This paper describes the system used by the Machine Learning Group of LTU in subtask 1 of the SemEval-2022 Task 4: Patronizing and Condescending Language (PCL) Detection. Our system consists of finetuning a pretrained text-to-text transfer transformer (T5) and innovatively reducing its out-of-class predictions. The main contributions of this paper are 1) the description of the implementation details of the T5 model we used, 2) analysis of the successes & struggles of the model in this task, and 3) ablation studies beyond the official submission to ascertain the relative importance of data split. Our model achieves an F1 score of 0.5452 on the official test set.

In today's interconnected world, the reliability and performance of network infrastructure are critical for business operations and service delivery. Network Guard, a Python-powered network monitoring solution, is designed to address these challenges by providing comprehensive, real-time insights into network performance. This thesis details the development, implementation, and evaluation of Network Guard, focusing on its ability to perform network discovery, connection monitoring, and bandwidth analysis. The tool leverages Python's flexibility and Scapy's powerful packet manipulation capabilities to offer a scalable, user-friendly interface that meets the diverse needs of modern network environments. Our evaluations demonstrate Network Guard's effectiveness in real-time diagnostics and its potential to enhance network stability and performance, making it a valuable asset for network engineers and administrators

The development of applications that use microservice architecture patterns is increasingrapidly, and this architecture is proven to be successful in many different areas, especiallyin cloud computing. The reason microservices and cloud computing are a great matchis the possibility of scaling and deploying individual services, which positively affects thecost and utilization. This architecture pattern includes some challenges for the devel-opers, such as placement optimisation and knowledge about how the applications aredeployed.This study intends to clarify how the applications in a multi-cluster environment are de-ployed. A service model was created, describing how applications built with microservicearchitecture patterns communicate to each other and which microservices the applicationcontains. More specifically, this can be seen as the first step of placement optimisationthat will be developed in the future. The test cases used to produce the service modelshave various characteristics, such as control planes, where applications were deployed,and numbers of replicas. These kinds of characteristics were varied so that the servicemodels could be relied on and such that the model created works independent of howthe deployment model is created. The created service models show that the applicationtopology is not restricted for the reverse engineering method to work. Independent ofthe number of control planes or replicas, this method worked. Furthermore, the servicemodels created for each test case gave the correct outcome for each application regardingmicroservices and the connections between each microservice.

Decentralization is essential when trust and performance must not depend on a single organization. Distributed Ledger Technologies (DLTs) and Decentralized Hash Tables (DHTs) are examples where the DLT is useful for transactional events, and the DHT is useful for large-scale data storage. The combination of these two technologies can meet many challenges. The blockchain is a DLT with immutable history protected by cryptographic signatures in data blocks. Identification is an essential issue traditionally provided by centralized trust anchors. Self-sovereign identities (SSIs) are proposed decentralized models where users can control and manage their identities with the help of DHT. However, slowness is a challenge among decentralized identification systems because of many connections and requests among participants. In this article, we focus on decentralized identification by DLT and DHT, where users can control their information and store biometrics. We survey some existing alternatives and address the performance challenge by comparing different decentralized identification technologies based on execution time and throughput. We show that the DHT and machine learning model (BioIPFS) performs better than other solutions such as uPort, ShoCard, and BBID.

Stormwater investigations are an important part of the detail plan that is necessary for companies and industries to write. The detail plan is used to show that an area is well suited for among other things, construction. Writing these detail plans is a costly and time consuming process and it is not uncommon they get rejected. This is because it is difficult to find information about the criteria you need to meet and what you need to address within the investigation.

This thesis aims to make this problem less ambiguous by applying the topic modeling algorithm LDA (latent Dirichlet allocation) in order to identify the structure of stormwater investigations. Moreover, sentences that contain words from the topic modeling will be extracted to give each word a perspective of how it can be used in the context of writing a stormwater investigation. Finally a knowledge graph will be created with the extracted topics and sentences.

The result of this study indicates that topic modeling and NLP (natural language processing) can be used to identify the structure of stormwater investigations. Furthermore it can also be used to extract useful information that can be used as a guidance when learning and writing stormwater investigations.

The vast amount of data generated by the Internet of Things (IoT) has made information and cyber security vital for not only personal privacy, but also for the sustainability of the IoT itself. Security and Privacy in the Internet of Things brings together high-quality research on IoT security models, architectures, techniques, and application domains. This concise yet comprehensive volume explores state-of-the-art mitigations in IoT security while addressing important security and privacy challenges across different IoT layers. The book provides timely coverage of IoT architecture, security technologies and mechanisms, and applications. The authors outline emerging trends in IoT security and privacy with a focus on areas such as smart environments and e-health. Topics include authentication and access control, attack detection and prevention, securing IoT through traffic modeling, human aspects in IoT security, and IoT hardware security. Presenting the current body of knowledge in a single volume, Security and Privacy in the Internet of Things:

• Discusses a broad range of IoT attacks and defense mechanisms • Examines IoT security and privacy protocols and approaches • Covers both the logical and physical security of IoT devices • Addresses IoT security through network traffic modeling • Describes privacy preserving techniques in smart cities • Explores current threat and vulnerability analyses

Security and Privacy in the Internet of Things: Architectures, Techniques, and Applications is essential reading for researchers, industry practitioners, and students involved in IoT security development and IoT systems deployment.

The use of cloud-native technologies has grown in popularity in recent years. With its ability to take advantage of the full benefits of cloud computing, cloud-native architecture has become a hot topic among developers and IT professionals. It refers to building and running applications using cloud services and architectures, including containerization, microservices, and automation tools such as Kubernetes to enable fast and continuous delivery of software applications. In Kubernetes, the desired state of a resource is described declaratively and then handles the details of how to get there. Databases are notoriously hard to deploy in such environments, and the Kubernetes operator pattern extends the resources it manages and how to get to the desired state, called reconcile function. Operators exist to manage PostgreSQL databases with backup and restore functionality, and some require a license to be used. Kubegres is a free-to-use open-source operator, but it lacks restore functionality.

