| 1. |
- Abid, Nosheen, 1993-
(författare)
-
Unsupervised Curriculum Learning Case Study: Earth Observation UCL4EO
- 2024
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- Earth Observation (EO) data, collected via satellites and remote sensing technologies, is crucial for understanding, managing, and conserving the Earth. It enables humankind to monitor environmental changes, such as natural disasters, urban growth, and climate shifts, assisting informed decisions and proactive measures. Early Remote Sensing (RS) 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 machine learning (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 an innovative Unsupervised Curriculum Learning (UCL) approach utilizing advanced deep learning (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, typically a Convolutional Neural Network (CNN) tailored for feature extraction from image data, ii) a clustering technique to cluster deep features, and iii) a selection operation to select 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 serves as a proof of concept in a simpler setting of detection tasks 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.In these investigations, two datasets are generated using Sentinel-2: one for water bodies - PakSAT and the other for Australian forest fire. Cloud cover significantly hinders the acquisition of satellite imagery depicting the Earth's surface. In preparing these datasets, this work employs available cloud masking solutions to avoid the images with cloud cover. Later, this thesis examines cloud detection and Cloud Optical Thickness (COT) estimation from Sentinel-2 imagery. We employed advanced machine-learning techniques, achieving state-of-the-art performance for cloud cover tasks.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 three publicly available datasets of Sentinel-2 and aerial imagery: EuroSAT, SAT-6, and RSSCN7. The evaluation of Multi-class UCL’s performance incorporates the concept of a confusion matrix to compare the predicted labels with the actual labels. Comprehensive experiments conducted on the specified datasets revealed better cross-domain adaptation capabilities compared to supervised methods, thereby demonstrating the effectiveness of Multi-class UCL.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.
|
|
| 2. |
- Afzaal, Muhammad, 1989-
(författare)
-
Explainable AI for Data-Driven Feedback and Intelligent Action Recommendations to Support Student Self-Regulation
- 2024
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- Self-regulated learning (SRL) is a cognitive ability with demonstrable significance in facilitating students’ ability to effectively strategize, monitor, and assess their own learning actions. Studies have indicated that a lack of selfregulated learning skills negatively impacts students’ academic performance. Effective data-driven feedback and action recommendations are considered crucial for SRL and significantly influence student learning and performance. However, the task of delivering personalised feedback to every student poses a significant challenge for teachers. Moreover, the task of identifying appropriate learning activities and resources for individualised recommendations poses a significant challenge for teachers, given the large number of students enrolled in most courses.To address these challenges, several studies have examined how learning analytics-based dashboards can support students’ self-regulation. These dashboards offered several visualisations (as feedback) on student success and failure. However, while such feedback may be beneficial, it does not offer insightful information or actionable recommendations to help students improve academically. Explainable artificial intelligence (xAI) approaches have been proposed to explain such feedback and generate insights from predictive models, with a focus on the relevant actions a student needs to take to improve in ongoing courses. Such intelligent activities could be offered to students as data-driven behavioural change recommendations.This thesis offers an xAI-based approach that predicts course performance and computes informative feedback and actionable recommendations to promote student self-regulation. Unlike previous research, this thesis integrates a predictive approach with an xAI approach to analyse and manipulate students’ learning trajectories. The aim is to offer detailed, data-driven actionable feedback to students by providing in-depth insights and explanations for the predictions provided by the approach. The technique provides students with more practical and useful knowledge compared to the predictions alone.The proposed approach was implemented in the form of a dashboard to support self-regulation by students in university courses, and it was evaluated to determine its effects on the students’ academic performance. The results revealed that the dashboard significantly enhanced students’ learning achievements and improved their self-regulated learning skills. Furthermore, it was found that the recommendations generated by the proposed approach positively affected students’ performance and assisted them in self-regulation.
