{ "exported_at": "2026-11-15 22:55:29", "count": 64, "records": [ { "id": 76, "title": "Quantifying Uncertainty in Deep Learning-based medical image segmentation", "title_ar": "", "title_en": "Quantifying Uncertainty in Deep Learning-based medical image segmentation", "author": "Benjamin Lambert", "supervisor": "", "degree_type": "phd", "degree_evidence": "نص السجل: «Ph.D»", "university": "", "university_source": "", "college": "", "department": "", "country": "France", "governorate": "", "publication_year": 2023, "language": "en", "research_domain": "medical-ai", "research_subdomain": "medical-imaging", "categories": "machine-learning; deep-learning; computer-vision; medical-ai; ai-applications; predictive-analytics; image-segmentation; medical-imaging; ai-industry", "keywords": "", "abstract": "Quantifying the inherent uncertainty in an automated medical image analysis is crucial to guarantee the safe deployment of deep learning models. However, these models, often referred to as black boxes, are known to produce errors with high confidence, potentially leading to misinformed conclusions. The operative goal of Pixyl, Grenoble Institute of Neurosciences, and Inria for this Ph.D. is to develop flexible uncertainty quantification tools that would address the opacity of deep learning models. In medical image analysis, uncertainty estimates are useful at different levels: the voxel, the lesion, the subject (input image or output segmentation), and the estimated volumes. At the voxel level, there is a multitude of solutions proposed in the literature. Through three different brain lesion segmentation tasks, we show the adequacy of the Deep Ensemble framework to detect incorrectly classified voxels. The resulting uncertainty map can be overlayed on the input image to identify the ambiguous regions. Second, specifically tailored for diseases involving the detection of multiple lesions such as Multiple Sclerosis, we propose a lesion-level uncertainty module that associates each identified brain lesion with a confidence score. This allows to draw the clinician's attention to these uncertain lesion instances, which may correspond to false positive detections. Then, as deep learning models lack robustness when the test image is not represented in the training dataset, we build two different subject-level quality control tools. Non-conform inputs are detected using a compressed latent representation, allowing to efficiently compute its distance to the training distribution in a low-dimensional manifold. Yet, this approach focuses on the conformity of the input image and thus is unhelpful in detecting poor-quality segmentations. To alleviate this, a second quality check focusing on segmentation quality is implemented, allowing to enrich the informativeness of the case-level uncertainty quantification. Lastly, we leverage the conformal prediction framework to equip lesional volume estimations with robust predictive intervals. The proposed framework, TriadNet, computes the segmentation and associated predictive intervals in a second, thus being ideal for industrial software. The tools developed during this PhD will allow to enhance medical image analysis software with useful uncertainty estimates, allowing increased trust in the automated results and enabling informed decision-making.", "source_url": "https:\/\/openalex.org\/W4406884857", "pdf_url": "", "doi": "", "handle": "", "repository_name": "HAL", "source_record_id": "W4406884857", "other_sources": "", "verification_status": "verified", "created_at": "2026-11-15 16:52:46", "updated_at": "2026-11-15 16:52:46" }, { "id": 91, "title": "New and Advanced Explainable Artificial Intelligence Approaches for Managing Project Risks", "title_ar": "", "title_en": "New and Advanced Explainable Artificial Intelligence Approaches for Managing Project Risks", "author": "Bodrunnessa Badhon", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "", "university_source": "", "college": "", "department": "", "country": "Australia", "governorate": "", "publication_year": 2026, "language": "en", "research_domain": "generative-ai", "research_subdomain": "xai", "categories": "machine-learning; generative-ai; responsible-ai; expert-systems; ai-applications; predictive-analytics; xai; fairness; ai-smart-cities", "keywords": "", "abstract": "Project Risk Management (PRM) is a critical component of modern project governance, ensuring that uncertainties are systematically identified, assessed, and mitigated to minimize their impact on project outcomes. However, existing PRM practices remain largely subjective and qualitative, relying on expert judgment and static checklists that fail to capture dynamic interdependencies and evolving risk interactions. Recent advances in Artificial Intelligence (AI), particularly Machine Learning (ML) have improved predictive accuracy in PRM, but the opacity of many high-performing models has hindered adoption in safety-critical and managerial contexts where transparency, traceability, and stakeholder trust are essential. This thesis addresses these challenges by developing a suite of Explainable Artificial Intelligence (XAI) frameworks to support trustworthy project risk analysis. XAI has emerged as a promising approach to enhance transparency and interpretability in AI-based risk analysis; however, existing methods still exhibit several critical limitations that motivate this research. Current XAI techniques struggle to capture interrelations and cascading effects among risk factors, often producing fragmented explanations that overlook the structural dependencies underlying overall project vulnerability. In PRM, this limitation is particularly critical as risks are inherently interdependent, and disturbances in one aspect can propagate across multiple dimensions of project performance. Without incorporating such relationships into XAI models, the underlying causes of risk cannot be effectively identified, leading to explanations that lack depth and meaningful insight. Moreover, many post-hoc XAI models exhibit inconsistent or unstable explanations due to random neighborhood sampling or sensitivity to input variation, leading to unreliable outputs that are difficult to reproduce. In addition, existing XAI approaches remain largely static, offering one-way descriptions that overlook the temporal evolution and causal propagation of risks. Finally, they lack mechanisms for interactive engagement, preventing decision-makers from dynamically querying model behavior or exploring alternative “what-if” scenarios through conversational interfaces. Collectively, these limitations underscore the need for an integrated XAI-driven framework that unifies interpretability, scalability, causality, and interactivity. To address these shortcomings, this thesis defines four interrelated research objectives. Each objective focuses on a specific methodological limitation, namely interpretability, scalability, causality, and interactivity, while collectively forming a cohesive, progressive framework for explainable, trustworthy project risk management. The research commences by addressing the lack of interpretability and structured reasoning in existing risk models through the Interpretable Risk Assessment Framework based on Belief Rule-Based systems (IRAF-BRB). Traditional AI models for risk prediction often sacrifice interpretability for performance, making their outputs difficult for decision-makers to trust. IRAF-BRB resolves this issue by integrating expert knowledge with data-driven learning under explicit interpretability constraints. It ensures that model parameters remain meaningful, consistent with domain expertise, and explainable in linguistic terms. By optimizing belief rules using a modified Differential Evolution strategy while preserving human-interpretable reasoning, IRAF-BRB achieves both transparency and predictive reliability. This framework establishes the foundation for interpretable AI in PRM by showing that high accuracy can coexist with auditable, expert-aligned rule-based reasoning. While the IRAF-BRB ensures interpretability, it remains limited in scalability and adaptability to high-dimensional data. To overcome these limitations, the second framework, the Multi-Module Explainable Artificial Intelligence for Project Risk Management (MMXAI-PRM), extends explainability to complex, data-rich environments. MMXAI-PRM integrates generative modeling and post-hoc explanation to handle nonlinear relationships and class imbalance that traditional interpretable models cannot address. By combining synthetic data generation and local explanation, it enables consistent, stable interpretive reasoning even when training data are limited or heterogeneous. The framework provides richer explanatory insight by identifying key risk interactions and quantifying their relative influence, thereby enabling more informed prioritization and mitigation strategies. This contribution advances scalability and generalizability in explainable risk modeling, bridging the gap between interpretable logic-based systems and high-capacity learning models. Despite these improvements, earlier frameworks remain limited to static state-based representations that do not explicitly model the temporal evolution or causal propagation of risks.", "source_url": "https:\/\/openalex.org\/W7153988587", "pdf_url": "", "doi": "10.26190\/unsworks\/32194", "handle": "", "repository_name": "UNSWorks (UNSW Sydney)", "source_record_id": "W7153988587", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:46", "updated_at": "2026-11-15 16:52:46" }, { "id": 92, "title": "Explainable Artificial Intelligence in the Image Domain and its Applications to the Medical Field", "title_ar": "", "title_en": "Explainable Artificial Intelligence in the Image Domain and its Applications to the Medical Field", "author": "Mirtha Lucas", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "DePaul University", "university_source": "repository_host", "college": "", "department": "", "country": "United States", "governorate": "", "publication_year": 2026, "language": "en", "research_domain": "medical-ai", "research_subdomain": "medical-imaging", "categories": "machine-learning; deep-learning; computer-vision; generative-ai; medical-ai; responsible-ai; self-semi-supervised; predictive-analytics; cnn; gan; medical-imaging; xai; fairness", "keywords": "", "abstract": "This dissertation investigates the development of Explainable Artificial Intelligence (XAI) methods for deep learning models in the image domain, with a particular focus on medical imaging applications. Although neural networks achieve high predictive performance, their lack of interpretability limits their adoption in critical domains such as healthcare, where transparency and trust are essential. This work addresses this challenge by proposing novel approaches that improve the interpretability and reliability of model predictions. A primary contribution is the introduction of Riemann–Stieltjes Integrated Grad-CAM (RSI Grad-CAM), a gradient-based attribution method that generates more relevant and spatially localized saliency maps. The method is evaluated using quantitative metrics and statistical analysis, demonstrating improved performance over established techniques such as Grad-CAM and Integrated Grad-CAM. A secondary contribution is the analysis of the interpretability of intermediate layers in convolutional neural networks, demonstrating that clinically meaningful features, such as spiculation in lung nodules, can be identified within specific layers of the model. A third contribution is the development of a human-centered framework based on a Variational Autoencoder-Auxiliary Classifier Wasserstein Generative Adversarial Network (VAE-AC-WGAN), which generates contrastive explanations aligned with clinical reasoning. Experimental results demonstrate that the proposed method produces higher-quality reconstructions and more informative anomaly maps than baseline approaches (e.g., medXGAN), as reflected in improved structural similarity (SSIM) and classifier-based evaluation metrics (AUROC). Furthermore, human