This thesis aims to extend the Kubegres operator to support database restoration using dump files. It includes how to create the restore process in Kubernetes, what modifications must be done to the current architecture, and how to make the reconcile function robust and self-healing yet customizable to fit many different needs. Research has been done to explore the design of other operators that already support database restoration. It inspired the design of the resource definition and the restoration process. A new resource definition was added to define the desired state of the database restoration and a new reconcile function to define how to act on it. The state is repeatedly created each time the reconcile function is triggered. During the restoration, a new database is always the target, and once completed, the resources to restore it are deleted, and only the PostgreSQL database is left.

The performance of the modified operator impact compared to the original operator was measured to evaluate the operator. The tests consisted of operations both versions of the operator supported, including PostgreSQL database creation, cluster scaling, and changing resource limits. The two collected metrics, CPU- and memory usage, increased by 0.058-0.4 mvCPU (12-33%) and 8.2 MB (29%), respectively. A qualitative evaluation of the operator using qualities such as robustness, self-healing, customizability, and correctness showed that the design fulfils most of the qualities.

Climate change pressures fossil-driven vehicles into electrification to sustain a healthy planet and environment for its inhabitants. Air traffic contributes to the ongoing climate crisis, and electric aircraft is taking its toll on the avionics market. The electrification of aircraft is complex and requires specialists and researchers to engage in the subject to haste the process of electric aircraft becoming commercial. 

The electrification of aircraft comes with challenges in electromagnetic compatibility (EMC). Part of the aim of this study was to get a broad view of what EMC problems electric aircraft are facing. The leading purpose was to understand electric drive with brushless DC (BLDC) motors and to design and construct an active cancellation filter (ACF) for reducing harmonic noise in electric propulsion systems.

A literature study including recent research on EMC issues onboard, outside of, and within the system of electric aircraft deepened the view of the specifications and limits in the electronic design of the airplane. The literature study also covered electric propulsion using BLDC motors and ACF topologies for suppressing harmonics originating from the power inverter in the electric propulsion system. 

Measurements on a Cypress motor kit with a BLDC motor showed ringings on the motor phase voltages. The parasitic drain-source capacitances in the MOSFETs of the motor inverter caused the ringings, and damping resistors managed to overcome the ringings in simulations and when measured. However, adding damping resistors line-to-line on the motor phase was determined to be ineffective, but the knowledge of the ringings and how to manage them will become relevant for future work.

This study proposed an ACF and a hybrid filter for attenuating harmonics caused by a square wave. The proposed hybrid filter was the ACF filter with an LC filter added to it. Simulations showed that the ACF suppresses harmonics up to 2 MHz, and the hybrid filter had no impact. The measured results showed that the ACF and hybrid filter suppress harmonics up to 10 MHz, and beyond 10 MHz, the filters stop the attenuation of harmonics. The purpose of the hybrid filter was to eliminate harmonics in the higher frequency band, where the performance of the ACF started to lack. However, the hybrid filter showed no influence on the ACF performance, which might have been due to software issues with the Fourier transform. 

Although the filters did attenuate harmonics from the voltage pulse source, it could not be stated whether the ACF and hybrid filter topology will perform the same when applied to the drive system of an electric aircraft. Future work will determine how the filters perform on the Cypress motor kit, and eventually, with electric aircraft.

SRAM or Static Random-Access Memory is the most vital memory technology. SRAM is fast and robust but faces design challenges in nanoscale CMOS such as high leakage, power consumption, and reliability. Quantum-dot Cellular Automata (QCA) is the alternative technology that can be used to address the challenges of conventional SRAM. In this paper, a cost-efficient single layer SRAM cell has been proposed in QCA. The design has 39 cells with a latency of 1.5 clock cycles and achieves an overall improvement in cell count, area, latency, and QCA cost compared to the reported designs. It can therefore be used to design nanoscale memory structures of higher order.

Since the discovery of Ceres, asteroids have been of immense scientific interest and intrigue. They hold answers to many of the fundamental questionsabout the formation and evolution of the Solar System. Therefore, a missionsurveying the asteroid belt with close encounter of carefully chosen asteroidswould be of immense scientific benefit. The trajectory of such an asteroidtour mission needs to be designed such that asteroids of a wide range ofcompositions and sizes are encountered; all with an extremely limited ∆Vbudget.This thesis presents a novel heuristic algorithm to optimize trajectoriesfor an asteroid tour mission with close range flybys (≤ 1000 km). The coresearch algorithm efficiently decouples combinatorial (i.e. choosing the asteroids to flyby)and continuous optimization (i.e. optimizing critical maneuversand events) of what is essentially a mixed integer programming problem.Additionally, different methods to generate a healthy initial population forthe combinatorial optimization are presented.The algorithm is used to generate a set of 1800 feasible trajectories withina 2029+ launch frame. A statistical analysis of these set of trajectories isperformed and important metrics for the search are set based on the statistics.Trajectories allowing flybys to prominent families of asteroids like Flora andNysa with ∆V as low as 4.99 km/s are obtained.Two modified implementations of the algorithm are presented. In a firstiteration, a large sample of trajectories is generated with a limited numberof encounters to the most scientifically interesting targets. While, a posteriori, trajectories are filled in with as many small targets as possible. Thisis achieved in two different ways, namely single step extension and multiplestep extension. The former fills in the trajectories with small targets in onestep, while the latter optimizes the trajectory by filling in with one asteroid per step. The thesis also presents detection of asteroids for successfullyperforming flybys. A photometric filter is developed which prunes out badlyilluminated asteroids. The best trajectory is found to perform well againstthis filter such that nine out of the ten planned flybys are feasible.

