In the progression of the digital world the urgency of securing data becomes increasingly important and the emergence of modern data pipelines has fueled advances in data management and analysis which provides new opportunities for insights. Thus, due to this age of extensive data sharing and andconnectivity leads to ensuring security for sensitive information while in transit is a critical concern. Among modern prominent data pipelines, Apache Kafka stands as a versatile streaming platform with the ability to manage vast data flows and real-time data processing which has allowed the platform to gain traction, though the growth in prominence has resulted in increased demands for data security.

This thesis, conducted in collaboration with Basalt AB and provides an exploration of the field of data security in the context of Apache Kafka and involves the development, research, and validation of a proposed prototype with a focus on protective measures during data transmission. The thesis presents a strategy and solution to address these challenges and consists of an encryption method and a customized approach suited for the Apache Kafka environment. The contribution includes showcasing the implementation and deployment of the encryption coupled with the presentation of practical suggestions for the proposed model with the aim of outlining measures to access means for protecting confidential information.

Battery storage is emerging as a key component of intelligent green electricitiy systems. The battery is monetized through market participation, which usually involves bidding. Bidding is a multi‐objective optimization problem, involving targets such as maximizing market compensation and minimizing penalties for failing to provide the service and costs for battery aging. In this article, battery participation is investigated on primary frequency reserve markets. Reinforcement learning is applied for the optimization. In previous research, only simplified formulations of battery aging have been used in the reinforcement learning formulation, so it is unclear how the optimizer would perform with a real battery. In this article, a physics‐based battery aging model is used to assess the aging. The contribution of this article is a methodology involving a realistic battery simulation to assess the performance of the trained RL agent with respect to battery aging in order to inform the selection of the weighting of the aging term in the RL reward formula. The RL agent performs day-ahead bidding on the Finnish Frequency Containment Reserves for Normal Operation market, with the objective of maximizing market compensation, minimizing market penalties and minimizing aging costs.

Battery storages are an essential element of the emerging smart grid. Compared to other distributed intelligent energy resources, batteries have the advantage of being able to rapidly react to events such as renewable generation fluctuations or grid disturbances. There is a lack of research on ways to profitably exploit this ability. Any solution needs to consider rapid electrical phenomena as well as the much slower dynamics of relevant electricity markets. Reinforcement learning is a branch of artificial intelligence that has shown promise in optimizing complex problems involving uncertainty. This article applies reinforcement learning to the problem of trading batteries. The problem involves two timescales, both of which are important for profitability. Firstly, trading the battery capacity must occur on the timescale of the chosen electricity markets. Secondly, the real-time operation of the battery must ensure that no financial penalties are incurred from failing to meet the technical specification. The trading-related decisions must be done under uncertainties, such as unknown future market prices and unpredictable power grid disturbances. In this article, a simulation model of a battery system is proposed as the environment to train a reinforcement learning agent to make such decisions. The system is demonstrated with an application of the battery to Finnish primary frequency reserve markets.

Data centers operations are notoriously energy-hungry, with the computing and cooling infrastructures drawing comparable amount of electrical power to operate. A direction to improve their efciency is to optimizethe cooling, in the sense of implementing cooling infrastructures controlschemes that avoid performing over-cooling of the servers.Towards this direction, this work investigates minimum cost linearquadratic control strategies for the problem of managing air cooled datacenters. We derive a physical and a black box model for a general datacenter, identify this model from real data, and then derive, present andtest in the eld a model based Linear-Quadratic Regulator (LQR) strategy that sets the optimal coolant temperature for each individual coolingunit. To validate the approach we compare the eld tests from the LQR strategy against classical Proportional-Integral-Derivative (PID) controlstrategies, and show through our experiments that it is possible to reducethe energy consumption with respect to the existing practices by severalpoints percent without harming the servers within the data center fromthermal perspectives.

The Internet of Things (IoT) provides the convenience and automation in our homes, but it also introduces significant security risks. This thesis investigates the security challenges in transmitting the data from Smart Home Controllers to cloud platforms like Amazon Web Services (AWS) IoT. For this purpose, a realistic testbed was set up to do some real-world attacks, such as Man-in-the-Middle attacks, Wi-Fi traffic interception, and Denial-of-Service (DoS) attacks. The study discovered the vulnerabilities in unencrypted MQTT communication, vulnerability of local networks to ARP spoofing, and the risks of unsecured Wi-Fi networks, confirming the findings of previous research on IoT security threats. The potential consequences of these vulnerabilities range from breaches of privacy to physical harm. This thesis proposes a set of comprehensive best practices for the data transmission from Smart Home Controllers to the cloud. These recommendations include the implementation of network segmentation, encryption of MQTT traffic, strong Wi-Fi security measures, and using MQTT proxies to mitigate impact of the DoS attacks. The findings of this thesis have many broader implications for the future of the IoT security. If the security in the design and Deployment of the IoT devices like Smart Home Controllers can be prioritized,  then we can create a future where IoT devices are intelligent and convenient as well as secure.

The rise of zero-trust security has emerged as a response to the evolving threat landscape,offering a more comprehensive approach to information protection compared totraditional perimeter-based security models. However, the implementation of Zero Trustcan introduce challenges related to user experience, potentially hindering user adoptionand the overall effectiveness of the security framework.This research addresses the gap in user-centric research for Zero Trust by investigatingthe impact of Zero Trust login systems on user experience and the factors influencinguser perceptions towards Zero Trust principles. The study employs a user-centric evaluationapproach, combining quantitative and qualitative methods to gain a comprehensiveunderstanding of user perceptions and experiences. The findings highlight the potentialtrade-offs between security and usability, emphasizing the importance of user educationand organizational support for successful adoption.This research contributes to the field by offering recommendations for organizationsto implement Zero Trust login systems in a user-centric manner, addressing potentialchallenges and fostering a positive user experience. By bridging the gap in user-centricresearch for Zero Trust, this study provides valuable insights for organizations seeking toadopt Zero Trust security while maintaining a positive user experience.

Background: The gearbox and machinery faults prediction are expensive both in terms of repair and loss output in production. These losses or faults may lead to complete machinery or plant breakdown. 

Objective: The goal of this study was to apply advanced machine learning techniques to avoid these losses and faults and replace them with predictive maintenance. To identify and predict the faults in industrial machinery using Machine Learning (ML)  and Deep Learning (DL) approaches. 

Methods: Our study was based on two types of datasets which includes gearbox and rotatory machinery dataset. These datasets were analyzed to predict the faults using machine learning and deep neural network models. The performance of the model was evaluated for both the datasets with binary and multi-classification problems using the different machine learning models and their statistics.

Results: In the case of the gearbox fault dataset with a binary classification problem, we observed random forest and deep neural network models performed equally well, with the highest F1-score and AUC score of around 0.98 and with the least error rate of 7%.  In addition to this, in the case of the multi-classification rotatory machinery fault prediction dataset, the random forest model outperformed the deep neural network model with an AUC score of 0.98. 