evaluation indicates that the generated explanations better capture clinically relevant features, such as spiculation, supporting their interpretability in medical imaging contexts. Overall, this work advances XAI by enhancing model transparency and supporting the trustworthy integration of AI systems into medical decision-making.", "source_url": "https:\/\/openalex.org\/W7172068913", "pdf_url": "", "doi": "", "handle": "", "repository_name": "Digital Commons-DePaul (DePaul University)", "source_record_id": "W7172068913", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:46", "updated_at": "2026-11-15 16:52:46" }, { "id": 94, "title": "A Novel Explainable Artificial Intelligence Framework with Improved Learning Mechanisms", "title_ar": "", "title_en": "A Novel Explainable Artificial Intelligence Framework with Improved Learning Mechanisms", "author": "Kabir, Sami", "supervisor": "", "degree_type": "phd", "degree_evidence": "نص السجل: «doctoral thesis»", "university": "Luleå University of Technology", "university_source": "repository_host", "college": "", "department": "", "country": "Sweden", "governorate": "", "publication_year": 2025, "language": "en", "research_domain": "medical-ai", "research_subdomain": "cnn", "categories": "machine-learning; deep-learning; computer-vision; medical-ai; responsible-ai; expert-systems; ai-applications; transfer-learning; self-semi-supervised; predictive-analytics; optimization; cnn; multimodal; xai; fairness; ai-smart-cities", "keywords": "", "abstract": "Artificial Intelligence (AI) models are increasingly being deployed in various critical domains, such as healthcare, finance, law, and autonomous systems, resulting in remarkable benefits for the society. Due to black-box (sub-symbolic) nature, these AI models do not provide explanation of the predictive output. This lack of explanation results in lack of transparency between human and machine. Such opacity of black-box AI models, particularly deep learning architectures, is a significant concern. To address this concern, Explainable Artificial Intelligence (XAI) has emerged as a vital research area. The aim of XAI is to enhance transparency, trust, and accountability of AI models. Various post-hoc XAI tools explain the outputs of black-box AI models based on training datasets. However, the use of training datasets, rather than domain knowledge, makes such explanations proxy. A biased training dataset may lead to a misleading post-hoc explanation as well. In contrast, an explanation based on domain knowledge will be more trustworthy to an end user. Motivated by this, we propose a novel XAI framework, consisting of Belief Rule Based Expert System (BRBES), to predict output and explain it with reference to domain knowledge. BRBES represents domain knowledge with its rule base. It outperforms other knowledge-driven AI models to handle uncertainty due to ignorance. To improve the accuracy of BRBES, we fine-tune its parameters and structure through evolutionary learning. Moreover, to overcome the scarcity of labeled dataset in this learning mechanism, we integrate semi-supervised and self-supervised learning with BRBES. We explain the output of BRBES with respect to the most influential rule of the rule base of BRBES. Thus, the output of our proposed XAI framework becomes not only accurate, but also explainable. This doctoral thesis delves into the challenges and opportunities surrounding XAI in order to provide a comprehensive understanding of the AI output. It presents a novel XAI framework to provide domain knowledge based energy consumption prediction with improved accuracy, and explain this predictive output in non-technical human language. Then it deals with multi-modal air quality data by integrating deep learning model with BRBES. Moreover, to reduce dependence on labeled data for evolutionary learning, this thesis integrates semi-supervised and self-supervised learning with BRBES. This thesis presents six significant contributions. First, we conduct a Systematic Literature Review (SLR) on XAI using the PRISMA guidelines, delving into the numerous challenges and opportunities of XAI. Extensive research is conducted to explore the definition, terminologies, taxonomy, and application domains of XAI. It highlights various challenges of XAI, such as, no universal definition, trade-off between accuracy and explainability, and lack of standardized evaluation metrics. To address this lack of standardized evaluation metrics, we also propose a unified framework to evaluate XAI. Secondly, we present an innovative explainable BRBES (eBRBES) framework, which offers accurate prediction of building energy consumption phenomenon, while providing insightful explanation and counterfactual based on domain knowledge. As part of eBRBES framework, we also present a novel Belief Rule Based adaptive Balance Determination (BRBaBD) algorithm to assess the optimal balance of the proposed framework between explainability and accuracy. Thirdly, we propose a mathematical model to integrate BRBES with the Convolutional Neural Network (CNN). We leverage the domain knowledge of BRBES, and image data patterns discovered by CNN with this integrated approach. We predict air quality with this integrated model using outdoor ground images and numerical sensor data. Fourth, we integrate two-layer BRBES with CNN to monitor air quality using satellite images, and relevant environmental parameters, such as, cloud, relative humidity, temperature, and wind speed. The two-layer BRBES showcases the strength of BRBES in conducting reasoning across multiple layers. Fifth, we enhance the learning mechanism of BRBES by utilizing the extensive unlabeled energy data, along with limited labeled data. For this purpose, we synthetically generate unlabeled data through weak and strong augmentation. We then pseudo-label these unlabeled data by integrating semi-supervised self-training model with BRBES. By learning from both labeled and pseudo-labeled data, initial BRBES transitions to semi-supervised BRBES. Sixth, to conduct the learning mechanism of BRBES without relying on any labeled data, we integrate self-supervised learning with it. To this effect, we pseudo-label the synthetically generated unlabeled energy data with BRBES in the pretext tasks of self-supervised learning. Then, the initial BRBES learns deeper representation from these pseudo-labeled data, and transitions to self-supervised BRBES.", "source_url": "https:\/\/openalex.org\/W7112031890", "pdf_url": "https:\/\/ltu.diva-portal.org\/smash\/get\/diva2:2007851\/FULLTEXT01", "doi": "", "handle": "", "repository_name": "DiVA (Luleå University of Technology)", "source_record_id": "W7112031890", "other_sources": "", "verification_status": "verified", "created_at": "2026-11-15 16:52:46", "updated_at": "2026-11-15 16:52:46" }, { "id": 96, "title": "Federated Learning of Explainable Artificial Intelligence models: real world applications and theoretical advances", "title_ar": "", "title_en": "Federated Learning of Explainable Artificial Intelligence models: real world applications and theoretical advances", "author": "F. Ruffini", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "University of Pisa", "university_source": "repository_host", "college": "", "department": "", "country": "Italy", "governorate": "", "publication_year": 2026, "language": "en", "research_domain": "responsible-ai", "research_subdomain": "federated-learning", "categories": "machine-learning; responsible-ai; predictive-analytics; xai; federated-learning; fairness", "keywords": "", "abstract": "This thesis aims to provide a personal contribution to the research area of Trustworthy Artificial Intelligence (AI), focusing on some aspects still largely unexplored. The objective is to develop systems that achieve high predictive accuracy while creating trust between stakeholders and the AI system. Since trustworthiness is a multidimensional topic, this work focuses on two of its key requirements: transparency and data privacy. In this context, the thesis explore the use of Fed-XAI models (explainable AI models trained through Federated Learning), to simultaneously ensure data privacy and provide explanations. At the time of writing, most existing works address these issues separately. Furthermore, this work provides a comparison between the explanations generated by post-hoc methods applied to opaque models and those derived from interpretable by design models. Such systematic comparison remains largely unexplored in the current literature, predominantly focusing on post-hoc methods alone. The thesis is structured as a three-act play. The first two acts examine the application of different Fed-XAI models to real-world datasets, initially focusing on interpretable-by-design approaches and subsequently comparing them with opaque models. These analyses highlight the limitations of existing strategies and motivate the methodological framework introduced in the third act, which represents the main theoretical contribution of this thesis.", "source_url": "https:\/\/openalex.org\/W7140268698", "pdf_url": "https:\/\/etd.adm.unipi.it\/theses\/available\/etd-01272026-184626\/unrestricted\/PhDAIThesis_RuffiniXVIII_pdfA.pdf", "doi": "", "handle": "", "repository_name": "Electronic Theses and Dissertations Repository (University of Pisa)", "source_record_id": "W7140268698", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:46", "updated_at": "2026-11-15 16:52:46" }, { "id": 98, "title": "Human-in-the-loop and Storytelling-based Explainable Frameworks for the Next Generation of Explainable Ensemble Methods and Artificial Intelligence Techniques", "title_ar": "", "title_en": "Human-in-the-loop and Storytelling-based Explainable Frameworks for the Next Generation of Explainable Ensemble Methods and Artificial Intelligence Techniques", "author": "Akshat Dubey", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "", "university_source": "", "college": "", "department": "", "country": "", "governorate": "", "publication_year": 2026, "language": "en", "research_domain": "nlp", "research_subdomain": "llm", "categories": "machine-learning; deep-learning; computer-vision; nlp; medical-ai; responsible-ai; predictive-analytics; llm; medical-imaging; xai; fairness", "keywords": "", "abstract": "Artificial intelligence (AI) is being adopted in diverse fields, including medical diagnosis and scientific research, to assist with decision-making thanks to its ability to learn underlying patterns from large amounts of data. However, these advances present a challenge. Modern AI systems, particularly deep learning (DL) models and large ensembles, are often difficult for audiences, such as domain experts, to understand due to their inherent black-box nature. This lack of transparency can reduce trust, impede adoption in critical fields, and cause problems with new regulations, such as the European Union’s (EU) AI Act, which impose strict transparency, accountability, and human oversight requirements. This thesis makes the following primary contributions: First, the thesis introduces a five-layer nested model for AI design and validation (Regulation, Domain, Data, Model, Prediction). This model systematically translates regulatory requirements into specific design and validation methodologies, surfacing threats to validity and compliance at each stage. Second, the thesis develops human-centered explainability frameworks, including Storytelling explainable AI (XAI) for medical imaging, which integrates multi-task distillation with visual and textual narratives. Third, the thesis introduces UbiQTree & UbiQVision, frameworks for uncertainty quantification in explainable AI (XAI). UbiQTree is instantiated for ensemble tree models, and UbiQVision is instantiated for deep vision models. They decompose shapley additive explanations (SHAP) based attributions into aleatoric, epistemic, and entangled components using Dirichlet processes and Dempster–Shafer theory (DST). This transformation shifts the focus from point-estimate explanations to distributions of explanations, allowing practitioners to calibrate explanations and