they need to make quick decisions. Therefore an intelligent decision support system that suggests how the fighter pilot should act in a specific situation is vital. The aim of this project is to investigate and evaluate grammatical evolution paired with behaviour trees to develop a decision support system. This support system should control a simulated fighter pilot during an airborne reconnaissance mission. This thesis evaluates the complexity of the evolved trees and the performance, and robustness of the algorithm. Key factors were identified for a successful system: scenario, fitness function, initialisation technique and control parameters. The used techniques were decided based on increasing performance of the algorithm and decreasing complexity of the tree structures. The initialisation technique, the genetic operators and the selection functions performed well but the fitness function needed more work. Most of the experiments resulted in local maxima. A desired solution could only be found if the initial population contained an individual with a BT succeeding the mission. However, the implementation behaved as expected. More and longer simulations are needed to draw a conclusion of the performance based on robustness, when testing the evolved BT:s on different scenarios. Several methods were studied to decrease the complexity of the trees and the experiments showed a promising variation of complexity through the generations when the best fitness was fixed. A feature was added to the algorithm, to promote lower complexity when equal fitness value. The results were poor and implied that pruning would be a better fit after the simulations. Nevertheless, this thesis suggests that it is suitable to implement a decision support system based on grammatical evolution paired with behaviour trees as framework.

2018
[48] Anton Hölscher. 2018.
A Cycle-Trade Heuristic for the Weighted k-Chinese Postman Problem.
Student Thesis. 23 pages. ISRN: LIU-IDA/LITH-EX-G--18/073--SE.

This study aims to answer whether a heuristic that trades cycles between the tours in a solution would show good results when trying to solve the Weighted k-Chinese Postman Problem for undirected graphs, of varying size, representing neighbourhoods in Sweden.A tabu search heuristic was implemented with each iteration consisting of giving a cycle from the most expensive tour to the cheapest. The heuristic performed increasingly well for graphs of increasing size, although the solution quality decreased when increasing the number of tours to be used in the solution. It is suspected that the cause for this behavior is due to the heuristic only giving cycles from the most expensive tour, not considering trading cycles from other tours in the solution. It is believed that a heuristic considering more than only the most expensive tour when trading cycles would produce even better solutions.

[47] Linus Kortesalmi. 2018.
Gaussian Process Regression-based GPS Variance Estimation and Trajectory Forecasting.
Student Thesis. 59 pages. ISRN: LIU-IDA/LITH-EX-A--18/040--SE.

Spatio-temporal data is a commonly used source of information. Using machine learning to analyse this kind of data can lead to many interesting and useful insights. In this thesis project, a novel public transportation spatio-temporal dataset is explored and analysed. The dataset contains 282 GB of positional events, spanning two weeks of time, from all public transportation vehicles in Östergötland county, Sweden. From the data exploration, three high-level problems are formulated: bus stop detection, GPS variance estimation, and arrival time prediction, also called trajectory forecasting. The bus stop detection problem is briefly discussed and solutions are proposed. Gaussian process regression is an effective method for solving regression problems. The GPS variance estimation problem is solved via the use of a mixture of Gaussian processes. A mixture of Gaussian processes is also used to predict the arrival time for public transportation buses. The arrival time prediction is from one bus stop to the next, not for the whole trajectory. The result from the arrival time prediction is a distribution of arrival times, which can easily be applied to determine the earliest and latest expected arrival to the next bus stop, alongside the most probable arrival time. The naĂŻve arrival time prediction model implemented has a root mean square error of 5 to 19 seconds. In general, the absolute error of the prediction model decreases over time in each respective segment. The results from the GPS variance estimation problem is a model which can compare the variance for different environments along the route on a given trajectory.

[46] Erik Hansson. 2018.
Temporal Task and Motion Plans: Planning and Plan Repair: Repairing Temporal Task and Motion Plans Using Replanning with Temporal Macro Operators.
Student Thesis. 128 pages. ISRN: LIU-IDA/LITH-EX-A--18/047--SE.

This thesis presents an extension to the Temporal Fast Downward planning system that integrates motion planning in it and algorithms for generating two types of temporal macro operators expressible in PDDL2.1. The extension to the Temporal Fast Downward planning system includes, in addition to the integration of motion planning itself, an extension to the context-enhanced additive heuristic that uses information from the motion planning part to improve the heuristic estimate. The temporal macro operators expressible in PDDL2.1 are, to the author's knowledge, an area that is not studied within the context of plan repair before. Two types of temporal macro operators are presented along with algorithms for automatically constructing and using them when solving plan repair problems by replanning. Both the heuristic extension and the temporal macro operators were evaluated in the context of simulated unmanned aerial vehicles autonomously executing reconnaissance missions to identify targets and avoiding threats in unexplored areas. The heuristic extension was proved to be very helpful in the scenario. Unfortunately, the evaluation of the temporal macro operators indicated that the cost of introducing them is higher than the gain of using them for the scenario.

[45] Rasmus Johns Johns. 2018.
Intelligent Formation Control using Deep Reinforcement Learning.
Student Thesis. 53 pages. ISRN: LLIU-IDA/LITH-EX-A--2017/001--SE.

In this thesis, deep reinforcement learning is applied to the problem of formation control to enhance performance. The current state-of-the-art formation control algorithms are often not adaptive and require a high degree of expertise to tune. By introducing reinforcement learning in combination with a behavior-based formation control algorithm, simply tuning a reward function can change the entire dynamics of a group. In the experiments, a group of three agents moved to a goal which had its direct path blocked by obstacles. The degree of randomness in the environment varied: in some experiments, the obstacle positions and agent start positions were fixed between episodes, whereas in others they were completely random. The greatest improvements were seen in environments which did not change between episodes; in these experiments, agents could more than double their performance with regards to the reward. These results could be applicable to both simulated agents and physical agents operating in static areas, such as farms or warehouses. By adjusting the reward function, agents could improve the speed with which they approach a goal, obstacle avoidance, or a combination of the two. Two different and popular reinforcement algorithms were used in this work: Deep Double Q-Networks (DDQN) and Proximal Policy Optimization (PPO). Both algorithms showed similar success.

[44] Fredrik Hćkansson and Carl-Johan Larsson. 2018.
User-Based Predictive Caching of Streaming Media.
Student Thesis. 58 pages. ISRN: LIU-IDA/LITH-EX-A--18/033—SE.

Note: This thesis is written as a joint thesis between two students from different universities. This means the exact same thesis is published at two universities (LiU and KTH) but with different style templates. The other report has identification number: TRITA-EECS-EX-2018:403

Streaming media is a growing market all over the world which sets a strict requirement on mobile connectivity. The foundation for a good user experience when supplying a streaming media service on a mobile device is to ensure that the user can access the requested content. Due to the varying availability of mobile connectivity measures has to be taken to remove as much dependency as possible on the quality of the connection. This thesis investigates the use of a Long Short-Term Memory machine learning model for predicting a future geographical location for a mobile device. The predicted location in combination with information about cellular connectivity in the geographical area is used to schedule prefetching of media content in order to improve user experience and to reduce mobile data usage. The Long Short-Term Memory model suggested in this thesis achieves an accuracy of 85.15% averaged over 20000 routes and the predictive caching managed to retain user experience while decreasing the amount of data consumed.