Conclusions: In conclusion classification efficiency of the Machine Learning (ML) and Deep Neural Network (DNN) model were tested and evaluated. Our results show Random Forest (RF) and Deep Neural Network (DNN) models have better fault prediction ability to identify the different types of rotatory machinery and gearbox faults as compared to the decision tree and AdaBoost. 

Keywords: Machine Learning, Deep Learning, Big Data, Predictive Maintenance, Rotatory Machinery Fault Prediction, Gearbox Fault Prediction, Machinery Fault Database, Internet of Things (IoT), Spectra quest machinery fault simulator, Cloud Computing, Industry 4.0

The popularity of network virtualization has recently regained considerable momentum because of the emergence of OpenFlow technology. It is essentially decouples a data plane from a control plane and promotes hardware programmability. Subsequently, OpenFlow facilitates the implementation of network virtualization. This study aims to provide an overview of different approaches to create a virtual network using OpenFlow technology. The paper also presents the OpenFlow components to compare conventional network architecture with OpenFlow network architecture, particularly in terms of the virtualization. A thematic OpenFlow network virtualization taxonomy is devised to categorize network virtualization approaches. Several testbeds that support OpenFlow network virtualization are discussed with case studies to show the capabilities of OpenFlow virtualization. Moreover, the advantages of popular OpenFlow controllers that are designed to enhance network virtualization is compared and analyzed. Finally, we present key research challenges that mainly focus on security, scalability, reliability, isolation, and monitoring in the OpenFlow virtual environment. Numerous potential directions to tackle the problems related to OpenFlow network virtualization are likewise discussed

Although most brain tumor diagnosis studies have focused on tumor segmentation and localization operations, few studies have focused on tumor detection as a time- and effort-saving process. This study introduces a new network structure for accurate glioma tumor detection and classification using two parallel deep convolutional neural networks (PDCNNs). The proposed structure is designed to identify the presence and absence of a brain tumor in MRI images and classify the type of tumor images as high-grade gliomas (HGGs, i.e., glioblastomas) or low-grade gliomas (LGGs). The introduced PDCNNs structure takes advantage of both global and local features extracted from the two parallel stages. The proposed structure is not only accurate but also efficient, as the convolutional layers are more accurate because they learn spatial features, and they are efficient in the testing phase since they reduce the number of weights, which reduces the memory usage and runtime. Simulation experiments were accomplished using an MRI dataset extracted from the BraTS 2017 database. The obtained results show that the proposed parallel network structure outperforms other detection and classification methods in the literature.

The use of medical images has been continuously increasing, which makes manual investigations of every image a difficult task. This study focuses on classifying brain magnetic resonance images (MRIs) as normal, where a brain tumor is absent, or as abnormal, where a brain tumor is present. A hybrid intelligent system for automatic brain tumor detection and MRI classification is proposed. This system assists radiologists in interpreting the MRIs, improves the brain tumor diagnostic accuracy, and directs the focus toward the abnormal images only. The proposed computer-aided diagnosis (CAD) system consists of five steps: MRI preprocessing to remove the background noise, image segmentation by combining Otsu binarization and K-means clustering, feature extraction using the discrete wavelet transform (DWT) approach, and dimensionality reduction of the features by applying the principal component analysis (PCA) method. The major features were submitted to a kernel support vector machine (KSVM) for performing the MRI classification. The performance evaluation of the proposed system measured a maximum classification accuracy of 100 % using an available MRIs database. The processing time for all processes was recorded as 1.23 seconds. The obtained results have demonstrated the superiority of the proposed system.

Computer-aided diagnosis (CAD) systems have become very important for the medical diagnosis of brain tumors. The systems improve the diagnostic accuracy and reduce the required time. In this paper, a two-stage CAD system has been developed for automatic detection and classification of brain tumor through magnetic resonance images (MRIs). In the first stage, the system classifies brain tumor MRI into normal and abnormal images. In the second stage, the type of tumor is classified as benign (Noncancerous) or malignant (Cancerous) from the abnormal MRIs. The proposed CAD ensembles the following computational methods: MRI image segmentation by K-means clustering, feature extraction using discrete wavelet transform (DWT), feature reduction by applying principal component analysis (PCA). The two-stage classification has been conducted using a support vector machine (SVM). Performance evaluation of the proposed CAD has achieved promising results using a non-standard MRIs database.

The successful early diagnosis of brain tumors plays a major role in improving the treatment outcomes and thus improving patient survival. Manually evaluating the numerous magnetic resonance imaging (MRI) images produced routinely in the clinic is a difficult process. Thus, there is a crucial need for computer-aided methods with better accuracy for early tumor diagnosis. Computer-aided brain tumor diagnosis from MRI images consists of tumor detection, segmentation, and classification processes. Over the past few years, many studies have focused on traditional or classical machine learning techniques for brain tumor diagnosis. Recently, interest has developed in using deep learning techniques for diagnosing brain tumors with better accuracy and robustness. This study presents a comprehensive review of traditional machine learning techniques and evolving deep learning techniques for brain tumor diagnosis. This review paper identifies the key achievements reflected in the performance measurement metrics of the applied algorithms in the three diagnosis processes. In addition, this study discusses the key findings and draws attention to the lessons learned as a roadmap for future research.

The utilization of different types of brain images has been expanding, which makes manually examining each image a labor-intensive task. This study introduces a brain tumor segmentation method that uses two parallel U-Net with an asymmetric residual-based deep convolutional neural network (TPUAR-Net). The proposed method is customized to segment high and low grade glioblastomas identified from magnetic resonance imaging (MRI) data. Varieties of these tumors can appear anywhere in the brain and may have practically any shape, contrast, or size. Thus, this study used deep learning techniques based on adaptive, high-efficiency neural networks in the proposed model structure. In this paper, several high-performance models based on convolutional neural networks (CNNs) have been examined. The proposed TPUAR-Net capitalizes on different levels of global and local features in the upper and lower paths of the proposed model structure. In addition, the proposed method is configured to use the skip connection between layers and residual units to accelerate the training and testing processes. The TPUAR-Net model provides promising segmentation accuracy using MRI images from the BRATS 2017 database, while its parallelized architecture considerably improves the execution speed. The results obtained in terms of Dice, sensitivity, and specificity metrics demonstrate that TPUAR-Net outperforms other methods and achieves the state-of-the-art performance for brain tumor segmentation.

Abstract Using prototypes when developing IT artifacts is not something new. With the help of prototypes, developers get a chance to show what the IT artifact can look like without investing too much time or money. The end user also gets the chance to see the IT artifact before it has been developed and gets a chance to add or remove features and requirements. This study examines how to use prototypes and how the use of prototypes improves the experience of mobile application interfaces for young people. In the study, young people from a leisure center were interviewed regarding the development of a mobile application. Three different prototypes have been presented and data has been collected on those prototypes to then see how the different prototypes can be used in a development process. After data collection, it turned out that the Lo-fi prototype did not contribute as much as the Hi-fi prototypes, so there is no need to develop Lo-fi prototypes in order to obtain data on user experience.