facilitate meaningful human intervention. Finally, the thesis demonstrates domain-centric explainability through PepTriX, a framework for explainable peptide analysis. PepTriX integrates protein language models (PLM), predicted 3D structures, graph attention networks (GAT), and cross-modal co-attention to yield mechanistically interpretable representations. Together, these contributions reflect the thesis title, \"Human-in-the-Loop and Storytelling-Based Explainable Frameworks for the Next Generation of Explainable Ensemble Methods and Artificial Intelligence Techniques\". The work argues that explainability should be a core design principle throughout the AI lifecycle, not just an afterthought. Treating explanations as a means of communication and storytelling — i.e., narratives emphasizing the human reasoning process — brings AI systems closer to becoming not only good at achieving enhanced performance metrics, but also accountable, understandable, and better suited to users’ needs. This thesis is based on peer-reviewed publications and submissions covering AI regulations, human-centered explainability, uncertainty quantification, and domain-specific applications. The thesis allows for interpretive\/interpretative human-in-the-loop (HITL) interaction, enabling domain users to inspect and calibrate model behavior as necessary. Together, these works demonstrate that developing trustworthy AI in critical fields requires a comprehensive approach encompassing regulatory compliance, human involvement, technical methodologies, and domain-specific relevance.", "source_url": "https:\/\/openalex.org\/W7172247662", "pdf_url": "", "doi": "10.17169\/refubium-53170", "handle": "", "repository_name": "", "source_record_id": "W7172247662", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:46", "updated_at": "2026-11-15 16:52:46" }, { "id": 100, "title": "Understanding and the Black Box Problem in Artificial Intelligence", "title_ar": "", "title_en": "Understanding and the Black Box Problem in Artificial Intelligence", "author": "Barnaby Crook", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "University of Bayreuth", "university_source": "repository_host", "college": "", "department": "", "country": "Germany", "governorate": "", "publication_year": 2025, "language": "en", "research_domain": "medical-ai", "research_subdomain": "xai", "categories": "machine-learning; deep-learning; medical-ai; responsible-ai; ai-applications; xai; fairness; ai-industry", "keywords": "", "abstract": "Artificial intelligence (AI) systems developed with automated machine learning procedures are used throughout industry, science, and society. However, the opacity of these systems threatens to undermine the satisfaction of critical desiderata including safety, trust, and fairness. This worry is known as the black box problem (BBP). In this dissertation, I develop a novel conceptual framework for analysing the BBP and, through five interrelated papers, address three key research questions about its nature and our prospects for solving it. First, I consider the central epistemic construct of understanding, arguing that an abilities-based view is best-suited to the requirements of BBP analysis. Conceptualising understanding as abilities, I claim, appropriately balances theoretical justification with practical utility and can help coordinate interdisciplinary efforts in explainable AI (XAI). Second, I address the potential of XAI, a research field which attempts to render AI systems comprehensible to human stakeholders. I focus on mechanistic interpretability, an approach which employs functional decomposition to characterise the internal structure of AI systems. Mechanistic interpretability can deliver detailed explanations, but at significant costs in time and labour. Additionally, I present a framework for evaluating explainability approaches, using it to show how performance-explainability trade-offs can be effectively analysed. Third, I address the solubility of the BBP in particular contexts. In neuroscience, I argue against critics that artificial neural network models, in conjunction with XAI, can enhance our understanding of the brain. In psychiatry, I argue that even if the opacity of AI systems is overcome, data-driven approaches are unlikely to deliver transformative improvements to patient outcomes. Overall, this dissertation demonstrates how an abilities-based view of understanding and a considered assessment of XAI’s potential can be combined to yield new insights into the BBP and inform its resolution across diverse contexts.", "source_url": "https:\/\/openalex.org\/W7143534649", "pdf_url": "https:\/\/epub.uni-bayreuth.de\/id\/eprint\/8992\/1\/Crook_Barnaby_Understanding.pdf", "doi": "10.15495\/epub_ubt_00008992", "handle": "", "repository_name": "EPub Bayreuth (University of Bayreuth)", "source_record_id": "W7143534649", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:46", "updated_at": "2026-11-15 16:52:46" }, { "id": 101, "title": "Explainable artificial intelligence for integrity assessment of civil engineering infrastructure: Insights from various engineering applications", "title_ar": "", "title_en": "Explainable artificial intelligence for integrity assessment of civil engineering infrastructure: Insights from various engineering applications", "author": "Masoud Haghbin", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "University of Granada", "university_source": "repository_host", "college": "", "department": "", "country": "Spain", "governorate": "", "publication_year": 2025, "language": "en", "research_domain": "computer-vision", "research_subdomain": "xai", "categories": "machine-learning; computer-vision; responsible-ai; predictive-analytics; xai; fairness", "keywords": "", "abstract": "The incorporation of Artificial Intelligence (AI) into civil engineering, especially structural engineering, has the potential to improve decision-making by making it more precise and data driven. However, relying on AI to make decisions without understanding its reasoning poses a serious risk: It undermines the essential concept of human judgment. This concern has led to global regulatory measures, including the European Union AI Act, which highlights the importance of explainability and human oversight in creating trustworthy AI. In response to these needs, eXplainable AI (XAI) has attracted significant attention from researchers over the past five years. It serves as a technical tool to shed light on the internal computing processes of AI systems, which are often seen as black boxes, addressing safety and ethical implications. Most research into the application of XAI in civil engineering is conducted under controlled and standardized conditions. However, real-world engineering problems often occur in complex and changing environments. These problems often span multiple domains, making their modeling challenging despite their importance. In addition, the presence of scarce and poor quality data and real-world behaviors that are difficult to replicate in laboratory experiments (e.g., in the presence of damage) add to these complications. These inherent issues challenge the fundamental assumptions of many AI models, which are usually trained and validated in controlled or synthetic environments. Consequently, both the predictive accuracy and the reliability of model explanations can deteriorate when these models are extended to real-world situations, a risk that is often underestimated or overlooked. In 2026, the European Union’s AI Act will come into force for critical infrastructures, such as the vast majority of civil engineering infrastructure, making explainability and human oversight a legal requirement. There is a significant technical gap, as well as pressing safety, ethical, and regulatory concerns, due to the lack of assessments of the robustness of XAI models in real-world applications. This thesis shifts the focus from presenting novel AI models that excel in standardized benchmark problems to understanding how cutting-edge but existing AI and XAI methods perform under the irregular and data-poor conditions typically found in real-world civil engineering systems. This research highlights the need for realistic domain-aware deployment strategies that prioritize robustness, interpretability, and practical insights over purely theoretical advances. Rather than relying on idealized datasets or conditions favorable to AI, such as clean, balanced, and well-annotated data, which do not accurately represent realworld engineering conditions, this thesis focuses on investigating, understanding, and improving the behavior of the AI model in complex environments using real data from various case studies. To achieve these objectives, the thesis begins by investigating a complex type of rail surface degradation problem known as corrugation. This phenomenon is affected by a combination of dynamic and static factors, including vehicle-track interaction, material properties, and environmental conditions. However, measuring or integrating these contributing variables presents significant challenges as they are often difficult to assess directly, expensive, or even unfeasible to acquire in real-world situations. In order to ensure a realistic and application-oriented approach, a minimal set of input parameters (acceleration and velocity), which are readily available and can be easily measured from the moving vehicle, have been selected to predict the rail corrugation profile based on time series data of these input parameters. To align with practical constraints in the field, multiple damaged zones were considered in order to capture the variability and complexity of actual conditions of the rail infrastructure. Despite the common practice of removing outliers as an initial step in data preprocessing, the research presented in this thesis intentionally refrains from doing so in this particular scenario. Retaining outliers is crucial for accurately representing the unpredictable nature of real-world engineering data, where noise, anomalies, and irregular patterns are inherent in reality. Excluding them could lead to an overly simplified understanding of the problem and diminish the relevance of AI models in practical applications. The XAI method, Grad-CAM, is applied to enhance transparency and increase human oversight in the prediction of rail corrugation under the aforementioned realistic conditions. This approach allows for visual interpretation of the model decision-making process, helping to bridge the gap between black-box predictions and domain-relevant understanding.", "source_url": "https:\/\/openalex.org\/W7154665432", "pdf_url": "", "doi": "", "handle": "", "repository_name": "Institutional Repository of the University of Granada (University of Granada)", "source_record_id": "W7154665432", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:46", "updated_at": "2026-11-15 16:52:46" }, { "id": 102, "title": "Bridging the Gap: From Artificial Intelligence and Optimization Theory to Action", "title_ar": "", "title_en": "Bridging the Gap: From Artificial Intelligence and Optimization Theory to Action", "author": "Periklis Petridis", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "Massachusetts Institute of Technology", "university_source": "repository_host", "college": "", "department": "", "country": "United States", "governorate": "", "publication_year": 2025, "language": "en", "research_domain": "medical-ai", "research_subdomain": "xai", "categories": "machine-learning; generative-ai; medical-ai; responsible-ai; ai-applications; predictive-analytics; xai; ai-industry; ai-smart-cities", "keywords": "", "abstract": "Despite significant theoretical advances in Operations Research (OR) and Artificial Intelligence (AI), a persistent gap remains between these developments and their practical implementation in real-world settings. Despite significant progress in these fields, many OR and ML approaches struggle to scale to realistic problem sizes, lack robustness to uncertainty, or fail to address implementation constraints faced by practitioners in industry. Through