|
|
| 3. |
- Ahmadpanah, Seyed Mohammad Mehdi, 1996
(författare)
-
Language-Based Security and Privacy in Web-driven Systems
- 2024
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- Modular programming is a core principle in software development, which demands reducing design complexity through independent code modules. A prime example of modular programming is systems offering various services and applications accessible through the web. Their complex nature, heavy dependence on third-party modules, and large user base call for principled approaches to user security and privacy. This thesis focuses on securing web-driven systems, practically targeting Trigger-Action Platforms (TAPs) and browser extensions. Both increasingly popular systems empower users to develop and publish applications that enhance digital lives through smart automation and personalized web browsing, respectively. Our approach to software security and privacy is through the lens of programming-language techniques. We identify vulnerabilities in popular TAP applications and prevent malicious behavior by sandboxing and fine-grained access control. To minimize data access for TAPs with user-configured applications, we also present a construction-by-design paradigm for on-demand data minimization using lazy computation. Besides access control and minimization, we study how sensitive information is processed once access is granted, using information-flow analysis. We identify privacy risks in browser extensions, such as exfiltration of cookies and browsing history over the network. We develop a static analysis framework to track flows from user-sensitive data to network requests in browser extensions. Moreover, we revisit information-flow policies that are not necessarily transitive, supporting coarse-grained policies where security labels are specified at the level of modules. We leverage flow-sensitive type systems to enforce granular security in module-based systems.
|
|
| 4. |
- Al-Saedi, Ahmed Abbas Mohsin, 1980-
(författare)
-
Resource-Aware and Personalized Federated Learning via Clustering Analysis
- 2024
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- Today’s advancement in Artificial Intelligence (AI) enables training Machine Learning (ML) models on the daily-produced data by connected edge devices. To make the most of the data stored on the device, conventional ML approaches require gathering all individual data sets and transferring them to a central location to train a common model. However, centralizing data incurs significant costs related to communication, network resource utilization, high volume of traffic, and privacy issues. To address the aforementioned challenges, Federated Learning (FL) is employed as a novel approach to train a shared model on decentralized edge devices while preserving privacy. Despite the significant potential of FL, it still requires considerable resources such as time, computational power, energy, and bandwidth availability. More importantly, the computational capabilities of the training devices may vary over time. Furthermore, the devices involved in the training process of FL may have distinct training datasets that differ in terms of their size and distribution. As a result of this, the convergence of the FL models may become unstable and slow. These differences can influence the FL process and ultimately lead to suboptimal model performance within a heterogeneous federated network.In this thesis, we have tackled several of the aforementioned challenges. Initially, a FL algorithm is proposed that utilizes cluster analysis to address the problem of communication overhead. This issue poses a major bottleneck in FL, particularly for complex models, large-scale applications, and frequent updates. The next research conducted in this thesis involved extending the previous study to include wireless networks (WNs). In WSNs, achieving energy-efficient transmission is a significant challenge due to their limited resources. This has motivated us to continue with a comprehensive overview and classification of the latest advancements in context-aware edge-based AI models, with a specific emphasis on sensor networks. The review has also investigated the associated challenges and motivations for adopting AI techniques, along with an evaluation of current areas of research that need further investigation. To optimize the aggregation of the FL model and alleviate communication expenses, the initial study addressing communication overhead is extended to include a FL-based cluster optimization approach. Furthermore, to reduce the detrimental effect caused by data heterogeneity among edge devices on FL, a new study of group-personalized FL models has been conducted. Finally, taking inspiration from the previously mentioned FL models, techniques for assessing clients' contribution by monitoring and evaluating their behavior during training are proposed. In comparison with the most existing contribution evaluation solutions, the proposed techniques do not require significant computational resources.The FL algorithms presented in this thesis are assessed on a range of real-world datasets. The extensive experiments demonstrated that the proposed FL techniques are effective and robust. These techniques improve communication efficiency, resource utilization, model convergence speed, and aggregation efficiency, and also reduce data heterogeneity when compared to other state-of-the-art methods.