Mixed data is a combination of continuous and categorical variables and occurs frequently in fields such as agriculture, remote sensing, biology, medical science, marketing, etc., but only limited work has been done with this type of data. In this study, data on continuous and categorical characters of 452 genotypes of cotton (Gossypium hirsutum) were obtained from an experiment conducted by the Central Institute of Cotton Research (CICR), Sirsa, Haryana (India) during the Kharif season of the year 2018–2019. The machine learning (ML) classifiers/models, namely k-nearest neighbor (KNN), Classification and Regression Tree (CART), C4.5, Naïve Bayes, random forest (RF), bagging, and boosting were considered for cotton genotypes classification. The performance of these ML classifiers was compared to each other along with the linear discriminant analysis (LDA) and logistic regression. The holdout method was used for cross-validation with an 80:20 ratio of training and testing data. The results of the appraisal based on hold-out cross-validation showed that the RF and AdaBoost performed very well, having only two misclassifications with the same accuracy of 97.26% and the error rate of 2.74%. The LDA classifier performed the worst in terms of accuracy, with nine misclassifications. The other performance measures, namely sensitivity, specificity, precision, F1 score, and G-mean, were all together used to find out the best ML classifier among all those considered. Moreover, the RF and AdaBoost algorithms had the highest value of all the performance measures, with 96.97% sensitivity and 97.50% specificity. Thus, these models were found to be the best in classifying the low- and high-yielding cotton genotypes.

Self-driving vehicles has become a hot topic in today's industry during the past years and companies all around the globe are attempting to solve the complex task of developing vehicles that can safely navigate roads and traffic without the assistance of a driver. 

As deep learning and computer vision becomes more streamlined and with the possibility of using fisheye cameras as a cheap alternative to external sensors some companies have begun researching the possibility for assisted driving on vehicles such as electrical scooters to prevent injuries and accidents by detecting dangerous situations as well as promoting a sustainable infrastructure. However training such a model requires gathering large amounts of data which needs to be labeled by a human annotator. This process is expensive, time consuming, and requires extensive quality checking which can be difficult for companies to afford.

This thesis presents an application that allows for semi-automatically labeling a dataset with the help of a human annotator and an object detector. The application trains an object detector together with an active learning framework on a small part of labeled data sampled from the woodscape dataset of fisheye distorted images and uses the knowledge of the trained model as well as using a human annotator as assistance to label more data.

This thesis examines the labeled data produced by using the application described in this thesis and compares them with the quality of the annotations in the woodscape dataset. Results show that the model can't make any quality annotations compared to the woodscape dataset and resulted in the human annotator having to label all of the data, and the model achieved an accuracy of 0.00099 mAP.

The importance of modelling the spreading of infectious diseases as part of a public health strategy has been highlighted by the ongoing coronavirus pandemic. This includes identifying the geographical areas or travel routes most likely to contribute to the spreading of an outbreak. These areas and routes can then be monitored as part of an early warning system, be part of intervention strategies, e.g. lockdowns, aiming to mitigate the spreading of the disease or be a focus of vaccination campaigns. 

This thesis focus on developing a network-based infection model between the municipalities of Sweden in order to identify the areas most likely to contribute to an epidemic. First, a human mobility model is constructed based on the well-known radiation model. Then a network-based SEIR compartmental model is employed to simulate epidemic outbreaks with various parameters. Finally, the adoption of the influence maximization problem known in network science to identify the municipalities having the largest impact on the spreading of infectious diseases. 

The resulting super-spreading municipalities point towards confirmation of the known fact that central highly populated regions in highly populated areas carry a greater risk than their neighbours initially. However, once these areas are targeted, the other resulting nodes show a greater variety in geographical location than expected. Furthermore, a correlation can be seen between increased infections time and greater variety, although more empirical data is required to support this claim.  

For further evaluation of the model, the mobility network was studied due to its central role in creating data for the model parameters. Commuting data in the Gothenburg region were compared to the estimations, showing an overall good accuracy with major deviations in few cases. 

Within the visual creation of film, projects are processed to reach an end goal that should be visualized in the same way wherever it is seen. Color management is needed to control the output colors of a film. Just as the technology inside cameras changes, so does the way to manage colors throughout a project. Color management systems today look completely different compared to what they looked like in the beginning. This essay will address the color management issues that arise for a local VFX production company consisting of 2 people in Skellefteå, Sweden. The method in the thesis identifies the constraints and challenges they face and develops improvements based on the structure of their current way of working. The method is based on today's industry standard systems, suchas ACES (Academy Color Encoding System). The thesis project is based on previously developed knowledge of academic literature. The results indicate that the developed workflow is improved, but further investigation with a clearer build-up could help smaller VFX productions switch from their current workflow to this one.

In recent years, there have been studies focusing on the use of different types of autoencoders (AEs) for monitoring complex nonlinear data coming from industrial and chemical processes. However, in many cases the focus was placed on detection. As a result, practitioners are encountering problems in trying to interpret such complex models and obtaining candidate variables for root cause analysis once an alarm is raised. This paper proposes a novel statistical process control (SPC) framework based on orthogonal autoencoders (OAEs). OAEs regularize the loss function to ensure no correlation among the features of the latent variables. This is extremely beneficial in SPC tasks, as it allows for the invertibility of the covariance matrix when computing the Hotelling T2 statistic, significantly improving detection and diagnosis performance when the process variables are highly correlated. To support the fault diagnosis and identification analysis, we propose an adaptation of the integrated gradients (IG) method. Numerical simulations and the benchmark Tennessee Eastman Process are used to evaluate the performance of the proposed approach by comparing it to traditional approaches as principal component analysis (PCA) and kernel PCA (KPCA). In the analysis, we explore how the information useful for fault detection and diagnosis is stored in the intermediate layers of the encoder network. We also investigate how the correlation structure of the data affects the detection and diagnosis of faulty variables. The results show how the combination of OAEs and IG represents a compelling and ready-to-use solution, offering improved detection and diagnosis performances over the traditional methods.

Bankruptcy prediction is an important research topic. The cost of incorrect decision making in companies and financial institutions can be great and could affect large parts of society. But while it is indeed a major research area, there are few studies which consider the effects of feature selection. This is an important step that could improve the performance of bankruptcy prediction models. This thesis therefore aims to find which feature selection methods perform best for bankruptcy prediction. Five feature selection methods will be compared and used to create datasets with fewer redundant features. To test these methods, a LSTM model is used to train on both an unaltered dataset and datasets created by the mentioned models. The predictive performance of these are then compared with the metrics AUC, Type I error, and Type II error. This study finds that the forward selection algorithm from the Stepwise regression method performed best with an increase in AUC score and decrease in both Type I and Type II error rates compared to the model trained on the unaltered dataset. 