[43] Adrian Sonnert. 2018.
Predicting inter-frequency measurements in an LTE network using supervised machine learning: a comparative study of learning algorithms and data processing techniques.
Student Thesis. 52 pages. ISRN: LIU-IDA/LITH-EX-A--18/017--SE.

With increasing demands on network reliability and speed, network suppliers need to effectivize their communications algorithms. Frequency measurements are a core part of mobile network communications, increasing their effectiveness would increase the effectiveness of many network processes such as handovers, load balancing, and carrier aggregation. This study examines the possibility of using supervised learning to predict the signal of inter-frequency measurements by investigating various learning algorithms and pre-processing techniques. We found that random forests have the highest predictive performance on this data set, at 90.7\% accuracy. In addition, we have shown that undersampling and varying the discriminator are effective techniques for increasing the performance on the positive class on frequencies where the negative class is prevalent. Finally, we present hybrid algorithms in which the learning algorithm for each model depends on attributes of the training data set. These algorithms perform at a much higher efficiency in terms of memory and run-time without heavily sacrificing predictive performance.

[42] Göran Svensson and Jonas Westlund. 2018.
Intravenous bag monitoring with Convolutional Neural Networks.
Student Thesis. 12 pages. ISRN: LIU-IDA/LITH-EX-G--2018/048--SE.

Drip bags are used in hospital environments to administerdrugs and nutrition to patients. Ensuring that they are usedcorrectly and are refilled in time are important for the safetyof patients. This study examines the use of a ConvolutionalNeural Network (CNN) to monitor the fluid levels of drip bagsvia image recognition to potentially form the base of an earlywarning system, and assisting in making medical care moreefficient. Videos of drip bags were recorded as they wereemptying their contents in a controlled environment and fromdifferent angles. A CNN was built to analyze the recordeddata in order to predict a bags fluid level with a 5% intervalprecision from a given image. The results show that the CNNused performs poorly when monitoring fluid levels in dripbags.

[41] Joel Odd and Emil Theologou. 2018.
Utilize OCR text to extract receipt data and classify receipts with common Machine Learning algorithms.
Student Thesis. 13 pages. ISRN: LIU-IDA/LITH-EX-G--18/043—SE.

This study investigated if it was feasible to use machine learning tools on OCR extracted text data to classify receipts and extract specific data points. Two OCR tools were evaluated, the first was Azure Computer Vision API and the second was Google Drive REST Api, where Google Drive REST Api was the main OCR tool used in the project because of its impressive performance. The classification task mainly tried to predict which of five given categories the receipts belongs to, and also a more challenging task of predicting specific subcategories inside those five larger categories. The data points we where trying to extract was the date of purchase on the receipt and the total price of the transaction. The classification was mainly done with the help of scikit-learn, while the extraction of data points was achieved by a simple custom made N-gram model.The results were promising with about 94 % cross validation score for classifying receipts based on category with the help of a LinearSVC classifier. Our custom model was successful in 72 % of cases for the price data point while the results for extracting the date was less successful with an accuracy of 50 %, which we still consider very promising given the simplistic nature of the custom model.

2017
[40] Simon Keisala. 2017.
Designing an Artificial Neural Network for state evaluation in Arimaa: Using a Convolutional Neural Network.
Student Thesis. 31 pages. ISRN: LIU-IDA/LITH-EX-G--17/024--SE.

Agents being able to play board games such as Tic Tac Toe, Chess, Go and Arimaa has been, and still is, a major difficulty in Artificial Intelligence. For the mentioned board games, there is a certain amount of legal moves a player can do in a specific board state. Tic Tac Toe have in average around 4-5 legal moves, with a total amount of 255168 possible games. Both Chess, Go and Arimaa have an increased amount of possible legal moves to do, and an almost infinite amount of possible games, making it impossible to have complete knowledge of the outcome.This thesis work have created various Neural Networks, with the purpose of evaluating the likelihood of winning a game given a certain board state. An improved evaluation function would compensate for the inability of doing a deeper tree search in Arimaa, and the anticipation is to compete on equal skills against another well-performing agent (meijin) having one less search depth.The results shows great potential. From a mere one hundred games against meijin, the network manages to separate good from bad positions, and after another one hundred games able to beat meijin with equal search depth.It seems promising that by improving the training and by testing different sizes for the neural network that a neural network could win even with one less search depth. The huge branching factor of Arimaa makes such an improvement of the evaluation beneficial, even if the evaluation would be 10 000 times more slow.

[39] Niclas Jonsson. 2017.
Implementation and testing of an FPT-algorithm for computing the h+ heuristic.
Student Thesis. 39 pages. ISRN: LIU-IDA/LITH-EX-G–17/077–SE.

We have implemented and benchmarked an FPT-algorithm, that has two input parameters, k and w besides the input problem instance, which is a planing instance, in this thesis. The algorithm has an exponential running time as a function of these two parameters. The implemented algorithm computes the heuristic value h^+(s) of a state s that belongs to a state space, which originates from a strips instance. The purpose of the project was to test if the algorithm can be used to compute the heuristic function h^+, i.e. the delete-relaxation heuristic, in practice. The delete-relaxation heuristic value for some state is the length of the optimal solution from the state to a goal in the delete-relaxed-instance, which is the original instance without all its negative effects. Planning instances was benchmarked with the search algorithm A^* to test the algorithms practical value. The heuristic function blind was benchmarked together with A^* with the same instances so that we could compare the quality of the benchmark result for the implemented algorithm. The conclusion of the project was that the implemented algorithm is too slow to be used in practise.

[38] Henrik Phung. 2017.
Software developers? performance awareness.
Student Thesis. 49 pages. ISRN: LIU-IDA/LITH-EX-A--17/016--SE.

Automated tests and non-functional requirements are two widely used terms in the software development sector. Both are essential for software development teams but rarely mentioned together. Today, most software development teams are utilizing the development practice continuous integration. A method where software is built in iterations and in each iteration small chunks of code are merged into the main repository. Continuous integration requires automated tests to verify that each chunk of code is compatible with the main chunk. Automated test is essential for continuous integration to detect anomalies in each chunk of code. Customer satisfaction is a result of how well the delivered product performs in terms of non-functional requirements. Although the term “non-functional requirement” has not been formally defined and the existing definitions are diverse. In this thesis, we define the non-functional requirement, response time with help from a user-centered evaluation of responsiveness study. We create a test suite that can be ran on an automated build with focus on user-action-response. Based on the test result and a conducted survey, we evaluate how aware developers are when it comes to causes to performance issues.

[37] Full text  Daniel Artchounin. 2017.
Tuning of machine learning algorithms for automatic bug assignment.
Student Thesis. 135 pages. ISRN: LIU-IDA/LITH-EX-A--17/022--SE.

In software development projects, bug triage consists mainly of assigning bug reports to software developers or teams (depending on the project). The partial or total automation of this task would have a positive economic impact on many software projects. This thesis introduces a systematic four-step method to find some of the best configurations of several machine learning algorithms intending to solve the automatic bug assignment problem. These four steps are respectively used to select a combination of pre-processing techniques, a bug report representation, a potential feature selection technique and to tune several classifiers. The aforementioned method has been applied on three software projects: 66 066 bug reports of a proprietary project, 24 450 bug reports of Eclipse JDT and 30 358 bug reports of Mozilla Firefox. 619 configurations have been applied and compared on each of these three projects. In production, using the approach introduced in this work on the bug reports of the proprietary project would have increased the accuracy by up to 16.64 percentage points.