|
|
| 5. |
- Alam, Mahbub Ul, 1988-
(författare)
-
Advancing Clinical Decision Support Using Machine Learning & the Internet of Medical Things : Enhancing COVID-19 & Early Sepsis Detection
- 2024
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- This thesis presents a critical examination of the positive impact of Machine Learning (ML) and the Internet of Medical Things (IoMT) for advancing the Clinical Decision Support System (CDSS) in the context of COVID-19 and early sepsis detection.It emphasizes the transition towards patient-centric healthcare systems, which necessitate personalized and participatory care—a transition that could be facilitated by these emerging fields. The thesis accentuates how IoMT could serve as a robust platform for data aggregation, analysis, and transmission, which could empower healthcare providers to deliver more effective care. The COVID-19 pandemic has particularly stressed the importance of such patient-centric systems for remote patient monitoring and disease management.The integration of ML-driven CDSSs with IoMT is viewed as an extremely important step in healthcare systems that could offer real-time decision-making support and enhance patient health outcomes. The thesis investigates ML's capability to analyze complex medical datasets, identify patterns and correlations, and adapt to changing conditions, thereby enhancing its predictive capabilities. It specifically focuses on the development of IoMT-based CDSSs for COVID-19 and early sepsis detection, using advanced ML methods and medical data.Key issues addressed cover data annotation scarcity, data sparsity, and data heterogeneity, along with the aspects of security, privacy, and accessibility. The thesis also intends to enhance the interpretability of ML prediction model-based CDSSs. Ethical considerations are prioritized to ensure adherence to the highest standards.The thesis demonstrates the potential and efficacy of combining ML with IoMT to enhance CDSSs by emphasizing the importance of model interpretability, system compatibility, and the integration of multimodal medical data for an effective CDSS.Overall, this thesis makes a significant contribution to the fields of ML and IoMT in healthcare, featuring their combined potential to enhance CDSSs, particularly in the areas of COVID-19 and early sepsis detection.The thesis hopes to enhance understanding among medical stakeholders and acknowledges the need for continuous development in this sector.
|
|
| 6. |
- Alizadeh, Morteza, 1987-
(författare)
-
Architectural Aspects of Identification in Decentralized Systems
- 2024
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- High-level systems need identification techniques, where higher security and scalability are considered requirements. Identification plays a significant role in systems where smart electronic devices increase in zero trust and open environments like decentralized systems. Also, decentralization has emerged as one of the most exciting domains in recent years, again after the first Internet was invented. Besides, decentralization in identification systems has gained popularity worldwide since cryptocurrencies became part of businesses. Distributed Ledger Technology (DLT) and Distributed Hash Tables (DHT) can be appropriate decentralized solutions that improve identification to be much more secure, scalable, and trustworthy.The decentralized nature of DLT and DHT ensures no single point of failure, making them highly resilient to attacks. Blockchain as a DLT solution can help devices communicate with each other securely and trustably by storing an immutable history of transactions, providing an additional layer of security to identification systems. DHT senable applications to keep files and information immutable in a decentralized manner. DHTs ensure that the data is replicated across multiple nodes, making it highly resilient to data loss. Moreover, mitigating high storage costs without memory limitations is the target of these technologies. In this context, a decentralized system paradigm that combines systems with DLT and DHTs can be highly beneficial.This thesis argues for such a paradigm, and the contributions include introducing the term decentralized networks and architectures and demonstrating the feasibility of using blockchain as a DLT solution in real-world scenarios. These scenarios can be applied to the Internet of Things (IoT) or other Peer to Peer networked systems. We explore different architectures in various systems and analyze the interaction in blockchain. This thesis contributes to developing decentralized identification systems that provide users’ trust in an open environment. It presents the challenges associated with decentralized identification, including registry and storage issues, and proposes solutions using DLT and DHT. The immutability of DLT and DHTs provides fast and secure solutions for decentralized identification systems. In particular, we show that a DHT-based architecture is feasible to maintain decentralization while avoiding memory constraints. However, there is still room for improvement in terms of performance. Our investigation shows that combining DHTs with blockchain in decentralized identifiers improves performance.By concealing blocks in the private blockchain, we show that query performance is better than other DHT and public blockchain-based solutions without concealed information. Moreover, our results show that DHT performs better than the public blockchain for scenarios with many records.These findings highlight the importance of selecting the appropriate technology for decentralized identification systems, considering the specific use case and the number of records to be stored.We also consider different decentralized identification systems and platforms built based on the recommendation of W3C Decentralized Identifiers (DIDs). We found low-efficiency issues using this technology, resulting from leveraging public DLT in the data registry part of DIDs. That model has searching time problems if the DLT grows. Finally, this thesis helps to analyze these issues and find better solutions. By choosing the right technology, we can ensure that decentralized identifiers are efficient, secure, and scalable, which enables users to trust them in an open environment.