Precision medicine (PM) is an innovative medical approach that considers differences in the individuals’ omics, medical histories, lifestyles, and environmental information in treating diseases. To fully achieve the envisaged gains of PM, various contemporary and future technologies have to be employed, among which are nanotechnology, sensor network, big data, and artificial intelligence. These technologies and other applications require a communication network that will enable them to work in tandem for the benefit of PM. Hence, communication technology serves as the nervous system of PM, without which the entire system collapses. Therefore, it is essential to explore and determine the candidate communication technology requirements that can guarantee the envisioned gains of PM. To the best of our knowledge, no work exploring how communication technology directly impacts the development and deployment of PM solutions exists. This survey paper is designed to stimulate discussions on PM from the communication engineering perspective. We introduce the fundamentals of PM and the demands in terms of quality of service that each of the enabling technologies of PM places on the communication network. We explore the information in the literature to suggest the ideal metric values of the key performance indicators for the implementation of the different components of PM. The comparative analysis of the suitability of the contemporary and future communication technologies for PM implementation is discussed. Finally, some open research challenges for the candidate communication technologies that will enable the full implementation of PM solutions are highlighted.

A high-end core router, such as the CiscoCRS-1, consists of multiple chassis, each of which is populated with multiple line cards that in turn have multiple high-speed network connections. The router’s control plane software must configure, control, and coordinate the set of interfaces to insure that the control software remains running at all times, that faults are detected and corrected, and that forwarding remains consistent across all interfaces. A key requirement is that changes in forwarding tables propagate to all parts of the router quickly without producing transient inconsistencies (the control software must be especially careful to avoid even short-terminternal routing loops). This paper considers the application of cluster management software to a core router. Specifically, we investigate OpenMPI and the ORCM system that uses OpenMPI. After a review of basics and definition of terms, the paper considers fault tolerance and describes ORCM capabilities and limitations. It then presents measurements of latency and throughput that characterize performance and overhead. We conclude with possible extensions and future work.

Along with the dramatic changes that have happened over the years, from the first-generation (1G) of mobile networks to the current fifth-generation (5G) of mobile networks. A wide range of different technologies has been seen as potential possibilities for changing and improving the 5G networks, whereas Software-defined networking (SDN) has been widely regarded as one of the significant enablers for this possibility. At the same time, it has been seen that lowering cost and increasing speed and coverage is not enough for the emerging market. Instead, higher flexibility and increased revenues are desirable and have been seen coming from being able to manage and make use of programmed mobile networks. This thesis investigates the principles and concepts of merging 5G networks with the SDN technology, in which ways those networks can make use of programming to make them more suitable to manage and use. Lastly, it explores the possibility to demonstrate the identified model with the use case for creating virtual private networks. This was mainly done by evaluating and experimenting with 5G networks and the SDN technology together with available tools. Alongside doing so, it was possible to present several principles and concepts suitable for such a programmed mobile network, where some of which were through the utilization of a programming language and a compiler. In addition, it was also possible to discover a compatible SDN controller that seamlessly could be integrated with the other components providing more efficient network management and enhanced usability. To then present the identified model, an implementation could be made by combining the principles and concepts to illustrate a programmed mobile network. The implementation contained two elements, each resembling a virtual private network, with each network further consisting of several user equipments (UEs). Furthermore, it was possible to control the communication between individual UEs and their respective base stations. Several useful pieces of information have thus been gathered in the same place towards answering those research questions, whereas the identified model has also been demonstrated with the use case for creating virtual private networks.

This master thesis investigates the utilisation of various graph based machine learning models for solving a customer segmentation problem, a task coupled to Customer Relationship Management, where the objective is to divide customers into different groups based on similar attributes. More specifically a customer segmentation problem is solved via an unsupervised machine learning technique named clustering, using the k-means clustering algorithm. Three different representations of customers as a vector of attributes are created and then utilised by the k-means algorithm to divide users into different clusters. The first representation is using a elementary feature vector and the other two approaches are using feature vectors produced by graph based machine learning models. Results show that similar grouping are found but that results vary depending on what data is included in the instantiation and training of the various approaches and their corresponding models.

With the increased convergence of technologies whereby a user can access, store and transmit data across different devices in real-time, risks will arise from factors such as lack of appropriate security measures in place and users not having requisite levels of security awareness and not fully understanding how security measures can be used to their advantage. In this paper, we adapt our previously published adversary model for digital rights management (DRM) apps and demonstrate how it can be used to detect vulnerable iOS devices and to analyse (non-DRM) apps for vulnerabilities that can potentially be exploited. Using our adversary model, we investigate several (jailbroken and non-jailbroken) iOS devices, Australian Government Medicare Expert Plus (MEP) app, Commonwealth Bank of Australia app, Western Union app, PayPal app, PocketCloud Remote Desktop app and Simple Transfer Pro app, and reveal previously unknown vulnerabilities. We then demonstrate how the identified vulnerabilities can be exploited to expose the user's sensitive data and personally identifiable information stored on or transmitted from the device. We conclude with several recommendations to enhance the security and privacy of user data stored on or transmitted from these devices.

Industrial processes have for a long time become more and more automated, this is no different in the mining industry. When excavating during mining operations special drill rigs are used to drill holes in the rock walls to be used for either explosives or bolts to support the structure. The study aimed to find out if it was possible to create an algorithm that would use the drill rigs telemetry data to estimate the number of holes it had created over specific time period. The main approach would be to see if machine learning could be used for the problem or if some other method could be theorised. Without the groundwork needed to create a proper machine learning algorithm a basic statistical approach was used to solve the problem, however since there were no actual reports containing the amount of holes a rig drilled the final solution is highly conjectural.

The detection of false data-injection attacks in industrial networks is a growing challenge in the industry because it requires knowledge of application and protocol specific behaviors. Profinet is a common communication standard currently used in the industry, which has the potential to encounter this type of attack. This motivates an examination on whether a solution based on machine learning with a focus on anomaly detection can be implemented and used to detect abnormal data in Profinet packets. Previous work has investigated this topic; however, a solution is not available in the market yet. Any solution that aims to be adopted by the industry requires the detection of abnormal data at the application level and to run the analytics on a resource-constrained device. This thesis presents an implementation, which aims to detect abnormal data in Profinet packets represented as online data streams generated in real-time. The implemented unsupervised learning approach is validated on data from a simulated industrial use-case scenario. The results indicate that the method manages to detect all abnormal behaviors in an industrial network. 