An integration of ICT advances into a conventional healthcare system is spreading extensively nowadays. This trend is known as Electronic health or E-Health. E-Health solutions help to achieve the sustainability goal of increasing the expected lifetime while improving the quality of life by providing a constant healthcare monitoring. Cardiovascular diseases are one of the main killers yearly causing approximately 17.7 million deaths worldwide. The focus of this work is on studying the detection of one of the cardiovascular diseases – Atrial Fibrillation (AF) arrhythmia.  This type of arrhythmia has a severe influence on the heart health conditions and could cause congestive heart failure (CHF), stroke, and even increase the risk of death. Therefore, it is important to detect AF as early as possible. In this thesis we focused on studying various machine learning techniques for AF detection using only short single lead Electrocardiography recordings. A web-based solution was built as a final prototype, which first simulates the reception of a recorded signal, conducts the preprocessing, makes a prediction of the AF presence, and visualizes the result. For the AF detection the relatively high accuracy score was achieved comparable to the one of the state-of-the-art. The work was based on the investigation of the proposed architectures and the usage of the database of signals from the 2017 PhysioNet/CinC Challenge. However, an additional constraint was introduced to the original problem formulation, since the idea of a future deployment on the resource-limited devices places the restrictions on the complexity of the computations being performed for achieving the prediction. Therefore, this constraint was considered during the development phase of the project.

In order to facilitate data-driven solutions for early detection of atrial fibrillation (AF), the 2017 CinC conference challenge was devoted to automatic AF classification based on short ECG recordings. The proposed solutions concentrated on maximizing the classifiers F ₁ score, whereas the complexity of the classifiers was not considered. However, we argue that this must be addressed as complexity places restrictions on the applicability of inexpensive devices for AF monitoring outside hospitals. Therefore, this study investigates the feasibility of complexity reduction by analyzing one of the solutions presented for the challenge.

 Air quality assessment plays a crucial role in understanding the impact of air pollution onhuman health and the environment. With the increasing demand for accurate assessment andprediction of air quality, advanced data analytics modelling techniques offer promisingsolutions. This thesis focuses on leveraging advanced data analytics to assess and analyse airpollution concentration levels in Italy over a 4km resolution using the FORAIR_IT datasetsimulated in ENEA on the CRESCO6 infrastructure, aiming to uncover valuable insights andidentifying the most appropriate AI models for predicting air pollution levels. The datacollection, understanding, and pre-processing procedures are discussed, followed by theapplication of big data training and forecasting using Apache Spark MLlib. The research alsoencompasses different phases, including descriptive and inferential analysis to understand theair pollution concentration dataset, hypothesis testing to examine the relationship betweenvarious pollutants, machine learning prediction using several regression models and anensemble machine learning approach and time series analysis on the entire dataset as well asthree major regions in Italy (Northern Italy – Lombardy, Central Italy – Lazio and SouthernItaly – Campania). The computation time for these regression models are also evaluated and acomparative analysis is done on the results obtained. The evaluation process and theexperimental setup involve the usage of the ENEAGRID/CRESCO6 HPC Infrastructure andApache Spark. This research has provided valuable insights into understanding air pollutionpatterns and improving prediction accuracy. The findings of this study have the potential todrive positive change in environmental management and decision-making processes, ultimatelyleading to healthier and more sustainable communities. As we continue to explore the vastpossibilities offered by advanced data analytics, this research serves as a foundation for futureadvancements in air quality assessment in Italy and the models are transferable to other regionsand provinces in Italy, paving the way for a cleaner and greener future. 

Industry 4.0, as well as the increasing use of artificial intelligence and machine learning, have made it possible to analyse large amounts of data and improve performance across many businesses and sectors. These sectors have significantly increased their reliance on data when making decisions. This thesis examines the influence of data-driven decision-making on road maintenance planning in Sweden. I gathered data related to road state in accordance with the Swedish Road Maintenance Standard, focussing on the International Roughness Index (IRI) and rut depth as primary factors. This data analysis enabled the identification of maintenance needs within three separate time frames: immediate, the next five years, and long-term. Overall, the model predicted maintenance needs based on the International Roughness Index (IRI) with up to 96% accuracy. However, the model's accuracy dropped to only 67% when predicting maintenance needs over the next five years. In contrast, the model that predicted maintenance needs based on rut depth demonstrated high accuracy across all three-time frames, achieving up to 92% accuracy.The model demonstrated that modern road condition variable data are crucial to prediction. In terms of predictions, 2023 IRI measurements were the most important. Based on our findings, this thesis improves data-driven decision-making in Swedish road maintenance, resulting in more effective resource allocation and a decrease in emergency maintenance expenses. Moreover, the study highlights the value of collecting and utilising more accurate and thorough road state data to enhance these models.

Industry 4.0 and the increasing use of artificial intelligence and machine learning have allowed the analysis of large amounts of data and improve performance across many businesses and sectors. These sectors have significantly increased their reliance on data when making decisions. This paper examines the use of data-driven decision-making on road maintenance planning in Sweden. Data related to road maintenance following the Swedish Road Maintenance Standard, were collected focusing on the International Roughness Index (IRI) and rut depth as primary features. Analyzing such data enabled the identification of maintenance needs within three separate timeframes: immediate, the next five years, and long-term. The model predicted maintenance needs based on the IRI with up to 96% accuracy. However, the model's accuracy dropped to only 67% when predicting maintenance needs over the next five years. In contrast, the model that predicted maintenance needs based on rut depth demonstrated high accuracy across all three timeframes, achieving up to 92% accuracy. The model demonstrated that modern road condition variables are crucial to prediction. In terms of predictions, 2023 IRI measurements were the most important. Based on our findings, this paper improves data-driven decision-making in Swedish road maintenance, resulting in more effective resource allocation and decreased emergency maintenance expenses. Moreover, the study highlights the value of collecting and utilizing more accurate and thorough road state data to enhance these models.

We participated in the WMT 2022 Large-Scale Machine Translation Evaluation for the African Languages Shared Task. This work de-scribes our approach, which is based on filtering the given noisy data using a sentence-pair classifier that was built by fine-tuning a pre-trained language model. To train the classifier, we obtain positive samples (i.e. high-quality parallel sentences) from a gold-standard curated dataset and extract negative samples (i.e.low-quality parallel sentences) from automatically aligned parallel data by choosing sentences with low alignment scores. Our final machine translation model was then trained on filtered data, instead of the entire noisy dataset. We empirically validate our approach by evaluating on two common datasets and show that data filtering generally improves overall translation quality, in some cases even significantly.