four distinct works conducted in collaboration with industry partners, this research demonstrates how methodological advancements can bridge this theory-practice divide while maintaining rigorous theoretical foundations and guarantees. In the first part, we focus on optimization methodologies that scale traditional OR approaches to handle real-world problem sizes and uncertainty. In Chapter 2, we develop a stochastic Benders decomposition scheme for large-scale network design problems, a class of problems ubiquitous in logistics, transportation, and energy sectors. By incorporating sampling techniques within the decomposition framework, we achieve deterministic optimality guarantees while reducing computational costs, enabling solutions for networks with 700 nodes—an order of magnitude larger than previously tractable instances—while achieving optimality gaps of 5-7% compared to 16-27% for traditional deterministic approaches. In Chapter 3, we present a holistic framework for industrial decarbonization, developed with a major phosphate producer planning to quadruple energy consumption while transitioning to renewable sources. Our robust optimization approach combines strategic capacity expansion planning over a 25-year horizon with adaptive operational models, providing 95% reliability guarantees while balancing solar and wind integration with battery storage to meet a projected 12 TWh annual demand. In the second part, we shift our focus to developing AI systems that address the unique challenges of medical data abstraction and clinical decision support. In Chapter 4, we address the challenge of automating clinical data abstraction from electronic health records, collaborating with the Society of Thoracic Surgeons to populate their Adult Cardiac Surgery Database. Our AI pipeline combines 31 models per target variable with a two-tiered quality control system, achieving over 99% accuracy while automatically extracting 43-50% of registry variables, demonstrating how AI can dramatically reduce manual abstraction burden while maintaining clinical standards. In Chapter 5, we extend this healthcare AI focus by developing xHAIM (Explainable Holistic AI in Medicine), which addresses the limitations of current clinical AI systems in handling extensive patient records, providing interpretability, and incorporating medical knowledge. Through semantic similarity techniques and generative AI, xHAIM improves predictive performance while generating clinically grounded explanations that enhance trust and adoption by healthcare practitioners.", "source_url": "https:\/\/openalex.org\/W7123375454", "pdf_url": "", "doi": "", "handle": "", "repository_name": "DSpace@MIT (Massachusetts Institute of Technology)", "source_record_id": "W7123375454", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:46", "updated_at": "2026-11-15 16:52:46" }, { "id": 104, "title": "Applications of artificial intelligence for explainability and uncertainty quantification for performance-based design and damage classification in civil engineering", "title_ar": "", "title_en": "Applications of artificial intelligence for explainability and uncertainty quantification for performance-based design and damage classification in civil engineering", "author": "Shabbir, Khurram", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "Aalto University", "university_source": "repository_host", "college": "", "department": "", "country": "Finland", "governorate": "", "publication_year": 2025, "language": "en", "research_domain": "deep-learning", "research_subdomain": "transformers", "categories": "machine-learning; deep-learning; computer-vision; medical-ai; responsible-ai; ai-applications; predictive-analytics; transformers; xai; fairness; ai-smart-cities", "keywords": "", "abstract": "Structural resilience and safety are critical concerns in modern infrastructure, particularly in the face of increasing seismic risks and aging buildings. While advanced artificial intelligence (AI) techniques offer promising solutions for structural design, damage assessment, and retrofitting, the \"black-box\" nature of many AI models limits their trustworthiness and adoption in safety-critical fields. This research integrates explainable AI (XAI) techniques to bridge the gap between complex machine learning models and engineering practice, developing innovative frameworks for enhanced interpretability in civil engineering applications. A primary contribution is Seismo-XAI, a web-based tool that delivers transparent, interpretable insights into building performance under seismic loads, empowering engineers to optimize designs and retrofits. The study also introduces a novel hybrid framework combining microstructure-informed finite element modelling with computer vision to advance damage classification in concrete, synergizing vision transformers' computational efficiency with physics-based finite element analysis. To address data limitations, a synthetic dataset of concrete microstructures was generated using stochastic Monte Carlo augmentation. For uncertainty quantification in seismic assessment, this research employs artificial neural networks with the quality-driven lower upper bound estimation method to establish prediction intervals for long-term ground motion effects. This distribution-free machine learning approach provides both local and global levels probabilistic damage assessment, enhancing reliability in post-earthquake evaluations and early warning systems. Validation through benchmark structures and real-world case studies demonstrates the frameworks’ effectiveness in promoting reliable design practices for seismic resilience and damage assessment. The findings highlight the potential of XAI to transform structural engineering by providing transparent, interpretable, and actionable insights, fostering trust in AI-driven solutions. This research lays the foundation for future advancements in and development of AI applications for structural health monitoring and disaster management for urban infrastructure.", "source_url": "https:\/\/openalex.org\/W7115275889", "pdf_url": "https:\/\/aaltodoc.aalto.fi\/bitstreams\/22ee9f47-ed26-415f-8ef3-c22a3082d694\/download", "doi": "", "handle": "", "repository_name": "Aaltodoc (Aalto University)", "source_record_id": "W7115275889", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:46", "updated_at": "2026-11-15 16:52:46" }, { "id": 106, "title": "PERBANDINGAN KINERJA CNN DALAM KLASIFIKASI TUMOR OTAK DENGAN PENDEKATAN EXPLAINABLE ARTIFICIAL INTELLIGENCE (XAI)", "title_ar": "", "title_en": "PERBANDINGAN KINERJA CNN DALAM KLASIFIKASI TUMOR OTAK DENGAN PENDEKATAN EXPLAINABLE ARTIFICIAL INTELLIGENCE (XAI)", "author": "NIM.: 23206051017 Amin Darussalam", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "Sunan Kalijaga State Islamic University", "university_source": "repository_host", "college": "", "department": "", "country": "Indonesia", "governorate": "", "publication_year": 2026, "language": "en", "research_domain": "medical-ai", "research_subdomain": "medical-imaging", "categories": "machine-learning; deep-learning; computer-vision; medical-ai; responsible-ai; transfer-learning; predictive-analytics; cnn; image-classification; medical-imaging; xai", "keywords": "", "abstract": "Brain tumors are life-threatening conditions that require accurate and timely diagnosis. In clinical practice, diagnosis is commonly performed through Magnetic Resonance Imaging (MRI) analysis by medical experts, which is time-consuming and subject to variability. Recent advances in Convolutional Neural Networks (CNN) have shown promising performance in medical image classification; however, their black-box nature limits clinical trust and interpretability. This study proposes a comparative analysis of CNN-based models integrated with Explainable Artificial Intelligence (XAI) to improve both classification performance and model transparency. A transfer learning approach is employed using two architectures, ResNet50 and MobileNetV2, on a multi-class brain tumor MRI dataset. To enhance interpretability, Gradient-weighted Class Activation Mapping (Grad-CAM) is utilized to visualize discriminative regions influencing model predictions. Experimental results demonstrate that ResNet50 achieves an Accuracy of 93,7%, outperforming MobileNetV2 with 92.93%. Furthermore, Grad-CAM visualizations consistently highlight tumor-relevant regions, indicating that the models learn meaningful spatial features for classification. These findings confirm that integrating CNN with XAI not only provides high classification Accuracy but also improves model interpretability, thereby increasing its potential applicability in supporting reliable and transparent medical diagnosis.", "source_url": "https:\/\/openalex.org\/W7214133979", "pdf_url": "https:\/\/digilib.uin-suka.ac.id\/id\/eprint\/78375\/1\/23206051017_BAB-I_IV-atau-V_DAFTAR-PUSTAKA.pdf", "doi": "", "handle": "", "repository_name": "Digital Library UIN Sunan Kalijaga (Sunan Kalijaga State Islamic University)", "source_record_id": "W7214133979", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:46", "updated_at": "2026-11-15 16:52:46" }, { "id": 111, "title": "Explainable artificial intelligence (XAI) for healthcare decision-making", "title_ar": "", "title_en": "Explainable artificial intelligence (XAI) for healthcare decision-making", "author": "Zhiwei Zeng", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "Nanyang Technological University", "university_source": "repository_host", "college": "", "department": "", "country": "Singapore", "governorate": "", "publication_year": 2022, "language": "en", "research_domain": "medical-ai", "research_subdomain": "xai", "categories": "machine-learning; medical-ai; responsible-ai; ai-applications; self-semi-supervised; xai; fairness; ai-cybersecurity", "keywords": "", "abstract": "With the increasing ubiquity of Artificial Intelligence (AI) in daily life and into more critical applications, explainability has been brought under the spotlight due to ethical, social, and legal concerns. Explainability is particularly important in application domains involving high-stakes decision-making, such as healthcare, medicine, law, and defence, as simply having high performance may not satisfy all the desiderata of stakeholders. Stakeholders in these domains often require greater insights into the decision-making process, better support for tracing decision trails and identifying errors, as well as greater trustworthiness from an AI system. An eXplainable AI (XAI) is one that has the ability to explain the outputs or functioning of a model in a clear and understandable manner to a human. In the explanation process of XAI, explainability approaches provide explanatory information to human stakeholders in order to facilitate their understanding of the outputs and functioning of an AI model. The renewed understanding then affects the extent to which the desiderata are satisfied. Over the past decade, a plethora of explainability approaches have been proposed, ranging from ante-hoc methods that seek to build explainability features into a model itself to post-hoc methods that use techniques to improve the explainability of an already developed model. Despite the rapid growth of the XAI field, three key open problems emerge from a review of the existing XAI methods. Firstly, post-hoc methods have been increasingly used for high-stakes decisions, but most of them provide explanatory information that is not faithful or not insightful to what the original model computes. Secondly, most existing methods can only provide associational explanations and lack the structure and reasoning ability to provide contrastive and counterfactual explanations. Thirdly, most existing methods do not provide explanations to suit different levels of understanding. Research problem 1 - Post-hoc explainability methods have quickly gained popularity as they are mostly model-agnostic and highly extensible. However, without elucidating or understanding the model’s inner workings, explanations generated by post-hoc methods may not faithfully represent the