Can compression parameters used in video encoding be estimated, given only the visual information of the resulting compressed video? If so, these parameters could potentially improve existing parametric video quality estimation models. Today, parametric models use information like bitrate to estimate the quality of a given video. This method is inaccurate since it does not consider the coding complexity of a video. The constant rate factor (CRF) parameter for h.264 encoding aims to keep the quality constant while varying the bitrate, if the CRF for a video is known together with bitrate, a better quality estimate could potentially be achieved. In recent years, artificial neural networks and specifically convolutional neural networks have shown great promise in the field of image processing. In this thesis, convolutional neural networks are investigated as a way of estimating the constant rate factor parameter for a degraded video by identifying the compression artifacts and their relation to the CRF used. With the use of ResNet, a model for estimating the CRF for each frame of a video can be derived, these per-frame predictions are further used in a video classification model which performs a total CRF prediction for a given video. The results show that it is possible to find a relation between the visual encoding artifacts and CRF used. The top-5 accuracy achieved for the model is at 61.9% with the use of limited training data. Given that today’s parametric bitrate based models for quality have no information about coding complexity, even a rough estimate of the CRF could improve the precision of them.

Continuous authentication using behavioral biometrics can provide an additional layer of protection against online account hijacking and fraud. Mouse dynamics classification is the concept of determining the authenticity of a user through the use of machine learning algorithms on mouse movement data. This thesis investigates the viability of state of the art deep learning technologies in mouse dynamics classification by designing convolutional neural network classifiers taking mouse movement images as input. For purposes of comparison, classifiers using the random forest algorithm and engineered features inspired by related works are implemented and tested on the same data set as the neural network classifier.

A technique for lowering bias toward the on-screen location of mouse movement images is introduced, although its effectiveness is questionable and requires further research to thoroughly investigate. This technique was named 'centering', and is used for the deep learning-based classification methods alongside images not using the technique. The neural network classifiers yielded single action classification accuracies of 66% for centering, and 78% for non-centering. The random forest classifiers achieved the average accuracy of 79% for single action classification, which is very close to the results of other studies using similar methods. In addition to single action classification, a set based classification is made. This is the method most suitable for implementation in an actual authentication system as the accuracy is much higher.

The neural network and random forest classifiers have different strengths. The neural network is proficient at classifying mouse actions that are of similar appearance in terms of length, location, and curvature. The random forest classifiers seem to be more consistent in these regards, although the accuracy deteriorates for especially long actions. As the different classification methods in this study have different strengths and weaknesses, a composite classification experiment was made where the output was determined by the least ambiguous output of the two models. This composite classification had an accuracy of 83%, meaning it outperformed both the individual models.

With the increasing integration of artificial intelligence (AI) technologies into educational settings, it becomes important to ensure responsible and effective use of these systems. This thesis addresses two critical challenges within AI-driven educational applications: the effortless integration of different Large Language Models (LLMs) and the mitigation of inappropriate content. An AI assistant chatbot was developed, allowing teachers to design custom chatbots and set rules for them, enhancing students’ learning experiences. Evaluation of LangChain as a framework for LLM integration, alongside various prompt engineering techniques including zero-shot, few-shot, zero-shot chain-of-thought, and prompt chaining, revealed LangChain’s suitability for this task and highlighted prompt chaining as the most effective method for mitigating inappropriate content in this use case. Looking ahead, future research could focus on further exploring prompt engineering capabilities and strategies to ensure uniform learning outcomes for all students, as well as leveraging LangChain to enhance the adaptability and accessibility of educational applications.

With the rise of microservice architectures, there is a need for an orchestration tool to manage containers. Kubernetes has emerged as one of the most popular alternatives, adopting widespread usage. But managing multiple Kubernetes clusters on its own have proven to be a challenging task. This difficulty has given rise to multiple cloud based alternatives which help streamline the managing process of a cluster environment and helps maintain an extreme high availability environment that is hard to replicate in an on premise environment. Using these cloud based platforms for hosting and managing ones system is great, but alleviating control of a system to a cloud provider masquerades any illicit behaviour performed on or through the system.

The scope of this thesis is on examining optional designs that will automate the process of executing a geographical failover between different locations to better sustain an on premise fault tolerant kubernetes environment. There already exists multiple tools in the area of kubernetes service mesh, but their focus is not primarily on increasing system resilience but to increase security, observability and performance. Linkerd is a sidecar oriented service mesh which supports geographical failover by manually announcing individual services between cluster(s) mirror gateways. Cilium offers an Container Networking Interface (CNI) which performs routing through eBPF and allows for seamless failover between clusters by managing cross cluster service endpoints. Both of the mentioned service mesh providers handle failover from inside the kubernetes cluster.

The contributions includes two new peer-to-peer designs that focus on external cluster geographical failover - both designs are compatible with preexisting kubernetes clusters without internal modifications. A fully repli-cated design was then realised into a proof of concept (POC), and tested against a Cilium multi cluster environment on the metric of north to south traffic latency. Due to the nature of the underlying hardware, the tests showed that the POC can be used for external geographical failover and it showed potential performance capabilities in a limited lab scale.

As the purpose of this thesis was not to determine the traffic throughput of a geographical failover solution; but to examine different approaches automatic geographical failover can be implemented, the tests were a success. Therefore, this thesis can conclude that there exists several working solutions, and the POC have shown that there are still undiscovered and unimplemented solutions to explore.