|
|
| 7. |
- Andersson, Axel
(författare)
-
Computational Methods for Image-Based Spatial Transcriptomics
- 2024
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- Why does cancer develop, spread, grow, and lead to mortality? To answer these questions, one must study the fundamental building blocks of all living organisms — cells. Like a well-calibrated manufacturing unit, cells follow precise instructions by gene expression to initiate the synthesis of proteins, the workforces that drive all living biochemical processes.Recently, researchers have developed techniques for imaging the expression of hundreds of unique genes within tissue samples. This information is extremely valuable for understanding the cellular activities behind cancer-related diseases. These methods, collectively known as image-based spatial transcriptomics (IST) techniques, use fluorescence microscopy to combinatorically label mRNA species (corresponding to expressed genes) in tissue samples. Here, automatic image analysis is required to locate fluorescence signals and decode the combinatorial code. This process results in large quantities of points, marking the location of expressed genes. These new data formats pose several challenges regarding visualization and automated analysis.This thesis presents several computational methods and applications related to data generated from IST methods. Key contributions include: (i) A decoding method that jointly optimizes the detection and decoding of signals, particularly beneficial in scenarios with low signal-to-noise ratios or densely packed signals; (ii) a computational method for automatically delineating regions with similar gene compositions — efficient, interactive, and scalable for exploring patterns across different scales; (iii) a software enabling interactive visualization of millions of gene markers atop Terapixel-sized images (TissUUmaps); (iv) a tool utilizing signed-graph partitioning for the automatic identification of cells, independent of the complementary nuclear stain; (v) A fast and analytical expression for a score that quantifies co-localization between spatial points (such as located genes); (vi) a demonstration that gene expression markers can train deep-learning models to classify tissue morphology.In the final contribution (vii), an IST technique features in a clinical study to spatially map the molecular diversity within tumors from patients with colorectal liver metastases, specifically those exhibiting a desmoplastic growth pattern. The study unveils novel molecular patterns characterizing cellular diversity in the transitional region between healthy liver tissue and the tumor. While a direct answer to the initial questions remains elusive, this study sheds illuminating insights into the growth dynamics of colorectal cancer liver metastases, bringing us closer to understanding the journey from development to mortality in cancer.
|
|
| 8. |
- Aranda Muñoz, Álvaro
(författare)
-
Collaborative Thinking with and through Technology : Materials, methods and perspectives
- 2024
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- The rapid development and integration of IoT, AI, and digital twin technologies into work environments create new demands and challenges for organisations, including the need to upskill and prepare their workforces for new technological applications and capabilities. The collaborative design tradition draws upon participatory notions of joint inquiry to help people in the ideation and conception of alternative futures; however, given the novelty and the rapidity of the technological transformations, there is an opportunity to engage people unfamiliar with technology and design in the ideation and conceptualisation of work-oriented improvements using these technologies.This dissertation investigates what dimensions are necessary to support participatory processes for identifying and creating work-oriented improvements with technology and how design practitioners can apply these dimensions to stage these joint inquiry situations. To achieve these aims, explorations of and reflections on design are guided by a research-through-design approach that builds on three collaborative design cases that address real-world situations in a variety of contexts and participants: the joint inquiry processes of factory workers in the ideation, conceptualisation and prototyping of IoT work-oriented improvements; the joint inquiry process of ideation and conceptualisation of a digital twin in a manufacturing environment; and the joint inquiry processes of workers (predominantly from the healthcare sector) in ideating, conceptualising, and prototyping roles, skills, and products relying on IoT and AI technologies for their work futures. The research and design practice is guided by Deweyan pragmatism, underscoring the role and nature of materials (design methods, tools, and practices) in participatory design processes. Drawing from the three collaborative design cases and these theoretical notions, this thesis addresses two research questions: “What dimensions are needed to support participants in creating work-oriented improvements using technology?” and “How can these dimensions inform designers in staging joint inquiry situations of work-oriented improvements using technology?” The research methods consist of audio-recorded interviews, field notes, and collective reflective sessions to analyse the empirical material and video recordings. The main contributions are the identification of dimensions that underscore technology and work-oriented themes in joint inquiry and the framework of “thinking with and through technology”, which integrates these dimensions into a guided reflective and analytical design process. These contributions can help design and innovation practitioners and researchers prepare and stage materials, methods, and perspectives of joint inquiry situations concerning technology. The framework presents a “thinking with” perspective that underscores the material properties of technology and what the technology can offer to participants, and a “thinking through” perspective to contest the role of technology in organisations and open the design space to consider more sustainable and responsible futures. These results contribute to the collaborative design domain by developing knowledge and new nuances when staging joint inquiry situations of work-oriented improvements with technology. New understandings of these dimensions can contribute to an organisational landscape where workers can exercise their creativity, upskill their capacities, and voice their ideas and concerns concerning the technologies being integrated into their work environments.