underlying relationships of the model and often do not provide enough information to facilitate the understanding of how the output is obtained. In this thesis, I propose an ante-hoc explainability methodology to support the making and explaining of high-stakes decisions in healthcare. It is an argumentative methodology that can infuse explainability into the model and generate explanations that are faithful and insightful to the inner workings and internal representations of the model. It follows a “debate-like” process that shares many similarities with the way humans deliberate and make decisions, thereby giving it unique advantages in generating human-understandable explanations. I instantiate the argumentative methodology in three explainable decision-making approaches for the modelling and reasoning of contexts, qualitative preferences as well as both attack and support relationships. Research problem 2 - Although contrastive and counterfactual explanations have attracted more research interest recently, existing XAI methods mostly generate associational explana- tions. Most existing methods are good at answering “Why X? ” but poor at answering questions that require higher-level inference, such as “Why not-X? ”. This can be attributed to the fact that most of them lack the necessary structures and mechanisms to perform high-level logical reasoning. Building on the structures and reasoning ability of the three proposed explain- able decision-making approaches, I construct different forms of contrastive and counterfactual explanations that can answer questions such as “Why X instead of Y? ” and “Why not-X? ”. Research problem 3 - Different levels of understanding are often required in most practical scenarios. However, most existing XAI methods do not provide explanations that can cater to the different levels of understanding. In this thesis, I propose and study formal properties of explanatory information in order to generate such explanations. I propose notions and computational constructs for generating explanations that can provide different amounts of justification or have different dialectical strengths, as well as focused explanations that only include selective information in the decision-making process. As a case study, I implement a proposed explainable decision-making approach for the diagnostics and prognostics of Alzheimer’s Disease (AD). Empirical evaluation with real-world AD datasets and human evaluations were conducted to assess the performance and explainability of the approach.", "source_url": "https:\/\/openalex.org\/W4224077403", "pdf_url": "https:\/\/dr.ntu.edu.sg\/bitstream\/10356\/155849\/4\/Thesis_final_Zhiwei%20Zeng_signed%20with%20watermark.pdf", "doi": "10.32657\/10356\/155849", "handle": "", "repository_name": "DR-NTU (Nanyang Technological University)", "source_record_id": "W4224077403", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:47", "updated_at": "2026-11-15 16:52:47" }, { "id": 147, "title": "Class Imbalance and Time-To-Detection in the Performance Analysis of Machine Learning-Based Intrusion Detection Systems", "title_ar": "", "title_en": "Class Imbalance and Time-To-Detection in the Performance Analysis of Machine Learning-Based Intrusion Detection Systems", "author": "Shabanfar, Mohammad Pasha", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "Concordia University", "university_source": "repository_host", "college": "", "department": "", "country": "Canada", "governorate": "", "publication_year": 2025, "language": "en", "research_domain": "computer-vision", "research_subdomain": "object-detection", "categories": "machine-learning; computer-vision; responsible-ai; ai-applications; object-detection; fairness; ai-cybersecurity; ai-industry; ai-smart-cities", "keywords": "", "abstract": "The increasing reliance on Industrial Control Systems (ICS) and Supervisory Control and Data Acquisition (SCADA) systems has raised critical concerns regarding their vulnerability to cyberattacks. While machine learning (ML) methods have emerged as effective tools for detecting such intrusions, their real-world applicability is challenged by two major issues: the imbalance in cybersecurity datasets and the limited focus on the time required to detect attacks—referred to as Time-To-Detection (TTD). To address these two issues and suggest better practices for ML-based IDS researchers, this thesis examines the gaps in the literature and, through two respective case studies, aims to suggest practices toward a more precise and practical performance assessment of ML-based intrusion detection systems (IDS). First, the thesis examines the class imbalance problem prevalent in popular Information Technology (IT) and Operational Technology (OT) cybersecurity datasets, where normal traffic often vastly outnumbers attack instances. This imbalance leads to biased model performance and inflated accuracy scores, which can over- or under-assess a model’s ability to identify the minority classes correctly. Through a case study with several ML models on a realistic dataset, we demonstrate how imbalanced classes should be considered in the performance evaluation, and how imbalance learning techniques like resampling should be properly utilized for robust performance of ML-based IDS. Second, this thesis examines TTD, a crucial but often understressed performance indicator that measures how promptly an ML-based detection system identifies the onset of an attack. In addition to traditional metrics that focus solely on classification performance in ML communities, this thesis proposed a TTD model based on real-world responsiveness in OT systems by defining vari ous stages of the detection process for ML researchers to quantify and measure temporal overheads accordingly. We also demonstrate how the proposed TTD model can be applied to OT datasets through a case study, thereby suggesting it as a best practice for a more comprehensive evaluation of ML-based intrusion detectors in practical OT use cases. Through the two studies above, the thesis offers a more comprehensive and practical approach to evaluating ML-based IDS. It demonstrates how thoughtful consideration and integration of class imbalance and detection timeliness in the development and assessment of ML-based IDS representation is essential to deploy trustworthy and efficient cybersecurity solutions in critical infrastructure systems.", "source_url": "https:\/\/openalex.org\/W7111482866", "pdf_url": "", "doi": "", "handle": "", "repository_name": "Spectrum Research Repository (Concordia University)", "source_record_id": "W7111482866", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:47", "updated_at": "2026-11-15 16:52:47" }, { "id": 151, "title": "Investigation of Machine Learning Algorithms for Intrusion Detection System in Cybersecurity", "title_ar": "", "title_en": "Investigation of Machine Learning Algorithms for Intrusion Detection System in Cybersecurity", "author": "Mohmmed Alrowaily", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "University of South Florida", "university_source": "repository_host", "college": "", "department": "", "country": "United States", "governorate": "", "publication_year": 2020, "language": "en", "research_domain": "responsible-ai", "research_subdomain": "fairness", "categories": "machine-learning; responsible-ai; ai-applications; fairness; ai-cybersecurity", "keywords": "", "abstract": "The proliferation in usage and complexity of modern communication and network systems, a large number of trustworthy online Services and systems have been deployed. Even so, cybersecurity threats are still growing. An Intrusion Detection System (IDS) play a vital role in ensuring the security of communication networks, and it is taken into account as the subsequent security gate after the firewall. The IDS informs the system or network administrator in order to take specific actions to evade the suspicious activities. Three significant contributions are made during the course of this research to illustrate the feasibility of these IDS approaches. In the first contribution, we investigate the effectiveness of using conventional machine learning techniques based intrusion detection systems. The second contribution proposes an ensemble learning algorithm for cybersecurity threat detection. The third contribution proposes a hybrid feature selection approach for improving network attack detection. All presented algorithms were evaluated on the recent public CICIDS2017 dataset, which consist of benign and the most cutting-edge common attacks, and compared with other approaches. This research considers several machine learning classifiers, and feature selection techniques in order to study their classification performance under attack over different metrics. The empirical results of the three implemented systems conclude that the chosen minimized features provide promising performance to develop IDS that is effective and efficient for network intrusion detection. Moreover, these models not only improves the classification accuracy but also reduces the false alarm rate in the classification of IDS attacks.", "source_url": "https:\/\/openalex.org\/W3216292950", "pdf_url": "", "doi": "", "handle": "", "repository_name": "Digital Commons - University of South Florida (University of South Florida)", "source_record_id": "W3216292950", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:47", "updated_at": "2026-11-15 16:52:47" }, { "id": 159, "title": "Trustable machine learning for intrusion detection systems", "title_ar": "", "title_en": "", "author": "Robin Duraz", "supervisor": "", "degree_type": "phd", "degree_evidence": "theses.fr: الفهرس الوطني الفرنسي لأطاريح الدكتوراه", "university": "", "university_source": "", "college": "", "department": "", "country": "France", "governorate": "", "publication_year": 2024, "language": "fr", "research_domain": "computer-vision", "research_subdomain": "ai-cybersecurity", "categories": "machine-learning; computer-vision; ai-applications; ai-cybersecurity", "keywords": "", "abstract": "Apprentissage automatique de confiance pour les systèmes de détection d'intrusion Les systèmes de détection d'intrusion sont des composants essentiels à la défense de notre écosystème numérique. Récemment, les avancées en apprentissage automatique ont permis de développer de nouveaux types de système de détection d'intrusion, permettant de s'éloigner du besoin de créer des règles de détection de plus en plus complexes. Ces systèmes de détection utilisant l'apprentissage automatique sont capables d'apprendre de façon autonome différents comportements, à la condition d'avoir un jeu de données bien calibré. Le contexte de la cybersécurité amène des besoins spécifiques, des besoins qui sont différents des tâches d'apprentissage automatique les plus courantes : la reconnaissance d'images et le traitement du langage. Cela implique qu'il faut adapter les différents procédés utilisés en apprentissage automatique pour répondre à ces attentes. Étant dans un environnement ayant d'importants enjeux, les systèmes de détection d'intrusion devraient plutôt être utilisés pour aider à la prise de décisions, mais cela reste essentiel de pouvoir leur faire confiance. De ce fait, dans cette thèse, nous avons dans un premier temps développé une nouvelle métrique basée sur les scores CVSS, permettant d'intégrer des connaissances de cybersécurité dans le processus d'évaluation des systèmes de détection d'intrusion. Nous nous sommes ensuite concentrés sur le besoin de comprendre des décisions qui ne sont pas compréhensibles. Même si le domaine de l'explicabilité n'est pas assez avancé pour correctement expliquer ces décisions, cela reste possible de vérifier la confiance en celles-ci d'une façon plus robuste, amenant à examiner ou corriger d'éventuelles erreurs. Finalement, nous nous sommes efforcés de compléter les approches existantes en utilisant des techniques d'apprentissage automatique récentes pour augmenter la capacité à détecter de nouvelles cyberattaques. Toutes ces méthodes contribuent donc à créer des systèmes de détection d'intrusion utilisant l'apprentissage