In the past decade, cloud computing has become one of the essential techniques of many business areas, including social media, online shopping, music streaming, and many more. It is difficult for cloud providers to provision their systems in advance due to fluctuating changes in input workload and resultant resource demand. Therefore, there is a need for auto-scaling technology that can dynamically adjust resource allocation of cloud services based on incoming workload. In this paper, we present a predictive auto-scaler for Kubernetes environments to improve the quality of service. Being based on a proactive model, our proposed auto-scaling method serves as a foundation on which to build scalable and resource-efficient cloud systems.

Wireless Sensor Networks (WSN) have been highly developed which can be used in agriculture to enable optimal irrigation scheduling. Since there is an absence of widely used available methods to support effective agriculture practice in different weather conditions, WSN technology can be used to optimise irrigation in the crop fields. This paper presents architecture of an irrigation system by incorporating interoperable IP based WSN, which uses the protocol stacks and standard of the Internet of Things paradigm. The performance of fundamental issues of this network is emulated in Tmote Sky for 6LoWPAN over IEEE 802.15.4 radio link using the Contiki OS and the Cooja simulator. The simulated results of the performance of the WSN architecture presents the Round Trip Time (RTT) as well as the packet loss of different packet size. In addition, the average power consumption and the radio duty cycle of the sensors are studied. This will facilitate the deployment of a scalable and interoperable multi hop WSN, positioning of border router and to manage power consumption of the sensors.

Wireless Sensor Networks (WSNs) are playing remarkable contribution in real time decision making by actuating the surroundings of environment. As a consequence, the contemporary agriculture is now using WSNs technology for better crop production, such as irrigation scheduling based on moisture level data sensed by the sensors. Since WSNs are deployed in constraints environments, the life time of sensors is very crucial for normal operation of the networks. In this regard routing protocol is a prime factor for the prolonged life time of sensors. This research focuses the performances analysis of some clustering based routing protocols to select the best routing protocol. Four algorithms are considered, namely Low Energy Adaptive Clustering Hierarchy (LEACH), Threshold Sensitive Energy Efficient sensor Network (TEEN), Stable Election Protocol (SEP) and Energy Aware Multi Hop Multi Path (EAMMH). The simulation is carried out in Matlab framework by using the mathematical models of those algortihms in heterogeneous environment. The performance metrics which are considered are stability period, network lifetime, number of dead nodes per round, number of cluster heads (CH) per round, throughput and average residual energy of node. The experimental results illustrate that TEEN provides greater stable region and lifetime than the others while SEP ensures more througput.

Because of the increased popularity and fast expansion of the Internet as well as Internet of things, networks are growing rapidly in every corner of the society. As a result, huge amount of data is travelling across the computer networks that lead to the vulnerability of data integrity, confidentiality and reliability. So, network security is a burning issue to keep the integrity of systems and data. The traditional security guards such as firewalls with access control lists are not anymore enough to secure systems. To address the drawbacks of traditional Intrusion Detection Systems (IDSs), artificial intelligence and machine learning based models open up new opportunity to classify abnormal traffic as anomaly with a self-learning capability. Many supervised learning models have been adopted to detect anomaly from networks traffic. In quest to select a good learning model in terms of precision, recall, area under receiver operating curve, accuracy, F-score and model built time, this paper illustrates the performance comparison between Naïve Bayes, Multilayer Perceptron, J48, Naïve Bayes Tree, and Random Forest classification models. These models are trained and tested on three subsets of features derived from the original benchmark network intrusion detection dataset, NSL-KDD. The three subsets are derived by applying different attributes evaluator’s algorithms. The simulation is carried out by using the WEKA data mining tool.

The thesis focuses on machine learning methods for Earth Observation (EO) data, more specifically, remote sensing data acquired by satellites and drones. EO plays a vital role in monitoring the Earth’s surface and modelling climate change to take necessary precautionary measures. Initially, these efforts were dominated by methods relying on handcrafted features and expert knowledge. The recent advances of machine learning methods, however, have also led to successful applications in EO. This thesis explores supervised and unsupervised approaches of Deep Learning (DL) to monitor natural resources of water bodies and forests. 

The first study of this thesis introduces an Unsupervised Curriculum Learning (UCL) method based on widely-used DL models to classify water resources from RGB remote sensing imagery. In traditional settings, human experts labeled images to train the deep models which is costly and time-consuming. UCL, instead, can learn the features progressively in an unsupervised fashion from the data, reducing the exhausting efforts of labeling. Three datasets of varying resolution are used to evaluate UCL and show its effectiveness: SAT-6, EuroSAT, and PakSAT. UCL outperforms the supervised methods in domain adaptation, which demonstrates the effectiveness of the proposed algorithm. 

The subsequent study is an extension of UCL for the multispectral imagery of Australian wildfires. This study has used multispectral Sentinel-2 imagery to create the dataset for the forest fires ravaging Australia in late 2019 and early 2020. 12 out of the 13 spectral bands of Sentinel-2 are concatenated in a way to make them suitable as a three-channel input to the unsupervised architecture. The unsupervised model then classified the patches as either burnt or not burnt. This work attains 87% F1-Score mapping the burnt regions of Australia, demonstrating the effectiveness of the proposed method. 

The main contributions of this work are (i) the creation of two datasets using Sentinel-2 Imagery, PakSAT dataset and Australian Forest Fire dataset; (ii) the introduction of UCL that learns the features progressively without the need of labelled data; and (iii) experimentation on relevant datasets for water body and forest fire classification. 

This work focuses on patch-level classification which could in future be expanded to pixel-based classification. Moreover, the methods proposed in this study can be extended to the multi-class classification of aerial imagery. Further possible future directions include the combination of geo-referenced meteorological and remotely sensed image data to explore proposed methods. Lastly, the proposed method can also be adapted to other domains involving multi-spectral and multi-modal input, such as, historical documents analysis, forgery detection in documents, and Natural Language Processing (NLP) classification tasks.

Earth Observation (EO) data is crucial for understanding, managing, and conserving our planet's ecosystem and its natural resources. This data enables humanity to monitor environmental changes, such as natural disasters, urban growth, and climate shifts, assisting informed decisions and proactive measures. Early EO heavily relied on statistical methods and expert domain knowledge, but the advent of machine learning has revolutionized EO data processing, enhancing efficiency and accuracy. Conventional ML models require expensive and labor-intensive data labeling. In contrast, unsupervised ML techniques can learn features from data without the need for manual labeling, making the process more efficient and cost-effective.

This thesis presents a UCL approach utilizing advanced DL models to classify EO data, referred to as UCL4EO. This approach eliminates the need for manual data labeling in training the DL model. The UCL framework comprises i) a DL model tailored for feature extraction from image data, ii) a clustering method to group deep features, and iii) a selection operation to capture representative samples from these clusters. The CNN extracts meaningful features from images, subjected to a clustering algorithm to create pseudo-labels. After identifying the initial clusters, representative samples from each cluster are chosen using the UCL selection operation to fine-tune the feature extractor. The stated process is repeated iteratively until convergence. The proposed UCL approach progressively learns and incorporates salient data features in an unsupervised manner by utilizing pseudo-labels.