[36] Fredrik Präntare. 2017.
Simultaneous coalition formation and task assignment in a real-time strategy game.
Student Thesis. 65 pages. ISRN: LIU-IDA/LITH-EX-A--17/032--SE.

In this thesis we present an algorithm that is designed to improve the collaborative capabilities of agents that operate in real-time multi-agent systems. Furthermore, we study the coalition formation and task assignment problems in the context of real-time strategy games. More specifically, we design and present a novel anytime algorithm for multi-agent cooperation that efficiently solves the simultaneous coalition formation and assignment problem, in which disjoint coalitions are formed and assigned to independent tasks simultaneously. This problem, that we denote the problem of collaboration formation, is a combinatorial optimization problem that has many real-world applications, including assigning disjoint groups of workers to regions or tasks, and forming cross-functional teams aimed at solving specific problems.The algorithm's performance is evaluated using randomized artificial problems sets of varying complexity and distribution, and also using Europa Universalis 4 – a commercial strategy game in which agents need to cooperate in order to effectively achieve their goals. The agents in such games are expected to decide on actions in real-time, and it is a difficult task to coordinate them. Our algorithm, however, solves the coordination problem in a structured manner.The results from the artificial problem sets demonstrates that our algorithm efficiently solves the problem of collaboration formation, and does so by automatically discarding suboptimal parts of the search space. For instance, in the easiest artificial problem sets with 12 agents and 8 tasks, our algorithm managed to find optimal solutions after only evaluating approximately 0.000003% of the possible solutions. In the hardest of the problem sets with 12 agents and 8 tasks, our algorithm managed to find a 80% efficient solution after only evaluating approximately 0.000006% of the possible solutions.

[35] Full text  Tova Linder and Ola Jigin. 2017.
Organ Detection and Localization in Radiological Image Volumes.
Student Thesis. 88 pages. ISRN: LIU-IDA/LITH-EX-A--17/024--SE.

Using Convolutional Neural Networks for classification of images and for localization and detection of objects in images is becoming increasingly popular. Within radiology a huge amount of image data is produced and meta data containing information of what the images depict is currently added manually by a radiologist. To aid in streamlining physician’s workflow this study has investigated the possibility to use Convolutional Neural Networks (CNNs) that are pre-trained on natural images to automatically detect the presence and location of multiple organs and body-parts in medical CT images. The results show promise for multiclass classification with an average precision 89.41% and average recall 86.40%. This also confirms that a CNN that is pre-trained on natural images can be succesfully transferred to solve a different task. It was also found that adding additional data to the dataset does not necessarily result in increased precision and recall or decreased error rate. It is rather the type of data and used preprocessing techniques that matter.

[34] Full text  Elena Moral López. 2017.
Muting pattern strategy for positioning in cellular networks.
Student Thesis. 65 pages. ISRN: LIU-IDA/LITH-EX-A--17/018--SE.

Location Based Services (LBS) calculate the position of the user for different purposes like advertising and navigation. Most importantly, these services are also used to help emergency services by calculating the position of the person that places the emergency phone call. This has introduced a number of requirements on the accuracy of the measurements of the position. Observed Time Difference of Arrival (OTDOA) is the method used to estimate the position of the user due to its high accuracy. Nevertheless, this method relies on the correct reception of so called positioning signals, and therefore the calculations can suffer from errors due to interference between the signals. To lower the probability of interference, muting patterns can be used. These methods can selectively mute certain signals to increase the signal to interference and noise ratio (SINR) of others and therefore the number of signals detected. In this thesis, a simulation environment for the comparison of the different muting patterns has been developed. The already existing muting patterns have been simulated and compared in terms of number of detected nodes and SINR values achieved. A new muting pattern has been proposed and compared to the others. The results obtained have been presented and an initial conclusion on which of the muting patterns offers the best performance has been drawn.

[33] Petra Öhlin. 2017.
Prioritizing Tests with Spotify?s Test & Build Data using History-based, Modification-based & Machine Learning Approaches.
Student Thesis. 43 pages. ISRN: LIU-IDA/LITH-EX-A--2017/021--SE.

This thesis intends to determine the extent to which machine learning can be used to solve the regression test prioritization (RTP) problem. RTP is used to order tests with respect to probability of failure. This will optimize for a fast failure, which is desirable if a test suite takes a long time to run or uses a significant amount of computational resources. A common machine learning task is to predict probabilities; this makes RTP an interesting application of machine learning. A supervised learning method is investigated to train a model to predict probabilities of failure, given a test case and a code change. The features investigated are chosen based on previous research of history- based and modification-based RTP. The main motivation for looking at these research areas is that they resemble the data provided by Spotify. The result of the report shows that it is possible to improve how tests run with RTP using machine learning. Nevertheless, a much simpler history- based approach is the best performing approach. It is looking at the history of test results, the more failures recorded for the test case over time, the higher priority it gets. Less is sometimes more.

[32] Full text  Joakim Gylling. 2017.
Transition-Based Dependency Parsing with Neural Networks.
Student Thesis. 10 pages. ISRN: LIU-IDA/LITH-EX-G--17/011--SE.

Dependency parsing is important in contemporary speech and language processing systems. Current dependency parsers typically use the multi-class perceptron machine learning component, which classifies based on millions of sparse indicator features, making developing and maintaining these systems expensive and error-prone. This thesis aims to explore whether replacing the multi-class perceptron component with an artificial neural network component can alleviate this problem without hurting performance, in terms of accuracy and efficiency. A simple transition-based dependency parser using the artificial neural network (ANN) as the classifier is written in Python3 and the same program with the classifier replaced by a multi-class perceptron component is used as a baseline. The results show that the ANN dependency parser provides slightly better unlabeled attachment score with only the most basic atomic features, eliminating the need for complex feature engineering. However, it is about three times slower and the training time required for the ANN is significantly longer.

[31] Full text  Maximilian Bragazzi Ihrén and Henrik Ingbrant Björs. 2017.
Visualizing atmospheric data on a mobile platform.
Student Thesis. 10 pages. ISRN: LIU-IDA/LITH-EX-G--17/010--SE.

Weather data is important for almost everyone today. Thedaily weather report, home thermometers, and a lot of otherthings affect our every day life. In order to develop betterand more efficient equipment, tools and algorithms, thepeople working with this data need to be able to access it inan easily accessible and easy to read format. In thisresearch, methods of visualizing data on mobile platformsare evaluated based on what researchers in the field wants,since their respective fields might want to use very specificvisualizations. The implementability of these visualizationsare then evaluated, based on the implementations madethroughout this paper. The results show that the researchersknow what they want, and that what they want isimplementable on mobile platforms given some limitationscaused by performance.

[30] Full text  Fredrik Jonsén and Alexander Stolpe. 2017.
The feasibility and practicality of a generic social media library.
Student Thesis. 8 pages. ISRN: LIU-IDA/LITH-EX-G--17/009--SE.