automatique qui sont plus fiables.", "source_url": "https:\/\/openalex.org\/W7152081341", "pdf_url": "https:\/\/theses.hal.science\/tel-04929212v1\/file\/2024IMTA0433_Duraz-Robin_Diffusion.pdf", "doi": "10.70675\/e2fe5f32z19e5z41caz9b55z7b33b874c3c0", "handle": "", "repository_name": "theses.fr (ABES)", "source_record_id": "W7152081341", "other_sources": "", "verification_status": "verified", "created_at": "2026-11-15 16:52:47", "updated_at": "2026-11-15 16:52:47" }, { "id": 188, "title": "Causal Inference Explanations for Graph Neural Networks", "title_ar": "", "title_en": "Causal Inference Explanations for Graph Neural Networks", "author": "Sahil Satish Kumar", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "", "university_source": "", "college": "", "department": "", "country": "Portugal", "governorate": "", "publication_year": 2025, "language": "en", "research_domain": "medical-ai", "research_subdomain": "gnn", "categories": "machine-learning; deep-learning; medical-ai; responsible-ai; predictive-analytics; gnn; xai; fairness", "keywords": "", "abstract": "The increasing popularity of Artificial Intelligence, notably Deep Learning, has prompted its application across diverse real-world scenarios. Graph Neural Networks, a type of Deep Learning model developed for graph-structured data, have gained significant popu- larity owing to their versatility and efficacy. Despite their high accuracy and performance, contemporary Deep Learning models often function as black box systems, lacking trans- parency and trustworthiness crucial for deployment in critical decision-making domains. In response, Explainable Artificial Intelligence has emerged, aiming to enhance the trustworthiness of black box models by devising explanation methods that clarify their inner workings. However, prevalent explanation techniques predominantly leverage correlation and association rather than employing causality, a significant aspect of human comprehension. In this thesis, we propose a novel explanation method using causal inference developed specifically for Graph Neural Networks. Our approach seeks to illuminate the decision- making process of Graph Neural Networks, thereby augmenting their transparency and trustworthiness. To show the generality of our approach we do a benchmark evaluation of our method against state-of-the-art explanation methods on a variety of datasets, tasks and Graph Neural Networks architectures. Furthermore, we apply and evaluate our explanation method in two real-world use cases: medical referral recommendation and space object conjunction prediction. In both cases, we demonstrate the applicability of our method and discuss its limitations.", "source_url": "https:\/\/openalex.org\/W7197425670", "pdf_url": "https:\/\/run.unl.pt\/bitstreams\/811956c8-4867-4ac0-a47e-557471695537\/download", "doi": "", "handle": "", "repository_name": "Portuguese National Funding Agency for Science, Research and Technology (RCAAP Project by FCT)", "source_record_id": "W7197425670", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:47", "updated_at": "2026-11-15 16:52:47" }, { "id": 209, "title": "Trustful AI through fair and privacy-preserving federated machine learning", "title_ar": "", "title_en": "", "author": "Nawel Benarba", "supervisor": "", "degree_type": "phd", "degree_evidence": "theses.fr: الفهرس الوطني الفرنسي لأطاريح الدكتوراه", "university": "", "university_source": "", "college": "", "department": "", "country": "France", "governorate": "", "publication_year": 2026, "language": "fr", "research_domain": "responsible-ai", "research_subdomain": "fairness", "categories": "machine-learning; responsible-ai; fairness", "keywords": "", "abstract": "Intelligence artificielle digne de confiance grâce à l'apprentissage fédéré équitable et respectueux de la vie privée L'augmentation massive du volume de données disponibles a transformé l'informatique moderne, contribuant à l'essor de l'apprentissage automatique dans plusieurs domaines tels que la santé, la finance et les systèmes autonomes. Cette dépendance croissante aux modèles d'apprentissage pour la prise de décision dans ces domaines soulève des préoccupations majeures en matière d'équité et de confidentialité, étant donné que les décisions qu'ils produisent peuvent influencer de manière significative les individus et les organisations dans des contextes d'application réels. Motivés par la nécessité de mieux comprendre et d'atténuer ces enjeux, la présente recherche étudie les défis associés à ces problématiques et propose des contributions notables en faveur de systèmes d'apprentissage plus responsables et plus fiables. La première partie de cette thèse porte sur l'équité, devenue une préoccupation centrale en apprentissage automatique en raison de son potentiel à amplifier les disparités et à renforcer les biais existants. Ce constat, associé à l'émergence de régulations telles que l'AI Act de l'Union européenne et l'Executive Order américain, souligne la nécessité d'intégrer l'équité comme principe fondamental dans la conception des systèmes d'apprentissage automatique. Compte tenu de cela, nous avons examiné comment les méthodes de sélection automatique de données, souvent utilisées pour améliorer l'efficacité en termes de temps en utilisant des sous-ensembles plus petits d'entraînement plutôt que le jeu de données complet, influencent l'équité et la qualité des modèles. Nous avons mené la première analyse à grande échelle révélant l'impact souvent ignoré de la sélection sur ces deux aspects, nous avons identifié les facteurs clés qui en sont à l'origine et nous avons avancé des pistes pertinentes pour favoriser un apprentissage plus équitable et efficace. Parallèlement aux régulations promouvant un apprentissage plus équitable, d'autres insistent sur la protection des données utilisées, tel que le règlement général sur la protection des données (RGPD), qui a favorisé l'adoption de l'apprentissage fédéré, permettant à plusieurs parties d'entraîner un modèle global sans partager leurs données. Cependant, la nature distribuée et collaborative de ce paradigme d'apprentissage amplifie les biais hérités de l'apprentissage centralisé et introduit également de nouveaux types de biais. Pour éclairer ces enjeux, nous nous avons proposé une revue approfondie de l'équité en apprentissage fédéré, identifiant les principales sources de biais, couvrant l'ensemble des stratégies de mitigation existantes, et mettant en évidence des pistes de recherche encore ouvertes pour faire progresser l'équité dans ce cadre. Bien que l'apprentissage fédéré permette à plusieurs parties d'entraîner un modèle sans partager leurs données locales, de nombreuses études montrent qu'il reste vulnérable à diverses menaces portant atteinte à la vie privée des utilisateurs. Parmi ces menaces figurent les attaques d'inférence d'appartenance, qui permettent à un adversaire de déterminer si un point de données faisait partie de l'ensemble d'entraînement, exposant potentiellement des informations confidentielles. Dans ce contexte, la deuxième partie de cette thèse se concentre sur les atteintes à la vie privée en apprentissage fédéré, en accordant une attention particulière aux attaques d'inférence d'appartenance. Pour contrer ces attaques, nous avons introduit GAPER, une méthode granulaire et adaptative fondée sur la confidentialité différentielle, offrant une protection efficace contre ce type d'attaque tout en préservant la qualité du modèle, et soutenue à la fois par des garanties théoriques et des évaluations empiriques. De manière plus générale, ces travaux font progresser le développement de systèmes d'apprentissage non seulement performants, mais également fondamentalement alignés avec les principes d'équité et de confidentialité.", "source_url": "https:\/\/openalex.org\/W7170059833", "pdf_url": "https:\/\/theses.fr\/2026ISAL0013\/document", "doi": "10.70675\/17b0463eza470z4230za6edze526eaf22206", "handle": "", "repository_name": "theses.fr (ABES)", "source_record_id": "W7170059833", "other_sources": "", "verification_status": "verified", "created_at": "2026-11-15 16:52:47", "updated_at": "2026-11-15 16:52:47" }, { "id": 213, "title": "Trustworthy AI Under Distribution Shifts: Enhancing Generalization and Fairness in Deep Learning", "title_ar": "", "title_en": "Trustworthy AI Under Distribution Shifts: Enhancing Generalization and Fairness in Deep Learning", "author": "Gezheng Xu", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "Western University", "university_source": "repository_host", "college": "", "department": "", "country": "Canada", "governorate": "", "publication_year": 2026, "language": "en", "research_domain": "medical-ai", "research_subdomain": "fairness", "categories": "machine-learning; deep-learning; computer-vision; generative-ai; medical-ai; responsible-ai; transfer-learning; predictive-analytics; optimization; fairness", "keywords": "", "abstract": "Recent advances in machine learning and deep learning have achieved remarkable success across a wide range of applications, yet they largely rely on data-driven paradigms that assume the training and test data share similar distributions. In real-world scenarios, however, this assumption often fails. Distribution shifts between training and deployment environments or among population subgroups can significantly compromise the performance and reliability of AI systems. Moreover, privacy constraints frequently prevent direct access to source data, and ethical concerns surrounding fairness raise additional challenges for the trustworthy use of AI in sensitive domains such as healthcare and finance. These factors together highlight the urgent need for a deeper and more principled understanding of trustworthy AI under distribution shifts. This thesis examines the influence of distribution shifts along two complementary dimensions of trustworthy AI: generalizability and fairness. On one hand, we focus on the noisy and unreliable signals caused by distribution shift, which place learning in a weak-supervision regime and thereby limit AI systems’ generalizability. On the other hand, we focus on the distribution shifts that arise from structural data imbalance or scarcity rooted in historical and societal factors, which give rise to socially unfavorable model behaviors and, in turn, exacerbate fairness issues in the decision-making process. From the generalizability perspective, the thesis studies the domain adaptation problem under data privacy constraints, formulated as the Source-Free Domain Adaptation (SFDA) problem. SFDA is examined in both classification and regression tasks, with a focus on pseudo-label noise mitigation and uncertainty modeling, respectively. For the classification problem, a noise-robust loss–based noise-and-variance control method (NVC-LLN) is proposed to mitigate the unbounded label noise inherent in SFDA. For the regression task, a histogram-based uncertainty reconstruction framework, MERCI, is proposed to refine continuous supervision signals and align feature presentations under distribution shift. For the fairness problem, this thesis focuses on the multiple-subgroup situation, ranging from a single sensitive attribute with multiple subgroups to multiple sensitive attributes with intersectional bias, covering both unfairness discovery and mitigation. For subgroup discovery, a Bias-Guided Generative Network (BGGN) is proposed to identify underrepresented subgroups instead of relying on traditional search methods. For mitigation, a generalization error bound is derived under the PAC-Bayes framework, motivating a probabilistic-predictor-based bilevel optimization algorithm, FAMS, which achieves both fairness and accuracy across multiple subgroups under limited-data conditions. Collectively, these studies advance the development of responsible, adaptive, and human-centric AI systems. These systems are capable of performing reliably across diverse and dynamic environments.", "source_url": "https:\/\/openalex.org\/W7205536362", "pdf_url": "", "doi": "", "handle": "", "repository_name": "Scholarship@Western (Western