UCL started as a proof of concept to show the viability of the method for binary classification on RS and aerial imagery. Specifically, the UCL framework is employed to identify water bodies using three RGB datasets, encompassing both low and high-resolution RS and aerial imagery. While UCL has been extensively examined with RGB imagery, it has been adapted to benefit from the enhanced capabilities of multi-spectral satellite imagery. This adaptation enables UCL to generalize to multi-spectral imagery from Sentinel-2 to detect forest fires in Australia. UCL undergoes subsequent improvements and is further investigated to identify utility poles in high-resolution UAV images. These gray-scale images of utility poles pose computer vision challenges, including issues like occlusion and cropping, where a significant portion of the image contains the background and only a slight appearance of the utility pole. Extensive experimentation on the mentioned tasks effectively showcases UCL's adaptive learning capabilities, producing promising results. The achieved accuracy surpassed those of supervised methods in cross-domain adaptation on similar tasks, underscoring the effectiveness of the proposed algorithm.

The scope of UCL has been extended to encompass multi-class classification tasks in the domain of RS data, referred to as Multi-class UCL. Multi-class UCL progressively acquires knowledge about various categories on multi-scale resolution. To investigate Multi-class UCL, we have used four publicly available datasets of Sentinel-2 and aerial imagery: EuroSAT, SAT-6, UCMerced, and RSSCN7. Comprehensive experiments conducted on the above-mentioned datasets revealed better cross-domain adaptation capabilities compared to supervised methods, thereby demonstrating the effectiveness of Multi-class UCL.

In these investigations, two datasets are generated using Sentinel-2 satellite imagery: one for water bodies - PakSAT and the other for Australian forest fires. However, cloud cover poses a significant challenge by obstructing the satellite's ability to capture clear images of the Earth's surface. To address this issue, available cloud masking techniques are employed to filter out images affected by cloud cover, ensuring the datasets contain only clear and usable data. Later, this thesis examines cloud detection and Cloud Optical Thickness (COT) estimation from Sentinel-2 imagery. We employed machine-learning techniques, achieving better performance than SCL designed by ESA for cloud cover tasks.

In addition to the application in RS data, UCL has been investigated in other domains of EO, such as undersea imagery. Furthermore, UCL has also been used for tasks like natural scene classification, medical imaging, and document analysis, demonstrating its versatility and broad applicability. Further exploration of UCL could involve improving the process of generating pseudo-labels through deep learning techniques.

The effective training of supervised deep learning models requires the labeling of extensive datasets, a process that is often costly and labor-intensive. Such models also face significant challenges with overfitting on the training data with true labels, leading to suboptimal performance on new datasets with slight variations in capturing sources or regions. This paper introduces Multi-class Unsupervised Curriculum Learning (Multi-class UCL), a novel deep learning framework. We demonstrate the effectiveness of this framework on the  case study of land use and cover classification that bypasses the need for labeled data, thereby improving adaptability across different datasets. Multi-class UCL leverages pseudo-labels generated from a clustering technique to train the model and incorporates a selection process that ensures an equal representation of samples from each cluster, addressing the issue of class imbalance. The study evaluates the effectiveness of Multi-class UCL through comprehensive experiments on four diverse publicly available datasets: EuroSAT, SAT-6, RSSCN7, and UCMerced. These datasets have varying resolutions, come from different capturing sources, and encompass different geographical areas.The results demonstrate that the framework effectively learns and generalizes important features from the data, showing superior adaptability and performance across various datasets compared to traditional supervised models.

In many real-world applications of computer vision complex domains, such as medical diagnostics and document analysis, the lack of labeled data often limits the effectiveness of traditional deep learning models. This study addresses these challenges by enhancing Unsupervised Curriculum Learning (UCL), a deep learning framework that automatically discovers meaningful patterns without the need for labeled data. Originally designed for remote sensing imagery, UCL has been expanded in this work to improve classification performance in a variety of domain-specific applications. UCL integrates a convolutional neural network, clustering algorithms, and selection techniques to classify images unsupervised. We introduce key improvements, such as spectral clustering, outlier detection, and dimensionality reduction, to boost the framework’s accuracy. Experimental results demonstrate significant performance gains, with F1-scores increasing from 68% to 94% on a three-class subset of the CIFAR-10 dataset and from 68% to 75% on a five-class subset. The updated UCL also achieved F1-scores of 85% in medical diagnosis, 82% in scene recognition, and 62% in historical document classification. These findings underscore the potential of UCL in complex real-world applications and point to areas where further advancements are needed to maximize its utility across diverse fields.

This paper introduces a machine learning-based approach for detecting electric poles, an essential part of power grid maintenance. With the increasing popularity of deep learning, several such approaches have been proposed for electric pole detection. However, most of these approaches are supervised, requiring a large amount of labeled data, which is time-consuming and labor-intensive. Unsupervised deep learning approaches have the potential to overcome the need for huge amounts of training data. This paper presents an unsupervised deep learning framework for utility pole detection. The framework combines Convolutional Neural Network (CNN) and clustering algorithms with a selection operation. The CNN architecture for extracting meaningful features from aerial imagery, a clustering algorithm for generating pseudo labels for the resulting features, and a selection operation to filter out reliable samples to fine-tune the CNN architecture further. The fine-tuned version then replaces the initial CNN model, thus improving the framework, and we iteratively repeat this process so that the model learns the prominent patterns in the data progressively. The presented framework is trained and tested on a small dataset of utility poles provided by “Mention Fuvex” (a Spanish company utilizing long-range drones for power line inspection). Our extensive experimentation demonstrates the progressive learning behavior of the proposed method and results in promising classification scores with significance test having p−value<0.00005 on the utility pole dataset.

Massive wildfires not only in Australia, but also worldwide are burning millions of hectares of forests and green land affecting the social, ecological, and economical situation. Widely used indices-based threshold methods like Normalized Burned Ratio (NBR) require a huge amount of data preprocessing and are specific to the data capturing source. State-of-the-art deep learning models, on the other hand, are supervised and require domain experts knowledge for labeling the data in huge quantity. These limitations make the existing models difficult to be adaptable to new variations in the data and capturing sources. In this work, we have proposed an unsupervised deep learning based architecture to map the burnt regions of forests by learning features progressively. The model considers small patches of satellite imagery and classifies them into burnt and not burnt. These small patches are concatenated into binary masks to segment out the burnt region of the forests. The proposed system is composed of two modules: 1) a state-of-the-art deep learning architecture for feature extraction and 2) a clustering algorithm for the generation of pseudo labels to train the deep learning architecture. The proposed method is capable of learning the features progressively in an unsupervised fashion from the data with pseudo labels, reducing the exhausting efforts of data labeling that requires expert knowledge. We have used the realtime data of Sentinel-2 for training the model and mapping the burnt regions. The obtained F1-Score of 0.87 demonstrates the effectiveness of the proposed model.