|
|
| 9. |
- Bampa, Maria, 1992-
(författare)
-
Data-Driven AI for Patient and Public Health : On the Use of Multisource and Multimodal Data in Machine Learning to Improve Healthcare
- 2024
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- The integration of artificial intelligence in healthcare has created a new era of advancements, reshaping patient care and revolutionizing public health interventions. Through artificial intelligence, healthcare providers and public health authorities can optimize interventions, leading to more precise and efficient responses that enhance patient outcomes and address public health challenges effectively. The past decade has witnessed a rapid digital transformation across industries, and healthcare is no exception. This evolution is evident in the widespread adoption of electronic health records and healthcare information systems and the integration of diverse technologies, including handheld, wearable, and smart devices.A central challenge in this digital shift lies in representing data from multiple sources and modalities for downstream machine learning tasks. This complexity stems from the varied longitudinal or contextual events in patients' historical records, encompassing lab tests, vital signs, diagnoses, and drug administration. Additionally, the challenge extends to predictive modeling and constructing robust models that accurately classify future health events, taking into consideration heterogeneous health-related data. Electronic phenotyping, crucial for identifying fine-grained disease/patient clusters, is also a central problem when utilizing multisource and multimodal information effectively to create meaningful patient profiles. In the context of public health interventions, exemplified by crises like the COVID-19 pandemic, decision-making requires a delicate balance between optimizing intervention effectiveness and considering economic and societal well-being.This Ph.D. thesis seeks to unravel the potential of multisource and multimodal health observational data in generating patient phenotypes and predictions for both individual health and public health surveillance. It addresses the following central question: How can multisource and multimodal observational health data be effectively harnessed, using machine learning, to enhance patient and public health? Comprising five studies, the thesis confronts challenges posed by diverse data sources and modalities, exploring strategies for creating comprehensive patient profiles, developing robust classification models, and employing clustering methods tailored to observational health data. The research seeks to provide valuable insights into integrating AI in healthcare, with a specific emphasis on the complexities of multisource and multimodal data integration. It underscores the importance of exploring heterogeneous health observational data to deepen our understanding of patient health and optimize machine learning applications. Emphasizing the intricate nature of health data, the thesis discusses careful data handling and innovative methodologies to maximize its potential impact on improving patient outcomes and informing public health strategies. The effective management of heterogeneous observational health data requires thoughtful consideration due to their varied sources and inherent complexities.
|
|
| 10. |
- Banerjee, Sourasekhar, 1992-
(författare)
-
Advancing federated learning : algorithms and use-cases
- 2024
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- Federated Learning (FL) is a distributed machine learning paradigm that enables the training of models across numerous clients or organizations without requiring the transfer of local data. This method addresses concerns about data privacy and ownership by keeping raw data on the client itself and only sharing model updates with a central server. Despite its benefits, federated learning faces unique challenges, such as data heterogeneity, computation and communication overheads, and the need for personalized models. Thereby results in reduced model performance, lower efficiency, and longer training times.This thesis investigates these issues from theoretical, empirical, and practical application perspectives with four-fold contributions, such as federated feature selection, adaptive client selection, model personalization, and socio-cognitive applications. Firstly, we addressed the data heterogeneity problems for federated feature selection in horizontal FL by developing algorithms based on mutual information and multi-objective optimization. Secondly, we tackled system heterogeneity issues that involved variations in computation, storage, and communication capabilities among clients. We proposed a solution that ranks clients with multi-objective optimization for efficient, fair, and adaptive participation in model training. Thirdly, we addressed the issue of client drift caused by data heterogeneity in hierarchical federated learning with a personalized federated learning approach. Lastly, we focused on two key applications that benefit from the FL framework but suffer from data heterogeneity issues. The first application attempts to predict the level of autobiographic memory recall of events associated with the lifelog image by developing clustered personalized FL algorithms, which help in selecting effective lifelog image cues for cognitive interventions for the clients. The second application is the development of a personal image privacy advisor for each client. Along with data heterogeneity, the privacy advisor faces data scarcity issues. We developed a daisy chain-enabled clustered personalized FL algorithm, which predicts whether an image should be shared, kept private, or recommended for sharing by a third party.Our findings reveal that the proposed methods significantly outperformed the current state-of-the-art FL algorithms. Our methods deliver superior performance, earlier convergence, and training efficiency.
|
|