Recently, the Internet of Things (IoT) has emerged as an important way to connect diverse physical devices to the internet. The IoT paves the way for a slew of new cutting-edge applications. Despite the prospective benefits and many security solutions offered in the literature, the security of IoT networks remains a critical concern, considering the massive amount of data generated and transmitted. The resource-constrained, mobile, and heterogeneous nature of the IoT makes it increasingly challenging to preserve security in routing protocols, such as the routing protocol for low-power and lossy networks (RPL). RPL does not offer good protection against routing attacks, such as rank, Sybil, and sinkhole attacks. Therefore, to augment the security of RPL, this article proposes the energy-efficient multi-mobile agent-based trust framework for RPL (MMTM-RPL). The goal of MMTM-RPL is to mitigate internal attacks in IoT-based wireless sensor networks using fog layer capabilities. MMTM-RPL mitigates rank, Sybil, and sinkhole attacks while minimizing energy and message overheads by 25–30% due to the use of mobile agents and dynamic itineraries. MMTM-RPL enhances the security of RPL and improves network lifetime (by 25–30% or more) and the detection rate (by 10% or more) compared to state-of-the-art approaches, namely, DCTM-RPL, RBAM-IoT, RPL-MRC, and DSH-RPL. 

Digitalizing cement production plants to improve operation parameters’ control might reduce energy consumption and increase process sustainabilities. Cement production plants are one of the extremest CO2 emissions, and the rotary kiln is a cement plant’s most energy-consuming and energy-wasting unit. Thus, enhancing its operation assessments adsorb attention. Since many factors would affect the clinker production quality and rotary kiln efficiency, controlling those variables is beyond operator capabilities. Constructing a conscious-lab “CL” (developing an explainable artificial intelligence “EAI” model based on the industrial operating dataset) can potentially tackle those critical issues, reduce laboratory costs, save time, improve process maintenance and help for better training operators. As a novel approach, this investigation examined extreme gradient boosting (XGBoost) coupled with SHAP (SHapley Additive exPlanations) “SHAP-XGBoost” for the modeling and prediction of the rotary kiln factors (feed rate and induced draft fan current) based on over 3,000 records collected from the Ilam cement plant. SHAP illustrated the relationships between each record and variables with the rotary kiln factors, demonstrated their correlation magnitude, and ranked them based on their importance. XGBoost accurately (R-square 0.96) could predict the rotary kiln factors where results showed higher exactness than typical EAI models.

Dot peen markings are used by Scania Ferruform to maintain a traceability of their products throughout the manufacturing. Quality inspection of the markings are performed to ensure that they are added correctly and readable. This is, however, done manually by workers, which they are looking to change. Machine vision, in combination with machine learning, could prove helpful in automating this process, which is where this thesis comes in. Images of two types of dot peen markings were gathered using different experimental setups and equipment. Amazon Rekognition and MVTec Halcon were both used to predict the characters of the images, in order to determine if the two systems could be used to demonstrate that the quality inspection can be automated. To improve the result, the images were also processed with varied techniques. The pretrained version of Amazon Rekognition and MVTec Halcon, with unprocessed image, performed the best. They both predicted all the characters correctly, and showed a high confidence in their predictions, with an average confidence of 96.41% and 99.87% respectively. When processing the images before predicting the confidence of the systems decreased and predictions were also made incorrectly. Custom training a model also showed a poor result, with the best combination of average precision and overall recall being at 0.733 and 0.561 respectively.

In this thesis, a proof of concept of a digital twin of a type of reheating furnace, the walking beam furnace, is presented. It is created by using a machine learning concept called a neural network. The digital twin is trained using real data from a walking beam furnace located in Swerim AB, Luleå, and is taught to predict the temperature in the furnace using air, fuel and pressure as inputs. The machine learning technique used is an artifical neural network in the form of a multilayer perceptron model. The resulting model consists of 3 layers, input, hidden and output layer. The hyperparameters is decided by using grid search cross validation. The hyperparameters chosen to use in this thesis was amount of epochs, optimizer, learning rate, batch size, activation function, regularizer and amount of neurons in the hidden layer. The final settings for these can be found in table. The digital twin is then evaluated comparing predicted temperatures and actual temperatures from the measured data. The end result shows that the twin performs reasonably well. The predictions differs from measured temperature with a percentage around 0.5% to 1.5%.

The times of manual labour are changing as automation grows larger and larger by the day. Self-driving vehicles being one of the more well known examples of automation (the vehicles in this thesis being those found in the construction industry), relies on a machine learning network to recognize its surroundings. To achieve this, the network needs a dataset. A dataset consists of two things, data, which usually come in the form of images, and annotated labels in order for it to learn what it sees. The labels is a descriptor that describes what objects exists in an image, and coordinates for where in the image these objects exists, and the area they occupy.

As data is collected, it needs to be manually annotated, which can take several months to finish. With this in mind, is it possible to set up some form of semi-automatic annotation step that does a majority of the work? If so, what techniques can be used to achieve this? How does it compare to a dataset which have been annotated by a human? and is it even worth implementing in the first place?

For this research, a dataset was collected where a remote controlled wheel loader approached a stationary dump truck, at various different angles, and during different conditions. Four videos were used in the trainingset, containing 679 images and their respective labels. Two other videos were used for the validationset, consisting of 120 images and their respective labels. 

The chosen object detector was YOLOv3, which has a low inference time and high accuracy. This helped with gathering results at a faster rate than what would've been possible if an older version was chosen. 

The method which was chosen for doing the automatic annotations was linear interpolation, which was implemented to work in conjunction with the labels of the trainingset to approximate the corresponding values. 

The interpolation was done at different frame gaps, a gap of 10 frames, a gap of 20 frames, all the way up to a gap of 60 frames. This was done in order to help locate a sweet spot, where the model had similar performance compared to the manually annotated dataset.

The results showed that the fully manually annotated dataset approached a precision value of 0.8, a recall of 0.96, and a mean average precision (mAP) value of 0.95. Some of the models which had interpolated frames between a set gap, achieved similar results in the metrics, where interpolating between every 10th frame, every 20th frame, and every 30th frame, showed the most promise. They all approached precision values of around 0.8, a recall of around 0.94, and an mAP value of around 0.9. 

During the COVID-19 pandemic the questions of the efficiency around meetings has been in the forefront of some discussion inside companies. One way to measure efficiency is to measure the interactivity between different participants. In order to measure this the participants need to be identified. With the recent spike of Machine learning advancements, is this something that can be done using facial and voice recognition? Another field that has risen to the top is cloud computing. Can machine learning and cloud computing be used to evaluate and monitor a meeting, thus handling both audio and video streams in a real time environment?