Seagrass ecosystems are pivotal in marine environments, serving as crucial habitats for diverse marine species and contributing significantly to carbon sequestration. Accurate classification of seagrass species from underwater images is imperative for monitoring and preserving these ecosystems. This paper introduces Unsupervised Curriculum Learning (UCL) to seagrass classification using the DeepSeagrass dataset. UCL progressively learns from simpler to more complex examples, enhancing the model's ability to discern seagrass features in a curriculum-driven manner. Experiments employing state-of-the-art deep learning architectures, convolutional neural networks (CNNs), show that UCL achieved overall 90.12 % precision and 89 % recall, which significantly improves classification accuracy and robustness, outperforming some traditional supervised learning approaches like SimCLR, and unsupervised approaches like Zero-shot CLIP. The methodology of UCL involves four main steps: high-dimensional feature extraction, pseudo-label generation through clustering, reliable sample selection, and fine-tuning the model. The iterative UCL framework refines CNN's learning of underwater images, demonstrating superior accuracy, generalization, and adaptability to unseen seagrass and background samples of undersea images. The findings presented in this paper contribute to the advancement of seagrass classification techniques, providing valuable insights into the conservation and management of marine ecosystems. The code and dataset are made publicly available and can be assessed here: https://github.com/nabid69/Unsupervised-Curriculum-Learning—UCL.

This paper presents a Convolutional Neural Networks (CNN) based Unsupervised Curriculum Learning approach for the recognition of water bodies to overcome the stated challenges for remote sensing based RGB imagery. The unsupervised nature of the presented algorithm eliminates the need for labelled training data. The problem is cast as a two class clustering problem (water and non-water), while clustering is done on deep features obtained by a pre-trained CNN. After initial clusters have been identified, representative samples from each cluster are chosen by the unsupervised curriculum learning algorithm for fine-tuning the feature extractor. The stated process is repeated iteratively until convergence. Three datasets have been used to evaluate the approach and show its effectiveness on varying scales: (i) SAT-6 dataset comprising high resolution aircraft images, (ii) Sentinel-2 of EuroSAT, comprising remote sensing images with low resolution, and (iii) PakSAT, a new dataset we created for this study. PakSAT is the first Pakistani Sentinel-2 dataset designed to classify water bodies of Pakistan. Extensive experiments on these datasets demonstrate the progressive learning behaviour of UCL and reported promising results of water classification on all three datasets. The obtained accuracies outperform the supervised methods in domain adaptation, demonstrating the effectiveness of the proposed algorithm.

The problem of full-order observer design for a class of delayed nonlinear systems with unknown inputs and Markovian jumping parameters is considered. The design method is formulated as solving a set of linear matrix inequalities (LMI's). Extending the results of nonlinear observer design to Markovian jump systems with time-varying delays is the main advantages of this paper. The sufficient LMI conditions are dependent on both the upper and lower bounds of delay. The effectiveness of the proposed method is shown via an illustrative example.

In this paper, a distributed home automation system will be demonstrated. Traditional systems are based on a central controller where all the decisions are made. The proposed control architecture is a solution to overcome the problems such as the lack of flexibility and re-configurability that most of the conventional systems have. This has been achieved by employing a method based on the new IEC 61499 function block standard, which is proposed for distributed control systems. This paper also proposes a wireless sensor network as the system infrastructure in addition to the function blocks in order to implement the Internet-of-Things technology into the area of home automation as a solution for distributed monitoring and control. The proposed system has been implemented in both Cyber (nxtControl) and Physical (Contiki-OS) level to show the applicability of the solution

A Monte Carlo simulation was used to study the outflow of O+ and H+ ions along three flight trajectories above the polar cap up to altitudes of about 15 RE. Barghouthi (2008) developed a model on the basis of altitude and velocity-dependent wave-particle interactions and a radial geomagnetic field which includes the effects of ambipolar electric field and gravitational and mirror forces. In the present work we improve this model to include the effect of the centrifugal force, with the use of relevant boundary conditions. In addition, the magnetic field and flight trajectories, namely, the central polar cap (CPC), nightside polar cap (NPC), and cusp, were calculated using the Tsyganenko T96 model. To simulate wave-particle interactions, the perpendicular velocity diffusion coefficients for O+ ions in each region were determined such that the simulation results fit the observations. For H+ ions, a constant perpendicular velocity diffusion coefficient was assumed for all altitudes in all regions as recommended by Nilsson et al. (2013). The effect of centrifugal acceleration was simulated by considering three values for the ionospheric electric field: 0 (no centrifugal acceleration), 50, and 100 mV/m. It was found that the centrifugal acceleration increases the parallel bulk velocity and decreases the parallel and perpendicular temperatures of both ion species at altitudes above about 4 RE. Centrifugal acceleration also increases the temperature anisotropy at high altitudes. At a given altitude, centrifugal acceleration decreases the density of H+ ions while it increases the density of O+ ions. This implies that with higher centrifugal acceleration more O+ ions overcome the potential barrier. It was also found that aside from two exceptions centrifugal acceleration has the same effect on the velocities of both ions. This implies that the centrifugal acceleration is universal for all particles. The parallel bulk velocities at a given value of ionospheric electric field were highest in the cusp followed by the CPC followed by the NPC. In this study a region of no wave-particle interaction was assumed in the CPC and NPC between 3.7 and 7.5 RE. In this region the perpendicular temperature was found to decrease with altitude due to perpendicular adiabatic cooling.

This book provides a comprehensive overview of computational intelligence methods for semantic knowledge management. Contrary to popular belief, the methods for semantic management of information were created several decades ago, long before the birth of the Internet. In fact, it was back in 1945 when Vannevar Bush introduced the idea for the first protohypertext: the MEMEX (MEMory + indEX) machine. In the years that followed, Bush’s idea influenced the development of early hypertext systems until, in the 1980s, Tim Berners Lee developed the idea of the World Wide Web (WWW) as it is known today. From then on, there was an exponential growth in research and industrial activities related to the semantic management of the information and its exploitation in different application domains, such as healthcare, e-learning and energy management. 

However, semantics methods are not yet able to address some of the problems that naturally characterize knowledge management, such as the vagueness and uncertainty of information. This book reveals how computational intelligence methodologies, due to their natural inclination to deal with imprecision and partial truth, are opening new positive scenarios for designing innovative semantic knowledge management architectures.