Many people today use social media in one way or another, and many of these platforms have released APIs developers can use to integrate social media in their applications. As many of these platforms share a lot of functionality we see a need for developing a library, to contain these, and ease the development process when working with the platforms. The purpose of this paper is to find common functionality and explore the possibility of generalization in this regard. We first look for common denominators between the top social media networks, and using this information we attempt to make an implementation to evaluate the practicality. After the development process we analyze our findings and discuss the usability and maintainability of such a library. Our findings show that the current state of the studied APIs are not suitable for generalization.

2016
[29] Full text  Marcus Johansson. 2016.
Online Whole-Body Control using Hierarchical Quadratic Programming: Implementation and Evaluation of the HiQP Control Framework.
Student Thesis. 76 pages. ISRN: LIU-IDA/LITH-EX-A--16/056--SE.

The application of local optimal control is a promising paradigm for manipulative robot motion generation.In practice this involves instantaneous formulations of convex optimization problems depending on the current joint configuration of the robot and the environment.To be effective, however, constraints have to be carefully constructed as this kind of motion generation approach has a trade-off of completeness.Local optimal solvers, which are greedy in a temporal sense, have proven to be significantly more effective computationally than classical grid-based or sampling-based planning approaches.In this thesis we investigate how a local optimal control approach, namely the task function approach, can be implemented to grant high usability, extendibility and effectivity.This has resulted in the HiQP control framework, which is compatible with ROS, written in C++.The framework supports geometric primitives to aid in task customization by the user.It is also modular as to what communication system it is being used with, and to what optimization library it uses for finding optimal controls.We have evaluated the software quality of the framework according to common quantitative methods found in the literature.We have also evaluated an approach to perform tasks using minimal jerk motion generation with promising results.The framework also provides simple translation and rotation tasks based on six rudimentary geometric primitives.Also, task definitions for specific joint position setting, and velocity limitations were implemented.

[28] Full text  Tomas Melin. 2016.
Implementation and Evaluation of a Continuous Code Inspection Platform.
Student Thesis. 100 pages. ISRN: LIU-IDA/LITH-EX-A--16/047—SE.

Establishing and preserving a high level of software quality is a not a trivial task, although the benefits of succeeding with this task has been proven profitable and advantageous. An approach to mitigate the decreasing quality of a project is to track metrics and certain properties of the project, in order to view the progression of the project’s properties. This approach may be carried out by introducing continuous code inspection with the application of static code analysis. However, as the initial common opinion is that these type of tools produce a too high number of false positives, there is a need to investigate what the actual case is. This is the origin for the investigation and case study performed in this paper. The case study is performed at Ida Infront AB in Linköping, Sweden and involves interviews with developers to determine the performance of the continuous inspection platform SonarQube, in addition to examine the general opinion among developers at the company. The author executes the implementation and configuration of a continuous inspection environment to analyze a partition of the company’s product and determine what rules that are appropriate to apply in the company’s context. The results from the investigation indicate the high quality and accuracy of the tool, in addition to the advantageous functionality of continuously monitoring the code to observe trends and the progression of metrics such as cyclomatic complexity and duplicated code, with the goal of preventing the constant increase of complex and duplicated code. Combining this with features such as false positive suppression, instant analysis feedback in pull requests and the possibility to break the build given specified conditions, suggests that the implemented environment is a way to mitigate software quality difficulties.

[27] Full text  Martin Estgren. 2016.
Lightweight User Agents.
Student Thesis. 36 pages. ISRN: LIU-IDA/LITH-EX-G--16/036--SE.

The unit for information security and IT architecture at The Swedish Defence Research Agency (FOI) conducts work with a cyber range called CRATE (Cyber Range and Training Environment). Currently, simulation of user activity involves scripts inside the simulated network. This solution is not ideal because of the traces it leaves in the system and the general lack of standardised GUI API between different operating systems. FOI are interested in testing the use of artificial user agent located outside the virtual environment using computer vision and the virtualisation API to execute actions and extract information from the system.This paper focuses on analysing the reliability of template matching, a computer vision algorithm used to localise objects in images using already identified images of said object as templates. The analysis will evaluate both the reliability of localising objects and the algorithms ability to correctly identify if an object is present in the virtual environment.Analysis of template matching is performed by first creating a prototype of the agent's sensory system and then simulate scenarios which the agent might encounter. By simulating the environment, testing parameters can be manipulated and monitored in a reliable way. The parameters manipulated involves both the amount and type of image noise in the template and screenshot, the agent’s discrimination threshold for what constitutes a positive match, and information about the template such as template generality.This paper presents the performance and reliability of the agent in regards to what type of image noise affects the result, the amount of correctly identified objects given different discrimination thresholds, and computational time of template matching when different image filters are applied. Furthermore the best cases for each study are presented as comparison for the other results.In the end of the thesis we present how for screenshots with objects very similar to the templates used by the agent, template matching can result in a high degree of accuracy in both object localization and object identification and that a small reduction of similarity between template and screenshot to reduce the agent's ability to reliably identifying specific objects in the environment.

[26] Full text  Rasmus Holm. 2016.
Cluster Analysis of Discussions on Internet Forums.
Student Thesis. 62 pages. ISRN: LIU-IDA/LITH-EX-G--16/037—SE.

The growth of textual content on internet forums over the last decade have been immense which have resulted in users struggling to find relevant information in a convenient and quick way.The activity of finding information from large data collections is known as information retrieval and many tools and techniques have been developed to tackle common problems. Cluster analysis is a technique for grouping similar objects into smaller groups (clusters) such that the objects within a cluster are more similar than objects between clusters.We have investigated the clustering algorithms, Graclus and Non-Exhaustive Overlapping <em>k</em>-means (NEO-<em>k</em>-means), on textual data taken from Reddit, a social network service. One of the difficulties with the aforementioned algorithms is that both have an input parameter controlling how many clusters to find. We have used a greedy modularity maximization algorithm in order to estimate the number of clusters that exist in discussion threads.We have shown that it is possible to find subtopics within discussions and that in terms of execution time, Graclus has a clear advantage over NEO-<em>k</em>-means.

[25] Full text  Erik Hansson. 2016.
Search guidance with composite actions: Increasing the understandability of the domain model.
Student Thesis. 98 pages. ISRN: LIU-IDA/LITH-EX--16/043--SE.

This report presents an extension to the domain definition language for Threaded Forward-chaining Partial Order Planner (TFPOP) that can be used to increase the understandability of domain models. The extension consists of composite actions which is a method for expressing abstract actions as procedures of primitive actions. TFPOP can then uses these abstract actions when searching for a plan. An experiment, with students as participants, was used to show that using composite action can increase the understandability for non-expert users. Moreover, it was also proved the planner can utilize the composite action to significantly decrease the search time. Furthermore, indications was found that using composite actions is equally fast in terms of search time as using existing equivalent methods to decrease the search time.

[24] Full text  Anna Boyer de la Giroday. 2016.
Automatic fine tuning of cavity filters.
Student Thesis. 49 pages. ISRN: LIU-IDA/LITH-EX-A--16/036--SE.