The conclusion of this thesis is that Artificial Intelligence(AI) can be used to monitor a meeting. To be able to do so Amazon Web Service (AWS) can be utilized. The choice of using a DeepLens was however not best choice. A hardware like DeepLens is required, but with better integration with cloud computing, as well with more freedom regarding the usage of several models for handling both feeds. With the usage of other models to automatic annotate data the time needed for training a new model can be reduced. The data generated during a single meeting is enough with the help of transfer learning from Amazon web service to build a model for facial identification and detection.

This thesis explores the creation of a scaling, containerized, advertisement server that will be used by Gold Town Games AB to better integrate ads into their application(s). The server is built as a Docker image that will be used to create server instances on AWS Elastic Container Service for automatic scaling and server resource configuration.

The server was created with the intention that GTG will have full control over what advertisements are shown in their application(s) and to seamlessly integrate sponsored logos onto jerseys or sports fields. This will not only serve as a source of income with advertisers paying for ad space, but it will also make the game elements more realistic as we have come to expect teams and stadiums to be sponsored and plastered with company logos. Another important part when displaying advertisement is to track statistics for the ads, since without a way to show advertisers that their ads are shown and that they are generating engagement it is very hard to sell the ad space.

Road surface temperature (RST) plays an essential role in analyzing road surface conditions during winter in countries with adverse winter climates. A reduction in RST can have a negative impact on road safety due to decreasing vehicle grip on the road surface. Therefore, decision makers need to monitor low surface temperatures and plan for winter road maintenance. However, RST sensors can fail for different reasons, such as power outages. RST sensor failure will lead to lack of information about the road surface, which can be problematic, especially for critical road segments. Hence, the novelty of this study is to use a deep learning algorithm to predict RSTs in road segments if a sensor fails at a station using historical data from two other road stations. The mean absolute error in the proposed model is 0.453 and the model explains 98.6% of observations. In addition, since the adjustment of deep learning parameters (e.g., hidden layers, optimizer, activation function, etc.) is associated with epistemic uncertainty, a semiquantitative approach is developed for uncertainty assessment. With this approach, the most important and uncertain parameters in RST prediction models can be identified. The results have shown that the optimizer is the most uncertain and important parameter.

Denna rapport innefattar en genomgång av den nya 802.11ax (WIFI6)-standarden med djupdykning och förklaringar på de nya funktionerna BSS Coloring, Target Wake Time och OFDMA med flera. Frågeställningen som presenteras är om det är en bra investering av små företag och hem-användareatt uppgradera sin utrustning till WIFI 6 kompatibel sådan.Efter genomgång av protokollet och de nya funktionerna så rekommenderas inte detta medmotiveringen att tekniken är för ny och antal enheter som mobil och laptops med standarden är iskrivande stund inte så många och de som finns kan kosta mycket att köpa och byta ut.Det görs även en undersökning av Medlefors Folkhögskola och där så upptäcks att kanalplaneringenär bristfällig och att det finns accesspunkter från två tillverkare som sänder samtidigt i lokalerna. Därför ges det ett förslag på uppgradering av samtliga enheter till WIFI 6-kompatibel sådan. Utrustningen som föreslås är av märket Ubiquiti och samtliga enheter föreslås att bytas ut.Slutligen diskuteras summeringen med att ytterligare tester på större nätverk måste göras för attfullt ut se att det som låter så bra i teorin också ska fungera vid skarp implementation på stora arenorfullsatt med folk, vilket framtiden får utvisa. 

Documentation for programming projects can vary both in quality and availability. The availability of documentation can vary more for a closed working environment, since fewer developers will read the documentation. Documenting programming projects can be demanding on worker hours and unappreciated among developers. It is a common conception that developers rather invest time on developing a project than documenting a project, and making the documentation process more effective would benefit developers. To move towards a more automated process of writing documentation, this work generated documentation for repositories which attempts to summarize the repositories in their use cases and functionalities. Two different implementations are created to generate documentation using an on-premise hosted large language model (LLM) as a tool. First, the embedded solution processes all available code in a project and creates the documentation based on multiple summarizations of files and folders. Second, the RAG solution attempts to use only the most important parts of the code and lets the LLM create the documentation on a smaller set of the codebase. The results show that generating documentation is possible, but unreliable and must be controlled by a person with knowledge about the codebase. The embedded solution seems to be more reliable and produce better results, but is more costly compared to the RAG solution.

The traditional way to coordinate multiple clients in a multiplayer based game on multiple platforms is to create an implementation on a per-platform basis - resulting in often four different implementations, one for each major platform - ie. iOS, Android, Windows, and Linux based operating systems. This report examines the possibility of replacing multiple so called “native apps” with a single web based implementation - granting users access on all devices that supports modern browsers, and discusses what tools were used in the development of the application and why. 

This master thesis investigates the utility of spaced repetition learning in engineering education, focusing on knowledge retention for students. The study involved the development of an application with a Nest.js backend integrated with a MySQL database and a Next.js frontend, designed with human-computer interaction (HCI) principles to ensure usability and user-friendliness. Surveys were conducted to gather feedback on HCI and spaced repetition learning, although the limited number of participants affected the statistical significance of the findings.

Initial responses indicate that the platform is perceived as easy to navigate but lack engagement due to its simplistic nature. The absence of distracting elements is noted, yet issues with font size consistency have been identified. Further feedback suggests that students are open to spaced repetition learning techniques.

The study draws on previous research demonstrating the effectiveness of spaced repetition learning in enhancing long-term memory retention. The thesis aims to address the challenge of rapid memory erosion associated with traditional massed learning approaches. Further research is recommended to gather more substantial data and refine the platform for improved usability and effectiveness in engineering education.

In hyperscale data center operations, automation is applied in many ways as it is becomes very hard to scale otherwise. There are however areas relating to understanding, grouping and diagnosing of error reports that are done manually at Facebook today. This master's thesis investigates solutions for applying unsupervised clustering methods to server error reports, server properties and historical data to speed up and enhance the process of finding and root causing systematic issues. By utilizing data representations that can embed both key-value data and historical event log data, the thesis shows that clustering algorithms together with data representations that capture syntactic and semantic structures in the data can be applied with good results in a real-world scenario.