The worldwide electronics market is focusing on developing innovative technologies that can lead to denser, more resilient, and tighter board-level integration. The consumer electronics market is trending toward miniaturization, with HDI-PCBs dominating. Electronics shrinking and scaling technology is the prime concern of all manufacturers. The PCBA industry is transforming its production practices which can reduce the solder joints, limit the usage of discrete and bulky components, reduce the packaging factor of printed boards by accommodating the maximum number of ICs, minimize the assembly span, optimize the latency, and so on. However, developments in production processes in the PCB manufacturing industry need more attention than those in  Silicon-based (ICs) fabrications. One of the issues in PCB fabrication is utilizing conventional metallization approaches. The majority of manufacturers continue to use standard Copper(Cu) laminates on the base substrate and lithography methods to shape the structures.In recent manufacturing technologies, semi-Additive process (SAP) or modified-SAP (mSAP) methods are being adopted to replace traditional subtractive print-and-etch procedures. To scale down the Lines and Spaces (L\&S) on PCBs comparable to that of IC-level, most smartphone makers use Substrate-like PCB (SLP) using mSAP methods. However, subtractive patterning has been used in the intermediate stages of fabrication in those methods. This thesis demonstrates a fully additive selective metallization-based production approach to bridge this technology gap between IC-level and board-level fabrications. The fabrication process has given the name 'Sequential Build-Up Covalent Bonded Metallisation' (SBU-CBM) method.

This dissertation presents a new approach to Cu metallization using a significant step reducing-pattern-transfer process. The patterning method activates a seed layer of CBM polymer chains on a polymer surface with optimal UV-Laser settings. This surface modification enables a strong Copper (Cu) bonding onto the modified surface by Cu-plating. The suggested approach generated a 2.5D surface pattern using a micrometer via laser ablation and subsequent sub-micrometer laser lithography. Furthermore, the surface characterization of each step involved in the fabrication process is analysed and presented to show the sequential growth of layers on top of each other. To investigate the mechanism of the process at the interfaces, characterizations such as EDS, SEM, and XRD characterizations were performed. This PCB manufacturing method can selectively add metallic layers to the finest feature sizes at considerably lower temperatures. Overall, the thesis has addressed two critical aspects i.e. miniaturization of interconnects at board-level and the feasibility of a fully-additive production approach for electronics packaging.

First, a subtractive method is shown to achieve Copper interconnects with feature size 3.0$\mu$m. This miniaturization corresponds to 70\% reduction in the feature size from 20 $\mu$m to 3 $\mu$m. Next, the proposed additive production process has produced Cu interconnects with feature sizes of 2.5 $\mu$m L\&S and via of diameter 10 $\mu$m. The scaling of the interconnects was achieved by optimizing the process parameters involved in the proposed fabrication recipe.

Second, the sequential build-up (SBU) procedure is adopted to realize the embedded passives with the minimum possible feature size ($<$ 10 $\mu$m). An embedded capacitor and a planar inductor were fabricated. The proposed method can be employed to achieve any desirable pattern on FR-4, and a few of them are shown in the thesis. This additive technique can further be investigated through electrical and reliability assessment to make it an industrially accepted method.

Consumer electronics market is escalating towards the miniaturization and the use of HDI-PCBs is dominating. Thus, the production technologies are adapting the Semi-Additive process (SAP) or modified-SAP (mSAP) methods over conventional subtractive print-and-etch methods. Most of the Smartphone manufacturers are using Substrate-like PCB (SLP) with mSAP techniques to scale down the Lines and Spaces (L&S) on PCBs equivalent to ICs. However, those processes still involve subtractive patterning in the intermediate stages of fabrication. In this paper, a fully additive multi-layer patterning process using an electroless copper plating has been investigated. This patterning process is based on modifying a polymer surface by activating a seed layer of grafting polymer chains on it using optimized UV-Laser parameters. This surface modification enables a strong bonding of Copper (Cu) onto the modified surface by Cu-plating. Using a micrometer via laser ablation and subsequent sub-micrometer laser lithography a 2.5D surface pattern has been achieved with the proposed technique.So far, using the proposed additive production process the feature sizes of 2.5 μm L&S and via of diameter 10 μm have been achieved.The via ablation and pattering were done by using 266nm and 375nm laser sources respectively.The substrates used are standard FR4 material and a layer of polyurethane of thickness 35μm coated on top of it. Analysis of the process parameters and their optimization has been done by factorial design method using Design Expert 12.0 software to show their contribution and significance in the production process.

Advancements in production techniques in PCB manufacturing industries are still required as compared to silicon-ICs fabrications. One of the concerned areas in PCBs fabrication is the use of conventional methodologies for metallization. Most of the manufacturers are still using the traditional Copper (Cu) laminates on the base substrate and patterning the structures using lithography processes. As a result, significant amounts of metallic parts are etched away during any mass production process, causing unnecessary disposables leading to pollution. In this work, a new approach for Cu metallization is demonstrated with considerable step-reducing pattern-transfer mechanism. In the fabrication steps, a seed layer of covalent bonded metallization (CBM) chemistry on top of a dielectric epoxy resin is polymerized using actinic radiation intensity of a 375 nm UV laser source. The proposed method is capable of patterning any desirable geometries using the above-mentioned surface modification followed by metallization. To metallize the patterns, a proprietary electroless bath has been used. The metallic layer grows only on the selective polymer-activated locations and thus is called selective metallization. The highlight of this production technique is its occurrence at a low temperature (20–45 °C). In this paper, FR-4 as a base substrate and polyurethane (PU) as epoxy resin were used to achieve various geometries, useful in electronics packaging. In addition, analysis of the process parameters and some challenges witnessed during the process development are also outlined. As a use case, a planar inductor is fabricated to demonstrate the application of the proposed technique.

Miniaturization is the call of the day. Electronics shrinking and scaling technology is the priority of all manufacturers. PCBA Industry is working towards the elimination of solder joints, reduction in use of discrete and bulky components, lowering of assemble span, minimized latency etc. Embedded passive technology is playing a significant role in this roadmap by providing better signal performance, reduced parasitic and crosstalk. In this work, the primary focus is to develop a cost-efficient and flexible fabrication methodology that will be suitable for bulk production. A sequential build up (SBU) procedure is adopted with an additive lithography process to realize the passives with minimum possible feature size (<; 10 μm). A low cost insulating material, promising grafting solution and Laser assisted writing machine with optimized fabrication parameters are the highlights of this production method. A Computer Aided Design (CAD) software i.e. clewin is used during this process to pattern the mask for the entire process. Covalent bonded metallization (CBM) is the key process for the adhesion of copper layer on the desired site of the pattern. In the CBM process, a polymer surface is modified by grafting. The position of the surface modification is optically defined using a laser lithography system. Such surface modified samples are, then treated in an electroless copper process. Resulting in copper metallization only at the locations with a CBM modified surface. The verification of the copper deposition on the substrate is investigated using a high-resolution microscope followed by scanning electron microscopy (SEM). The confirmation of passive formation has been checked using kethley's source (electrical two-probe measurement). The first-order measured results showed the capacitance formed in the range of 0.3-8 pF. Further concrete measurements using standard methods are undergoing. One of the key advantage of this proposed process is its easiness and feasibility of at room temperature.