Cavity filters are a necessary component in base stations used for telecommunication. Without these filters it would not be possible for base stations to send and receive signals at the same time. Today these cavity filters require fine tuning by humans before they can be deployed. This thesis have designed and implemented a neural network that can tune cavity filters. Different types of design parameters have been evaluated, such as neural network architecture, data presentation and data preprocessing. While the results was not comparable to human fine tuning, it was shown that there was a relationship between error and number of weights in the neural network. The thesis also presents some rules of thumb for future designs of neural network used for filter tuning.

2015
[23] Full text  Stefan Bränd. 2015.
Using Rigid Landmarks to Infer Inter-Temporal Spatial Relations in Spatio-Temporal Reasoning.
Student Thesis. 32 pages. ISRN: LIU-IDA/LITH-EX-G--15/074--SE.

Spatio-temporal reasoning is the area of automated reasoning about space and time and is important in the field of robotics. It is desirable for an autonomous robot to have the ability to reason about both time and space. ST0 is a logic that allows for such reasoning by, among other things, defining a formalism used to describe the relationship between spatial regions and a calculus that allows for deducing further information regarding such spatial relations. An extension of ST0 is ST1 that can be used to describe the relationship between spatial entities across time-points (inter-temporal relations) while ST0 is constrained to doing so within a single time-point. This allows for a better ability of expressing how spatial entities change over time. A major obstacle in using ST1 in practise however, is the fact that any observations made regarding spatial relations between regions is constrained to the time-point in which the observation was made, so we are unable to observe inter-temporal relations. Further complicating things is the fact that deducing such inter-temporal relations is not possible without a frame of reference. This thesis examines one method of overcoming these problems by considering the concept of rigid regions which are assumed to always be unchanging and using them as the frame of reference, or as landmarks. The effectiveness of this method is studied by conducting experiments where a comparison is made between various landmark ratios with respect to the total number of regions under consideration. Results show that when a high degree of intra-temporal relations are fully or partially known, increasing the number of landmark regions will reduce the percentage of inter-temporal relations to be completely unknown. Despite this, very few inter-temporal relations can be fully determined even with a high ratio of landmark regions.

[22] Full text  Valberg Joakim. 2015.
Document Separation in Digital Mailrooms.
Student Thesis. 47 pages. ISRN: LIU-IDA/LITH-EX-A-15/056-SE.

The growing mail volumes for businesses worldwide is one reason why theyare increasingly turning to digital mailrooms. A digital mailroom automaticallymanages the incoming mails, and a vital technology to its success isdocument classication. A problem with digital mailrooms and the documentclassication is separating the input stream of pages into documents.This thesis investigates existing classication theory and applies it to createan algorithm which solves the document separation problem. This algorithmis evaluated and compared against an existing algorithmic solution, over adataset containing real invoices.

[21] Full text  Erik Sommarström. 2015.
I am the Greatest Driver in the World!: -Does self-awareness of driving ability affect traffic safety behaviour?.
Student Thesis. 43 pages. ISRN: LIU-IDA/KOGVET-A--15/008—SE.

This simulator study aims to investigate if there is a relationship between self-awareness of driving ability and traffic safety behaviour. Self-awareness in this study is accurate self-evaluation of one’s abilities. By letting 97 participants (55-75 years old) drive the simulator and answering the Driver Skill Inventory (DSI; Warner et al., 2013) as well as the Multidimensional locus of control (T-loc; Özkan &amp; Lajunen, 2005). A measure of self-awareness was computed using the residuals from regression line. Furthermore, this measure could show if a participant over-estimated or under-estimated their ability. Four self-awareness measures were made. The self-awareness measures were compared to traffic safety behaviour. Three different traffic safety measures were computed using specific events in the simulator scenario. The self-awareness measures were grouped into three groups; under-estimators, good self-awareness and over-estimators. These groups were then compared to each other with respect to traffic safety. A multivariate ANOVA was made to test for differences between the self-awareness groups but no significant main difference was found. The results showed no difference in traffic safety behaviour given the different levels of self-awareness. Furthermore, this could be a result of the old age of the sample group as self-awareness may only be relevant in a learning context. The conclusion of the study is that the analysis shows that there is no difference between over-estimators and under-estimators of driving ability, at least not in experienced older drivers.

[20] Tina Danielsson. 2015.
Portering frćn Google Apps REST API till Microsoft Office 365 REST API.
Student Thesis. 10 pages. ISRN: LiTH-IDA/ERASMUS-G--15/003--SE.

Stress pÄ arbetsplatsen relaterat till mÄnga inkommande och utgÄende kommunikationskanaler Àr ett reellt problem. Applikationer som samlar alla kanaler i samma verktyg kan hjÀlpa till pÄ det hÀr omrÄdet. För att förenkla vid utveckling av en sÄdan applikation kan ett modulÀrt system skapas, dÀr varje modul ser liknande ut och enkelt kan kopplas in i en huvudapplikation. Den hÀr studien undersöker de problem som kan uppstÄ nÀr flera tjÀnster ska integreras, mer specifikt genom att titta pÄ hur en befintlig modul för e-post via Google Apps kan porteras för att stödja e-post via Microsoft Office 365. Arbetet har skett enligt metoder för testdriven portering och varje steg i porteringen har dokumenterats noggrant. Ett antal problemomrÄden har identifierats och möjliga lösningar föreslÄs. UtfrÄn de problem som uppstÄtt dras slutsatsen att de Àr av en sÄdan karaktÀr att de inte utgör nÄgot hinder för en portering.

[19] Full text  Jonas Hietala. 2015.
A Comparison of Katz-eig and Link-analysis for Implicit Feedback Recommender Systems.
Student Thesis. 85 pages. ISRN: LIU-IDA/LITH-EX-A--15/026--SE.

Link: http://www.jonashietala.se/masters_thesi...

Recommendations are becoming more and more important in a world where there is an abundance of possible choices and e-commerce and content providers are featuring recommendations prominently. Recommendations based on explicit feedback, where user is giving feedback for example with ratings, has been a popular research subject. Implicit feedback recommender systems which passively collects information about the users is an area growing in interest. It makes it possible to generate recommendations based purely from a user's interactions history without requiring any explicit input from the users, which is commercially useful for a wide area of businesses. This thesis builds a recommender system based on implicit feedback using the recommendation algorithms katz-eig and link-analysis and analyzes and implements strategies for learning optimized parameters for different datasets. The resulting system forms the foundation for Comordo Technologies' commercial recommender system.

[18] Full text  Patrik Bergström. 2015.
Automated Setup of Display Protocols.
Student Thesis. 44 pages. ISRN: LIU-IDA/LITH-EX-A--15/014--SE.

Link: http://urn.kb.se/resolve?urn=urn:nbn:se:...

Radiologists' workload has been steadily increasing for decades. As digital technology matures it improves the workflow for radiology departments and decreases the time necessary to examine patients. Computer systems are widely used in health care and are for example used to view radiology images. To simplify this, display protocols based on examination data are used to automatically create a layout and hang images for the user. To cover a wide variety of examinations hundreds of protocols must be created, which is a time-consuming task and the system can still fail to hang series if strict requirements on the protocols are not met. To remove the need for this manual step we propose to use machine learning based on past manually corrected presentations. The classifiers are trained on the metadata in the examination and how the radiologist preferred to hang the series. The chosen approach was to create classifiers for different layout rules and then use these predictions in an algorithm for assigning series types to individual image slots according to categories based on metadata, similar to how display protocol works. The resulting presentations shows that the system is able to learn, but must increase its prediction accuracy if it is to be used commercially. Analyses of the different parts show that increased accuracy in early steps should improve overall success.