In this thesis, the focus is on enhancing EMI's products, Onboard and PocketMine leading software solutions in the mining sector. This study explores how the extensive data gathered by Onboard can be used to develop a more precise and reliable positioning system for PocketMine and to create the foundation of redundancy for Onboard using machine learning models. Furthermore, it will explore what machine learning model performs optimally with this data. This thesis is motivated by the potential of data-driven methodologies to enhance the safety and accuracy of EMI’s products, significantly improving operational safety and precision in challenging underground environments but also contributing to the broader field of positioning technology.

The goals for this thesis are achieved by comparing four different ML models on three distinct datasets based on locations in the mine to decide which models the final solution will be using. Additionally, the idea of creating a model encapsulating the entire mine is examined and compared to the POI-specific models to see if it is feasible for one model to learn the intricacies of the mine. In addition to this, the deployment strategy will be discussed.

Upon comparing the models against each other and the mine-wide model, it was decided to move on with Weighted K Nearest Neighbors as the model of choice based on several evaluation metrics. The large scale of the mine proved too great to be handled by one model so the decision to cluster the mine into 100 distinct clusters and create one model for each cluster was made. 

The results show that the proposed solution made a great improvement in positional accuracy over the current positioning algorithm of PocketMine. This improvement suggests in line with testing it against Onboard that the proposed model could effectively serve as a reliable backup system for Onboard.

Whole Slide Images (WSIs) are digital scans containing rich pathology information. There are many available WSI datasets that can be used for a wide range of purposes such as diagnostic tasks and analysis, but the availability of labeled WSI datasets is very limited since the annotation process is both very costly and time consuming. Self-supervised learning is a way of training neural networks to learn and predict the underlying structure of input data without any labels. 

AstraZeneca have developed a self-supervised learning feature extractor, the Drug-development Image Model Embeddings (DIME) pipeline, that trains on unlabeled WSIs and produces numerical embedding representations of WSI.

This thesis applies the DIME-embeddings to a binary tumour classification task on the annotated Camelyon16 dataset by using the DIME pipeline as a feature extractor and train a simple binary classifier on the embedding representations instead of the WSI patches. The results are then compared to previous fully-supervised learning approaches to see if the embedding features generated by the DIME pipeline are sufficiently predictive with simple classifiers for the downstream task of binary classification. 

The DIME embeddings were trained using Logistic Regression, Multi-Layer Perceptron and Gradient Boosting and the best performing model, a Multi-Layer Perceptron neural network trained on the DIME embeddings produced with an inpainting algorithm achieved a patch-level classification accuracy of 97.3%. This is very competitive results to the fully-supervised algorithms trained on the Camelyon16 dataset, beating some of them, while having a 1.1% gap to the best performing fully-supervised model. In addition to this, the performance of the DIME embeddings on reduced training sets also shows that the features captured in the DIME embeddings are sufficient. 

The concept of Metaverse is emerging in the industry. Metaverse is expected to be important in industrial asset management and sustainable operation and maintenance. Some of the potentials of implementing Metaverse in the industry can be related to virtual co-creation and design, remote and virtual inspection and maintenance, skills development and training, simulation, safety, and security. Additionally, Metaverse integrated with Artificial Intelligence (AI) and digital technologies will augment human perception, facilitating the Human-System-Interaction (HSI). The traditional HSI carries limitations regarding usability, immersiveness, and connectivity when it comes to the interaction between the virtual, augmented, and real world. An improved HSI in such cyberspace applications may lead to a better understanding of the system and eventually reduced faults. However, implementing Metaverse in industrial contexts is challenging and has not yet been explored thoroughly and systematically. Hence, this paper aims to systematically identify and investigate the various issues and challenges in the implementation of Metaverse in industrial contexts from an HSI perspective. The paper will further provide a taxonomy of these issues and challenges. The research methodology has been based on literature surveys, active and passive observations, and experiments done in the eMaintenanceLAB at Luleå University of Technology. The findings from this paper can be used to increase the effectiveness and efficiency of implementing Metaverse in various industrial contexts.

Digitalisation is changing the railway globally. One of the major concerns in digital transformation of the railway is the increased exposure to cyberattacks. The railway is vulnerable to these cyberattacks because the number of digital items and number of interfaces between digital and physical components in these systems keep growing. Increased number of digital items and interfaces require new methodologies, frameworks, models, concepts, and architectures to ensure the railway system’s resilience with respect to cybersecurity challenges, such as adoption and convergence of Information Technology (IT) and Operational Technology (OT) technology within the railway. This convergence has brought significant benefits in reliability, operational efficiency, capacity as well as improvements in passenger experience but also increases the vulnerability towards cyberattacks from individuals, organizations, and governments. This paper proposes a methodology on how to deals with OT security in the railway signalling using failure mode, effects and criticality analysis (FMECA) and ISA/IEC 62443 security risk assessment methodologies.

I det här arbetet ger vi en introduktion till opartiska kombinatoriska spel, där optimala strategier lyfts fram. Hur kan vi med matematiska metoder bevisa att en spelare kan tvinga fram en vinst, och hur räknar vi ut det optimala draget? För alla spel som beskrivs kan en vinst tvingas fram om spelet är i en så kallad P-position. Grundy-värdet kan användas för bestämning av utfallet på ett spel, där ett Grundy-värde på noll motsvarar P-position. Grundy-värdet definieras rekursivt och tester har implementerats där en rekursiv och en iterativ implementering jämförs. Resultatet visar att den iterativa metoden är mycket mer effektiv. Vi kan lägga ihop flera spel för att skapa ett summa-spel, där flera spel spelas samtidigt. Grundy-värdet för ett summa-spel beräknas av Nimsumman mellan de separata spelens Grundy-värden. Nim-summan mellan två heltal är en binär xor-operation mellan heltalen. En applikation har utvecklats för tester på teorin där spelen Nim, Silver-Dollar, ett subtraktions-spel och Wythoffs spel är implementerade. Applikationen ger möjlighet att tillämpa de optimala strategierna på de separata spelen eller på ett eget kombinerat summa-spel.