As per the latest roadmap of iNEMI, the global electronics market is emphasizing to identify disruptive technologies that can contribute towards denser, robust and tighter integration on the board level. Therefore, reduction in packaging factor of printed board can accommodate greater number of ICs to support miniaturization. This paper has shown an experimental method to pattern the metallic layer on a Printed circuit Board (PCB) to the smallest feature size. To investigate this, a commercially available FR-4 PCB with photosensitive material coat and a Copper (Cu) layer on it, is used. A reverse-mode Laser assisted writing is implemented to pattern the desired copper tracks. Soon after, a well-controlled development and chemical etching of the Laser-activated regions are done using Sodium Hydroxide solution followed by an aqueous solution of Sodium Persulfate. Current PCB interconnects used by the industries are of the order (~20 μm). Whereas the present work is a contribution towards achieving Copper interconnects with feature size 3.0μm. This miniaturization corresponds to 70% reduction in the feature size from 20 μm to 3μm. The natural adhesion of the Cu layer has remained intact even after the etching, shows the efficiency of the method adopted. Also, variation in the parameters such as etching time, etchant solution concentrations, temaperature, gain and exposure time of Laser beam and their corresponding effects are discussed. Other highlights of this subtractive method includes its cost-efficiency, lesser production time and repeatability.

To bridge the technology gap between IC-level and board-level fabrications, a fully additive selective metallization has already been demonstrated in the literature. In this article, the surface characterization of each step involved in the fabrication process is outlined with bulk metallization of the surface. This production technique has used polyurethane as epoxy resin and proprietary grafting chemistry to functionalize the surface with covalent bonds on an FR-4 base substrate. The surface was then metalized using an electroless copper (Cu) bath. This sequential growth of layers on top of each other using an actinic laser beam and palladium (Pd) ions to deposit Cu is analyzed. State-of-the-art material characterization techniques were employed to investigate process mechanism at the interfaces. Density functional theory calculations were performed to validate the experimental evidence of covalent bonding of the layers. This manufacturing approach is capable of adding metallic layers in a selective manner to the printed circuit boards at considerably lower temperatures. A complete analysis of the process using bulk deposition of the materials is illustrated in this work.

Despite evidence of rising popularity of video on the web (or VOW), little is known about how users access video. However, such a characterization can greatly benefit the design of multimedia systems such as web video proxies and VOW servers. Hence, this paper presents an analysis of trace data obtained from an ongoing VOW experiment in Lulea University of Technology, Sweden. This experiment is unique as video material is distributed over a high bandwidth network allowing users to make access decisions without the network being a major factor. Our analysis revealed a number of interesting discoveries regarding user VOW access. For example, accesses display high temporal locality: several requests for the same video title often occur within a short time span. Accesses also exhibited spatial locality of reference whereby a small number of machines accounted for a large number of overall requests. Another finding was a browsing pattern where users preview the initial portion of a video to find out if they are interested. If they like it, they continue watching, otherwise they halt it. This pattern suggests that caching the first several minutes of video data should prove effective. Lastly, the analysis shows that, contrary to previous studies, ranking of video titles by popularity did not fit a Zipfian distribution.

IoT, IIoT, 5G technologies are a source of huge revenue streams to many existing and upcoming players across several industries globally. IoT specifically promises granularity of data, real-time visibility of information, and defying geographical and natural obstacles to allow collecting, processing, and storing large amounts of useful data. Furthermore, there are many upcoming and ever-evolving mandatory standards and regulations such as NIS2, CRA, EU MDR, and many others with regards to the security and safety of some IoT technologies. For such devices to qualify and get to the market, a set of tests and assessments of their security posture will be required. There are also several other technological challenges such as interoperability that need to be tested and approved to be allowed for sale as products in some instances. Therefore, the motivation for this thesis is to develop a unified structure for assessing IoT security posture that responds and accommodates for the constantly changing regulatory and industrial landscape of requirements.

The main approach used was to explore the already existing methodologies, standards, regulations, and best practices regarding security testing and assessment of products, we further examined how these have been used with regards to IoT testing to produce an all-in-one framework to evaluate the security posture of IoT products. The results were a unified frame of reference to evaluate industry methodologies, standards regulations, and best practices for use in security assessment of IoT devices. Secondly, we proposed a frame of reference in this thesis that resulted from the evaluation of IoT technologies, architecture, known IoT security issues, and industry security assessment standards, regulations, and best practices. The framework specifies the main steps involved in security assessment, the current useful tools to conduct assessments, and several techniques to do so.  Thirdly, from the security assessment of the chosen IoT device, an x5 mini spy wireless camera, a threat analysis and risk assessment were performed and documented. The device security was found to be poor and corrective security controls to strengthen device security were also proposed.

In summary, we explored existing industry security assessment methodologies, standards, and best practices with a goal of drawing out a unified framework to tackle the security assessment and penetration testing challenges for IoT devices. It was found out that there is hardly a single method, tool, or technique to use to sufficiently analyze IoT security with accuracy and reproducibility. Rather, employing more than one method, tool and technique provides a broader and more comprehensive assessment of potential vulnerabilities. This approach ensures cross-referencing of results so that no single method's limitations get to compromise the whole assessment, which eventually leads to achieving higher precision, reliability and consistency in the security evaluation of IoT systems. 

Our contribution is a framework against which we analyzed existing assessment standards and methodologies. Subsequently, we adapted the framework for IoT security assessment and used it to conduct a proof of concept on a chosen IoT device. This research lays the ground for future examination and inclusion of several other standards, frameworks, and technologies such as OWASP and cloud. Secondly, the tools to conduct assessments are evolving faster than the techniques due to the ever-evolving technological landscape. The techniques and tools we have suggested in the framework should be used interchangeably; to verify the assessment and it is therefore not a good idea to depend on a result generated using a single tool or techniques.

Online social media has completely transformed how we communicate with each other. While online discussion platforms are available in the form of applications and websites, an emergent outcome of this transformation is the phenomenon of ‘opinion leaders’. A number of previous studies have been presented to identify opinion leaders in online discussion networks. In particular, Feng (2016 Comput. Hum. Behav. 54, 43–53. (doi:10.1016/j.chb.2015.07.052)) has identified five different types of central users besides outlining their communication patterns in an online communication network. However, the presented work focuses on a limited time span. The question remains as to whether similar communication patterns exist that will stand the test of time over longer periods. Here, we present a critical analysis of the Feng framework both for short-term as well as for longer periods. Additionally, for validation, we take another case study presented by Udanor et al. (2016 Program 50, 481–507. (doi:10.1108/PROG-02-2016-0011)) to further understand these dynamics. Results indicate that not all Feng-based central users may be identifiable in the longer term. Conversation starter and influencers were noted as opinion leaders in the network. These users play an important role as information sources in long-term discussions. Whereas network builder and active engager help in connecting otherwise sparse communities. Furthermore, we discuss the changing positions of opinion leaders and their power to keep isolates interested in an online discussion network.