[17] Full text  Karl Nygren. 2015.
Trust Logics and Their Horn Fragments: Formalizing Socio-Cognitive Aspects of Trust.
Student Thesis. 93 pages. ISRN: LiTH-MAT-EX--2015/01--SE.

This thesis investigates logical formalizations of Castelfranchi and Falcone's (C&amp;F) theory of trust [9, 10, 11, 12]. The C&amp;F theory of trust defines trust as an essentially mental notion, making the theory particularly well suited for formalizations in multi-modal logics of beliefs, goals, intentions, actions, and time.Three different multi-modal logical formalisms intended for multi-agent systems are compared and evaluated along two lines of inquiry. First, I propose formal definitions of key concepts of the C&amp;F theory of trust and prove some important properties of these definitions. The proven properties are then compared to the informal characterisation of the C&amp;F theory. Second, the logics are used to formalize a case study involving an Internet forum, and their performances in the case study constitute grounds for a comparison. The comparison indicates that an accurate modelling of time, and the interaction of time and goals in particular, is integral for formal reasoning about trust.Finally, I propose a Horn fragment of the logic of Herzig, Lorini, Hubner, and Vercouter [25]. The Horn fragment is shown to be too restrictive to accurately express the considered case study.

2014
[16] Full text  Mattias Tiger. 2014.
Unsupervised Spatio-Temporal Activity Learning and Recognition in a Stream Processing Framework.
Student Thesis. 103 pages. ISRN: LIU-IDA/LITH-EX-A--14/059--SE.

Learning to recognize and predict common activities, performed by objects and observed by sensors, is an important and challenging problem related both to artificial intelligence and robotics.In this thesis, the general problem of dynamic adaptive situation awareness is considered and we argue for the need for an on-line bottom-up approach.A candidate for a bottom layer is proposed, which we consider to be capable of future extensions that can bring us closer towards the goal.We present a novel approach to adaptive activity learning, where a mapping between raw data and primitive activity concepts are learned and continuously improved on-line and unsupervised. The approach takes streams of observations of objects as input and learns a probabilistic representation of both the observed spatio-temporal activities and their causal relations. The dynamics of the activities are modeled using sparse Gaussian processes and their causal relations using probabilistic graphs.The learned model supports both estimating the most likely activity and predicting the most likely future (and past) activities. Methods and ideas from a wide range of previous work are combined to provide a uniform and efficient way to handle a variety of common problems related to learning, classifying and predicting activities.The framework is evaluated both by learning activities in a simulated traffic monitoring application and by learning the flight patterns of an internally developed autonomous quadcopter system. The conclusion is that our framework is capable of learning the observed activities in real-time with good accuracy.We see this work as a step towards unsupervised learning of activities for robotic systems to adapt to new circumstances autonomously and to learn new activities on the fly that can be detected and predicted immediately.

[15] Full text  Per Jonsson. 2014.
Design och implementation av webbenkäter: kvalitet, svarsfrekvens och underhćll.
Student Thesis. 10 pages. ISRN: LIU-IDA/LITH-EX-G--14/049--SE.

En webbapplikation för analys och administration av webbenkÀter har designats och implementerats. Dess syfte Àr att maximera svarskvalitet och svarsfrekvens samt att vara underhÄllbar. Uppdragsgivaren Ericsson Linköping har utfÀrdat kravspecifikationen för applikationen. HÀnsyn har tagits till aspekterna webbenkÀtdesign och under-hÄllbarhet av mjukvara. UnderhÄllbarhetsmodeller för mjukvara med tillhörande metriker, samt designmodeller och rekommendationer för webbenkÀter har studerats. Arbetets bidrag till dessa omrÄden Àr en praktisk modell som tillÀmpar rÄdande forskning, i form av en webbapplikation. Applikationen har testats mot modeller och rekommendationer för underhÄllbarhet och enkÀtdesign. Applikationen uppvisar hög grad av analyserbarhet, förÀndringsbarhet och testbarhet, men inte stabilitet. Effekten av enkÀtdesignen har inte utvÀrderats. Modellen för underhÄllbarhet kan klarlÀgga orsak och verkan i mjukvarusystem och bidra till utveckling av programvara med hög kvalitet.

[14] Full text  Anders Wikström. 2014.
Resource allocation of drones flown in a simulated environment.
Student Thesis. 24 pages. ISRN: LIU-IDA/LITH-EX-G--14/003—SE.

In this report we compare three different assignment algorithms in how they can be used to assign a set of drones to get to a set of goal locations in an as resource efficient way as possible. An experiment is set up to compare how these algorithms perform in a somewhat realistic simulated environment. The Robot Operating system (ROS) is used to create the experimental environment. We found that by introducing a threshold for the Hungarian algorithm we could reduce the total time it takes to complete the problem while only sightly increasing total distance traversed by the drones.

2013
[13] Jakob Pogulis. 2013.
Testramverk för distribuerade system.
Student Thesis. 46 pages. ISRN: LIU-IDA/LITH-EX-G--13/010--SE.

When developing software that is meant to be distributed over several different computers and several different networks while still working together against a common goal there is a challenge in testing how updates within a single component will affect the system as a whole. Even if the performance of that specific component increases that is no guarantee for the increased performance of the entire system. Traditional methods of testing software becomes both hard and tedious when several different machines has to be involved for a single test and all of those machines has to be synchronized as well.This thesis has resulted in an exemplary application suite for testing distributed software. The thesis describes the method used for implementation as well as a description of the actual application suite that was developed. During the development several important factors and improvements for such a system was identified, which are described at the end of the thesis even though some of them never made it into the actual implementation. The implemented application suite could be used as a base when developing a more complete system in order to distribute tests and applications that has to run in a synchronized manner with the ability to report the results of each individual component.

[12] Full text  Christopher Bergdahl. 2013.
Modeling Air Combat with Influence Diagrams.
Student Thesis. 64 pages. ISRN: LIU-IDA/LITH-EX-A--13/031--SE.

Air combat is a complex situation, training for it and analysis of possible tactics are time consuming and expensive. In order to circumvent those problems, mathematical models of air combat can be used. This thesis presents air combat as a one-on-one influence diagram game where the influence diagram allows the dynamics of the aircraft, the preferences of the pilots and the uncertainty of decision making in a structural and transparent way to be taken into account. To obtain the players’ game optimal control sequence with respect to their preferences, the influence diagram has to be solved. This is done by truncating the diagram with a moving horizon technique and determining and implementing the optimal controls for a dynamic game which only lasts a few time steps.The result is a working air combat model, where a player estimates the probability that it resides in any of four possible states. The pilot’s preferences are modeled by utility functions, one for each possible state. In each time step, the players are maximizing the cumulative sum of the utilities for each state which each possible action gives. These are weighted with the corresponding probabilities. The model is demonstrated and evaluated in a few interesting aspects. The presented model offers a way of analyzing air combat tactics and maneuvering as well as a way of making autonomous decisions in for example air combat simulators.

[11] Full text  Johan Fredborg. 2013.
Spam filter for SMS-traffic.
Student Thesis. 82 pages. ISRN: LIU-IDA/LITH-EX-A--13/021-SE.

Communication through text messaging, SMS (Short Message Service), is nowadays a huge industry with billions of active users. Because of the huge userbase it has attracted many companies trying to market themselves through unsolicited messages in this medium in the same way as was previously done through email. This is such a common phenomenon that SMS spam has now become a plague in many countries.This report evaluates several established machine learning algorithms to see how well they can be applied to the problem of filtering unsolicited SMS messages. Each filter is mainly evaluated by analyzing the accuracy of the filters on stored message data. The report also discusses and compares requirements for hardware versus performance measured by how many messages that can be evaluated in a fixed amount of time.The results from the evaluation shows that a decision tree filter is the best choice of the filters evaluated. It has the highest accuracy as well as a high enough process rate of messages to be applicable. The decision tree filter which was found to be the most suitable for the task in this environment has been implemented. The accuracy in this new implementation is shown to be as high as the implementation used for the evaluation of this filter.Though the decision tree filter is shown to be the best choice of the filters evaluated it turned out the accuracy is not high enough to meet the specified requirements. It however shows promising results for further testing in this area by using improved methods on the best performing algorithms.

2012
[10] Full text  Jonatan Olofsson. 2012.
Towards Autonomous Landing of a Quadrotorusing Monocular SLAM Techniques.
Student Thesis. 102 pages. ISRN: LIU-IDA/LITH-EX-A--12/026--SE.

Use of Unmanned Aerial Vehicles have seen enormous growth in recent years due to the advances in related scientific and technological fields. This fact combined with decreasing costs of using UAVs enables their use in new application areas. Many of these areas are suitable for miniature scale UAVs - Micro Air Vehicles(MAV) - which have the added advantage of portability and ease of deployment. One of the main functionalities necessary for successful MAV deployment in real-world applications is autonomous landing. Landing puts particularly high requirements on positioning accuracy, especially in indoor confined environments where the common global positioning technology is unavailable. For that reason using an additional sensor, such as a camera, is beneficial. In this thesis, a set of technologies for achieving autonomous landing is developed and evaluated. In particular, state estimation based on monocular vision SLAM techniques is fused with data from onboard sensors. This is then used as the basis for nonlinear adaptive control as well trajectory generation for a simple landing procedure. These components are connected using a new proposed framework for robotic development. The proposed system has been fully implemented and tested in a simulated environment and validated using recorded data. Basic autonomous landing was performed in simulation and the result suggests that the proposed system is a viable solution for achieving a fully autonomous landing of a quadrotor.

[9] Full text  Anders Hongslo. 2012.
Stream Processing in the Robot Operating System framework.
Student Thesis. 79 pages. ISRN: LIU-IDA/LITH-EX-A--12/030--SE.

Streams of information rather than static databases are becoming increasingly important with the rapid changes involved in a number of fields such as finance, social media and robotics. DyKnow is a stream-based knowledge processing middleware which has been used in autonomous Unmanned Aerial Vehicle (UAV) research. ROS (Robot Operating System) is an open-source robotics framework providing hardware abstraction, device drivers, communication infrastructure, tools, libraries as well as other functionalities.This thesis describes a design and a realization of stream processing in ROS based on the stream-based knowledge processing middleware DyKnow. It describes how relevant information in ROS can be selected, labeled, merged and synchronized to provide streams of states. There are a lot of applications for such stream processing such as execution monitoring or evaluating metric temporal logic formulas through progression over state sequences containing the features of the formulas. Overviews are given of DyKnow and ROS before comparing the two and describing the design. The stream processing capabilities implemented in ROS are demonstrated through performance evaluations which show that such stream processing is fast and efficient. The resulting realization in ROS is also readily extensible to provide further stream processing functionality.

[8] Full text  Viet Ha Nguyen. 2012.
Design Space Exploration of the Quality of Service for Stream Reasoning Applications.
Student Thesis. 35 pages. ISRN: LIU-IDA/LITH-EX-A--12/027--SE.

An Unmanned Aerial Vehicle (UAV) is often an aircraft with no crew that can fly independently by a preprogrammed plan, or by remote control. Several UAV applications, like autonomously surveillance and traffic monitoring, are real-time applications. Hence tasks in these applications must complete within specied deadlines.Real Time Calculus (RTC) is a formal framework for reasoning about realtime systems and in particular streaming applications. RTC has its mathematical roots in Network Calculus. It supports timing analysis, estimating loads and predicting memory requirements.In this thesis, a formal analysis of real-time stream reasoning for UAV applications is conducted. The performance analysis is based on RTC using an abstract performance model of the streaming reasoning on board a UAV. In this study, we consider two dierent scheduling methods, first-in-first-out (FIFO) and fixed priority (FP). In the FIFO scheduling model the priorities of the tasks are assigned and processed based on the order of their arrival, while in the FP scheduling model the priorities of the tasks are preassigned. The Quality of Service (QoS) of these applications is calculated and analyzed in a proposed design space exploration framework.QoS can be defined dierently depending on what field we are studying and in this thesis we are interested in studying the delays of the real-time stream reasoning applications when (i) we fix jitters and number of instances and vary the periods, (ii) we fix the periods and number of instances and vary the jitters, and (iii) we fix the periods, jitters and vary the number of instances.

[7] Full text  Daniel Lazarovski. 2012.
Extending the Stream Reasoning in DyKnow with Spatial Reasoning in RCC-8.
Student Thesis. 74 pages. ISRN: LIU-IDA/LITH-EX-A--12/008--SE.

Autonomous systems require a lot of information about the environment in which they operate in order to perform different high-level tasks. The information is made available through various sources, such as remote and on-board sensors, databases, GIS, the Internet, etc. The sensory input especially is incrementally available to the systems and can be represented as streams. High-level tasks often require some sort of reasoning over the input data, however raw streaming input is often not suitable for the higher level representations needed for reasoning. DyKnow is a stream processing framework that provides functionalities to represent knowledge needed for reasoning from streaming inputs. DyKnow has been used within a platform for task planning and execution monitoring for UAVs. The execution monitoring is performed using formula progression with monitor rules specified as temporal logic formulas. In this thesis we present an analysis for providing spatio-temporal functionalities to the formula progressor and we extend the formula progression with spatial reasoning in RCC-8. The result implementation is capable of evaluating spatio-temporal logic formulas using progression over streaming data. In addition, a ROS implementation of the formula progressor is presented as a part of a spatio-temporal stream reasoning architecture in ROS.

2011
[6] Full text  Zlatan Dragisic. 2011.
Semantic Matching for Stream Reasoning.
Student Thesis. 110 pages. ISRN: LIU-IDA/LITH-EX-A--11/041--SE.

Autonomous system needs to do a great deal of reasoning during execution in order to provide timely reactions to changes in their environment. Data needed for this reasoning process is often provided through a number of sensors. One approach for this kind of reasoning is evaluation of temporal logical formulas through progression. To evaluate these formulas it is necessary to provide relevant data for each symbol in a formula. Mapping relevant data to symbols in a formula could be done manually, however as systems become more complex it is harder for a designer to explicitly state and maintain thismapping. Therefore, automatic support for mapping data from sensors to symbols would make system more flexible and easier to maintain.DyKnow is a knowledge processing middleware which provides the support for processing data on different levels of abstractions. The output from the processing components in DyKnow is in the form of streams of information. In the case of DyKnow, reasoning over incrementally available data is done by progressing metric temporal logical formulas. A logical formula contains a number of symbols whose values over time must be collected and synchronized in order to determine the truth value of the formula. Mapping symbols in formula to relevant streams is done manually in DyKnow. The purpose of this matching is for each variable to find one or more streams whose content matches the intended meaning of the variable.This thesis analyses and provides a solution to the process of semantic matching. The analysis is mostly focused on how the existing semantic technologies such as ontologies can be used in this process. The thesis also analyses how this process can be used for matching symbols in a formula to content of streams on distributed and heterogeneous platforms. Finally, the thesis presents an implementation in the Robot Operating System (ROS). The implementation is tested in two case studies which cover a scenario where there is only a single platform in a system and a scenario where there are multiple distributed heterogeneous platforms in a system.The conclusions are that the semantic matching represents an important step towards fully automatized semantic-based stream reasoning. Our solution also shows that semantic technologies are suitable for establishing machine-readable domain models. The use of these technologies made the semantic matching domain and platform independent as all domain and platform specific knowledge is specified in ontologies. Moreover, semantic technologies provide support for integration of data from heterogeneous sources which makes it possible for platforms to use streams from distributed sources.

[5] Full text  Marjan Alirezaie. 2011.
Semantic Analysis Of Multi Meaning Words Using Machine Learning And Knowledge Representation.
Student Thesis. 74 pages. ISRN: LiU/IDA-EX-A- -11/011- -SE.

The present thesis addresses machine learning in a domain of naturallanguage phrases that are names of universities. It describes two approaches to this problem and a software implementation that has made it possible to evaluate them and to compare them.In general terms, the system's task is to learn to 'understand' the significance of the various components of a university name, such as the city or region where the university is located, the scienti c disciplines that are studied there, or the name of a famous person which may be part of the university name. A concrete test for whether the system has acquired this understanding is when it is able to compose a plausible university name given some components that should occur in the name.In order to achieve this capability, our system learns the structure of available names of some universities in a given data set, i.e. it acquires a grammar for the microlanguage of university names. One of the challenges is that the system may encounter ambiguities due to multi meaning words. This problem is addressed using a small ontology that is created during the training phase.Both domain knowledge and grammatical knowledge is represented using decision trees, which is an ecient method for concept learning. Besides for inductive inference, their role is to partition the data set into a hierarchical structure which is used for resolving ambiguities.The present report also de nes some modi cations in the de nitions of parameters, for example a parameter for entropy, which enable the system to deal with cognitive uncertainties. Our method for automatic syntax acquisition, ADIOS, is an unsupervised learning method. This method is described and discussed here, including a report on the outcome of the tests using our data set.The software that has been implemented and used in this project has been implemented in C.

2010
[4] Full text  Fredrik Ćslin. 2010.
Evaluation of Hierarchical Temporal Memory in algorithmic trading.
Student Thesis. 32 pages. ISRN: LIU-IDA/LITH-EX-G--10/005--SE.

This thesis looks into how one could use Hierarchal Temporal Memory (HTM) networks to generate models that could be used as trading algorithms. The thesis begins with a brief introduction to algorithmic trading and commonly used concepts when developing trading algorithms. The thesis then proceeds to explain what an HTM is and how it works. To explore whether an HTM could be used to generate models that could be used as trading algorithms, the thesis conducts a series of experiments. The goal of the experiments is to iteratively optimize the settings for an HTM and try to generate a model that when used as a trading algorithm would have more profitable trades than losing trades. The setup of the experiments is to train an HTM to predict if it is a good time to buy some shares in a security and hold them for a fixed time before selling them again. A fair amount of the models generated during the experiments was profitable on data the model have never seen before, therefore the author concludes that it is possible to train an HTM so it can be used as a profitable trading algorithm.

2009
[3] Mikael Nilsson. 2009.
Spanneröar och spannervägar.
Student Thesis. 126 pages. ISRN: -.

In this Master Thesis the possibility to efficiently divide a graph into spanner islands is examined. Spanner islands are islands of the graph that fulfill the spanner condition, that the distance between two nodes via the edges in the graph cannot be too far, regulated by the stretch constant, compared to the Euclidian distance between them. In the resulting division the least number of nodes connecting to other islands is sought-after. Different heuristics are evaluated with the conclusion that for dense graphs a heuristic using MAX-FLOW to divide problematic nodes gives the best result whereas for sparse graphs a heuristic using the single-link clustering method performs best. The problem of finding a spanner path, a path fulfilling the spanner condition, between two nodes is also investigated. The problem is proven to be NP-complete for a graph of size n if the spanner constant is greater than n^(1+1/k)*k^0.5 for some integer k. An algorithm with complexity O(2^(0.822n)) is given. A special type of graph where all the nodes are located on integer locations along the real line is investigated. An algorithm to solve this problem is presented with a complexity of O(2^((c*log n)^2))), where c is a constant depending only on the spanner constant. For instance, the complexity O(2^((5.32*log n)^2))) can be reached for stretch 1.5.

[2] Full text  Tommy Persson. 2009.
Evaluating the use of DyKnow in multi-UAV traffic monitoring applications.
Student Thesis. 75 pages. ISRN: LIU-IDA/LITH-EX-A--09/019--SE.

This Master’s thesis describes an evaluation of the stream-based knowledge pro-cessing middleware framework DyKnow in multi-UAV traffic monitoring applica-tions performed at Saab Aerosystems. The purpose of DyKnow is “to providegeneric and well-structured software support for the processes involved in gen-erating state, object, and event abstractions about the environments of complexsystems.\" It does this by providing the concepts of streams, sources, computa-tional units (CUs), entity frames and chronicles.This evaluation is divided into three parts: A general quality evaluation ofDyKnow using the ISO 9126-1 quality model, a discussion of a series of questionsregarding the specific use and functionality of DyKnow and last, a performanceevaluation. To perform parts of this evaluation, a test application implementinga traffic monitoring scenario was developed using DyKnow and the Java AgentDEvelopment Framework (JADE).The quality evaluation shows that while DyKnow suffers on the usability side,the suitability, accuracy and interoperability were all given high marks.The results of the performance evaluation high-lights the factors that affect thememory and CPU requirements of DyKnow. It is shown that the most significantfactor in the demand placed on the CPU is the number of CUs and streams. Italso shows that DyKnow may suffer dataloss and severe slowdown if the CPU istoo heavily utilized. However, a reasonably sized DyKnow application, such as thescenario implemented in this report, should run without problems on systems atleast half as fast as the one used in the tests.

1996
[1] Jonas Kvarnström. 1996.
A New Tractable Planner for the SAS+ Formalism.
Student Thesis. In series: LiTH-IDA-Ex #9625. 283 pages. ISRN: LiTH-IDA-Ex-9625.

Computer science


Page responsible: Patrick Doherty
Last updated: 2014-04-30