University)", "source_record_id": "W7205536362", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:47", "updated_at": "2026-11-15 16:52:47" }, { "id": 215, "title": "Bias and Fairness in AI Decision Making for Hiring Processes", "title_ar": "", "title_en": "Bias and Fairness in AI Decision Making for Hiring Processes", "author": "Muhammad Edwin Dwi Putra", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "KTH Royal Institute of Technology", "university_source": "repository_host", "college": "", "department": "", "country": "Sweden", "governorate": "", "publication_year": 2025, "language": "en", "research_domain": "responsible-ai", "research_subdomain": "xai", "categories": "machine-learning; responsible-ai; predictive-analytics; xai; fairness", "keywords": "", "abstract": "Artificial intelligence algorithms have been found to perpetuate societal biases that have been embedded in the system over decades due to historical discrimination. In the recruitment sector, this often manifested in biased hiring decisions, where the AI-driven hiring tools inherited patterns that led to unfair outcomes. The purpose of this research was to detect and mitigate such bias by providing transparent explanations for an AI-driven decision, which supports fairness in recruitment processes. A machine learning model was developed to assess candidate suitability based on multiple recruitment-related features. The SHAP (SHapley Additive exPlanations) framework was applied to interpret the model's predictions and to identify the influence of individual features, with particular attention to gender as a sensitive attribute. The purpose of using the XAI (explainable AI) technique was to identify and examine potential bias in order to mitigate it. Examination and analysis of sensitive variables behind an outcome, such as gender, revealed a measurable influence on model predictions. These findings helped identify potential biases in the recruitment model and informed modifications aimed to reduce their impact. Incorporating XAI-based transparency into recruitment algorithms provided a systematic and actionable approach to bias detection. This approach enhanced the fairness of the AI recruitment tools and fostered greater trust in their adoption.", "source_url": "https:\/\/openalex.org\/W7116823264", "pdf_url": "https:\/\/ltu.diva-portal.org\/smash\/get\/diva2:2007317\/FULLTEXT01", "doi": "", "handle": "", "repository_name": "KTH Publication Database DiVA (KTH Royal Institute of Technology)", "source_record_id": "W7116823264", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:47", "updated_at": "2026-11-15 16:52:47" }, { "id": 218, "title": "Explainable AI: Enhancing Machine Learning Model Interpretability with Generative AI", "title_ar": "", "title_en": "Explainable AI: Enhancing Machine Learning Model Interpretability with Generative AI", "author": "Castelhano, Inês Dinis", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "", "university_source": "", "college": "", "department": "", "country": "Portugal", "governorate": "", "publication_year": 2025, "language": "en", "research_domain": "nlp", "research_subdomain": "llm", "categories": "machine-learning; nlp; generative-ai; medical-ai; responsible-ai; ai-applications; llm; xai; fairness; ai-cybersecurity", "keywords": "", "abstract": "In today’s data-driven world, machine learning (ML) powers critical decisions across sectors such as healthcare and finance, but the opacity of complex “black-box” models undermines trust, accountability, and adoption. This thesis tackles the urgent challenge of explainability by exploring how Large Language Models (LLMs) can transform post-hoc explanations into clear, actionable insights for diverse stakeholders. Moving beyond traditional techniques like SHAP and Counterfactuals, which often overwhelm with complexity, this research introduces a dynamic framework that integrates LLMs as narrative explainers. The methodology combines robust ML pipelines with post-hoc interpreters, enhanced through prompt engineering, to generate explanations in both technical and business-friendly formats. Experiments on real-world datasets, including emergency healthcare and bank fraud detection, benchmarked leading LLMs such as GPT-4o, Claude 3, LLaMA 3, and DeepSeek. Results show that GPT-4o consistently delivers the most accurate, fluent, and stakeholderaligned explanations, while local open-weight models offer competitive, privacy-preserving alternatives. The evaluation, comprising linguistic heuristics, semantic similarity metrics, and human judgment, demonstrated significant gains in clarity, completeness, and trustworthiness over conventional explainers. Crucially, counterfactual-based narratives proved highly intuitive for decision-making, while SHAP-based explanations achieved greater technical depth. By reframing LLMs as interpretable mediators rather than mere translators, this study provides empirical evidence that generative AI can close the gap between ML performance and human understanding. The contributions extend beyond academic insight, offering practical guidelines for deploying Explainable AI in high-stakes domains where transparency is not optional but essential for fairness, accountability, and trust.", "source_url": "https:\/\/openalex.org\/W7112581040", "pdf_url": "https:\/\/run.unl.pt\/bitstreams\/3e76fd1f-bd42-49bf-922d-b79fcffa9a17\/download", "doi": "", "handle": "", "repository_name": "Portuguese National Funding Agency for Science, Research and Technology (RCAAP Project by FCT)", "source_record_id": "W7112581040", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:47", "updated_at": "2026-11-15 16:52:47" }, { "id": 220, "title": "RESPONSIBLE RISK ASSESSMENT OF MACHINE LEARNING TOOLS FOR PROMOTION BOARDS", "title_ar": "", "title_en": "RESPONSIBLE RISK ASSESSMENT OF MACHINE LEARNING TOOLS FOR PROMOTION BOARDS", "author": "Michael R. Burke", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "Naval Postgraduate School", "university_source": "repository_host", "college": "", "department": "", "country": "United States", "governorate": "", "publication_year": 2025, "language": "en", "research_domain": "responsible-ai", "research_subdomain": "fairness", "categories": "machine-learning; robotics; responsible-ai; predictive-analytics; fairness", "keywords": "", "abstract": "This thesis evaluates an artificial intelligence (AI) model's ability to support U.S. Marine Corps promotion boards, investigating how data drift and feedback loops impact its performance and reliability over time. Using the DataRobot automated machine learning (AutoML) platform, a model was trained on fiscal year (FY) 2023 Major promotion board data and subsequently tested against FY24 and FY25 data to measure performance degradation and detect statistical drift in key predictors. The model demonstrated strong but brittle performance, maintaining high ranking accuracy across years while relying heavily on administrative features like promotion zone, which exhibited significant data drift. The research concludes that without robust governance, the model's dependence on procedural proxies risks amplifying biases and capturing administrative timing rather than a holistic view of merit, potentially undermining the promotion system's integrity. It is recommended that the Marine Corps use the AI tool for pre-screening to improve efficiency, while implementing annual retraining, continuous drift monitoring, and explicit fairness thresholds to ensure long-term effectiveness and trust.", "source_url": "https:\/\/openalex.org\/W7134863984", "pdf_url": "", "doi": "", "handle": "", "repository_name": "Calhoun: The Naval Postgraduate School Institutional Archive (Naval Postgraduate School)", "source_record_id": "W7134863984", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:47", "updated_at": "2026-11-15 16:52:47" }, { "id": 222, "title": "Towards Trustworthy Machine Learning: Fairness Under Dynamics and Counterfactuals", "title_ar": "", "title_en": "Towards Trustworthy Machine Learning: Fairness Under Dynamics and Counterfactuals", "author": "Ozgur Guldogan", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "", "university_source": "", "college": "", "department": "", "country": "United States", "governorate": "", "publication_year": 2026, "language": "en", "research_domain": "nlp", "research_subdomain": "llm", "categories": "machine-learning; nlp; generative-ai; responsible-ai; few-zero-shot; predictive-analytics; llm; fairness", "keywords": "", "abstract": "Machine learning systems increasingly make decisions with long-term consequences for individuals and society. Designing trustworthy machine learning systems therefore requires going beyond static accuracy and one-shot fairness criteria. Decisions may alter future populations through feedback, individuals may improve and reapply, uncertainty may call for set-valued predictions rather than point estimates, and large-scale deployment may be constrained by the efficiency of modern generative models. This thesis explores these challenges through contributions spanning fairness under dynamics, counterfactual fairness under predictive uncertainty, and efficient inference serving for large language models. We first study fairness in dynamic decision-making settings such as lending, hiring, and admissions, where rejected individuals may invest effort to improve and reapply. We introduce Equal Improvability, a fairness notion requiring that rejected individuals from different groups have comparable opportunities to become accepted after bounded effort. We develop optimization-based methods for learning classifiers that satisfy this criterion and show that they can promote more equitable long-term population dynamics. Next, we consider sequential selection by multiple decision-makers from a shared applicant pool. We propose fair-greedy policies that balance utility with long-term representation targets, characterize convergence under positive reinforcement dynamics, and show that policy design must account for more complex feedback mechanisms to avoid unintended negative reinforcement. We then address fairness under counterfactual reasoning and predictive uncertainty. Counterfactually Fair Conformal Prediction constructs prediction sets that are both statistically valid and counterfactually fair. By aggregating conformity scores across counterfactual interventions on protected attributes, the method produces prediction sets that remain invariant under counterfactual changes to the protected attribute while preserving finite-sample marginal coverage guarantees. Finally, we study computational efficiency in generative machine learning systems. Multi-Bin Batching is an inference-serving policy for large language models that groups requests by predicted generation lengths to reduce inefficiency from heterogeneous runtimes. Queueing-theoretic analysis characterizes the throughput benefits of length-aware batching and shows that the effective service rate approaches the idealized static-batching capacity as the number of bins increases. Experiments on simulated and real inference workloads demonstrate substantial throughput improvements over static batching. Together, these contributions provide fairness notions and algorithms that remain meaningful under feedback, uncertainty, and counterfactual interventions, while also developing principled serving mechanisms that support scalable deployment of generative models.", "source_url": "https:\/\/openalex.org\/W7213476610", "pdf_url": "https:\/\/escholarship.org\/content\/qt0r1882jr\/qt0r1882jr.pdf", "doi": "", "handle": "", "repository_name": "eScholarship (California Digital Library)", "source_record_id": "W7213476610", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:47", "updated_at": "2026-11-15 16:52:47" }, { "id": 260, "title": "Artificial Intelligence for Medical Diagnosis Using Machine Learning for Brain Tumor MRI Classification : магистерская диссертация", "title_ar": "", "title_en": "Artificial Intelligence for Medical Diagnosis Using Machine Learning for Brain Tumor MRI Classification : магистерская диссертация", "author": "М. Ш. С. Мако ; M. S. S. Mako", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "Ural Federal University", "university_source": "repository_host", "college": "", "department": "", "country": "Russia", "governorate": "", "publication_year": 2025, "language": "en", "research_domain": "medical-ai", "research_subdomain": "medical-imaging", "categories": "machine-learning; deep-learning; computer-vision; medical-ai; responsible-ai; transfer-learning; cnn; medical-imaging; xai", "keywords": "", "abstract": "The integration of Artificial Intelligence (AI) into the domain of medical diagnostics has ushered in a new era of precision, speed, and reliability, particularly in the analysis of medical imaging. Among the most promising applications is the use of Machine Learning (ML) techniques for the classification and diagnosis of brain tumors from Magnetic Resonance Imaging (MRI) scans. Brain tumors, whether malignant or benign, pose significant health threats and require early, accurate detection to improve prognosis and treatment outcomes. Traditional methods for diagnosing brain tumors are often time-consuming, subject to inter-observer variability, and reliant on the expertise of radiologists. This thesis explores the use of AI-driven machine learning models to automate and enhance the diagnostic process by classifying brain tumor types from MRI images. This research presents a comprehensive study on the implementation of various machine learning algorithms—including Convolutional Neural Networks (CNNs), Support Vector Machines (SVM), and ensemble models—for the classification of different types of brain tumors, such as gliomas, meningiomas, and pituitary tumors. The dataset employed comprises a substantial number of pre-processed and labeled MRI images sourced from publicly available medical imaging repositories. The study begins with a detailed examination of MRI imaging modalities, the clinical characteristics of common brain tumors, and the importance of accurate classification. A robust preprocessing pipeline was designed to handle data augmentation, normalization, and artifact removal, which are critical for enhancing image quality and improving model generalization. Feature extraction techniques, both handcrafted and automated via deep learning, were analyzed to assess their effectiveness in capturing the unique spatial and textural patterns of different tumor types. Subsequently, multiple supervised learning algorithms were trained and tested using cross-validation to compare their accuracy, precision, recall, F1-score, and area under the receiver operating characteristic (ROC) curve. CNN-based architectures, particularly those leveraging transfer learning (e.g., ResNet, VGGNet, and EfficientNet), demonstrated superior performance compared to traditional classifiers, achieving classification accuracies exceeding 90% in test scenarios. In addition, model interpretability techniques such as Grad-CAM and SHAP were applied to visualize and explain the decision-making process of the neural networks, addressing one of the key challenges in deploying AI systems in clinical environments: trust and transparency.", "source_url": "https:\/\/openalex.org\/W7117637577", "pdf_url": "https:\/\/elar.urfu.ru\/bitstream\/10995\/147597\/1\/m_th_m.s.s.mako_2025.pdf", "doi": "", "handle": "", "repository_name": "Electronic scientific archive of UrFU (Ural Federal University)", "source_record_id": "W7117637577", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:48", "updated_at": "2026-11-15 16:52:48" }, { "id": 355, "title": "Predicting refractive index structure parameter in the marine wave boundary layer using machine learning models: a thesis in Data Science", "title_ar": "", "title_en": "Predicting refractive index structure parameter in the marine wave boundary layer using machine learning models: a thesis in Data Science", "author": "Bhavana Chikkamuduvadi Renuka Gowda", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "", "university_source": "", "college": "", "department": "", "country": "", "governorate": "", "publication_year": 2026, "language": "en", "research_domain": "medical-ai", "research_subdomain": "predictive-analytics", "categories": "machine-learning; medical-ai; ai-applications; predictive-analytics; ai-education", "keywords": "", "abstract": "The refractive index structure parameter, ??ₙ², quantifies optical turbulence and strongly influences the performance of electromagnetic and free-space optical systems operating within the marine wave boundary layer. Predicting ??ₙ² remains challenging due to nonlinear interactions among atmospheric variables and shows strong changes between daytime and nighttime, especially in coastal environments. This thesis examines these challenges using high-frequency meteorological data from ATI sonic anemometers and co-located scintillometer observations collected at the SMAST Pier in Buzzards Bay, Massachusetts, during the summer and fall of 2025. A random forest (RF) modeling framework is developed to evaluate not only predictive accuracy, but also the conditions under which machine-learning-based estimates of ??ₙ² agree or disagree with observations. Atmospheric data are strongly time-dependent, meaning that common random train-test splits can introduce information leakage and overestimate model skill. To avoid this problem, the analysis uses a blocked, leakage-free train-gap-test strategy based on decorrelation time scales. Low-frequency and seasonal background trends are removed using variance-preserving spectral filtering, and time-lagged features are included based on correlation analyses to better represent the physical memory of the system. Separate RF models developed for summer and fall are evaluated using within-season validation. Results show moderate overall predictive skill, with clear differences between daytime and nighttime performance. Nighttime predictions consistently explain more of the observed variability in log₁₀(??ₙ²) than daytime predictions, particularly during summer. This improved nighttime performance may be linked to more stable and less varying turbulence conditions after sunset. In contrast, daytime predictions show larger errors, likely due to stronger convection and rapidly changing surface-layer processes. Overall, this study shows that the reliability of machine-learning-based ??ₙ² predictions depend strongly on temporal structure, diurnal regime, and how autocorrelation is handled. The iv findings highlight the importance of using regime-aware validation and variance-based diagnostics rather than relying only on single accuracy metrics. This work provides practical insight into when ML-based turbulence predictions can be trusted in coastal marine environments and offers guidance for developing more robust and physically consistent forecasting approaches.", "source_url": "https:\/\/openalex.org\/W7143310057", "pdf_url": "", "doi": "10.62791\/20574", "handle": "", "repository_name": "", "source_record_id": "W7143310057", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:48", "updated_at": "2026-11-15 16:52:48" }, { "id": 503, "title": "Fake news detection with Graph Neural Networks", "title_ar": "", "title_en": "Fake news detection with Graph Neural Networks", "author": "Go, Jin Ho", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "", "university_source": "", "college": "", "department": "", "country": "Australia", "governorate": "", "publication_year": 2025, "language": "en", "research_domain": "deep-learning", "research_subdomain": "gnn", "categories": "deep-learning; gnn", "keywords": "", "abstract": "The rapid advancement of technology and the accelerating dissemination of information have contributed to a significant increase in the prevalence of fake news, defined as mislead- ing or deliberately deceptive news articles. This proliferation poses a significant challenge, as it can erode trust in institutions and the media, incite violence and hatred, influence elections and public policy, discourage civic engagement, and fracture social cohesion. Therefore, developing e\"ective and e!cient strategies for detecting and intercepting on- line fake news is crucial. This thesis makes three research contributions focused on fake news detection, each employing Graph Neural Networks (GNNs) and leveraging diverse graph data features to model social network. The first contribution addresses the evolving relationships between various entity types within a social network over time. We model the social network as a Dynamic Hetero- geneous Information Network (DHIN), where \"dynamic\" captures temporal changes and \"heterogeneous\" represents multiple node and edge types. To mitigate the high compu- tational cost typically associated with DHINs, we designed a GNN model incorporating a dynamic update mechanism. This approach significantly reduces the training time re- quired for DHINs by 35% to 45% compared to existing techniques. The second contribution focuses on early-stage fake news detection by examining prop- agation patterns during the initial phases of news dissemination. We developed a GNN architecture that separately trains a homogeneous information network (representing news propagation) and a heterogeneous information network (representing the broader social context). A fusion mechanism integrates the representations from both networks at desig- nated intervals. Findings demonstrate that early-stage propagation patterns o\"er strong indicators of a news item’s veracity. Within the initial two-hour period, our model achieves approximately 80% accuracy on real-world datasets. The third contribution investigates the largely unexplored influence of the author-topic relationship on news veracity. We utilize author-topic modelling to establish these re- lationships and represent them using hypergraphs, which capture complex interactions among multiple vertices. Furthermore, by establishing relationships between hyperedges, we model higher-order interactions without introducing additional nodes. On datasets containing author information, our model demonstrates outstanding performance com- pared to existing methods. Consequently, a significant deviation in an author’s observed topic distribution from their expected distribution, as derived from the author-topic rela- tionship, can serve as a strong signal for fake news detection.", "source_url": "https:\/\/openalex.org\/W7106819264", "pdf_url": "https:\/\/unsworks.unsw.edu.au\/bitstreams\/1b056a5d-aa7f-4879-8058-572735bd50b3\/download", "doi": "10.26190\/unsworks\/31844", "handle": "", "repository_name": "UNSWorks (UNSW Sydney)", "source_record_id": "W7106819264", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:50", "updated_at": "2026-11-15 16:52:50" }, { "id": 541, "title": "TOWARDS TRUSTWORTHY NATURAL LANGUAGE PROCESSING", "title_ar": "", "title_en": "TOWARDS TRUSTWORTHY NATURAL LANGUAGE PROCESSING", "author": "HANNAH AUGHT SCHWARTZ BROWN", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "", "university_source": "", "college": "", "department": "", "country": "Singapore", "governorate": "", "publication_year": 2026, "language": "en", "research_domain": "nlp", "research_subdomain": "fairness", "categories": "nlp; responsible-ai; fairness", "keywords": "", "abstract": "", "source_url": "https:\/\/openalex.org\/W7163603768", "pdf_url": "", "doi": "", "handle": "", "repository_name": "National University of Singapore", "source_record_id": "W7163603768", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:50", "updated_at": "2026-11-15 16:52:50" }, { "id": 590, "title": "Discriminator-driven diffusion generative adversarial network for synthetic medical image generation", "title_ar": "", "title_en": "Discriminator-driven diffusion generative adversarial network for synthetic medical image generation", "author": "Dennis Chia Yin Lim", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "", "university_source": "", "college": "", "department": "", "country": "Australia", "governorate": "", "publication_year": 2026, "language": "en", "research_domain": "medical-ai", "research_subdomain": "medical-imaging", "categories": "deep-learning; computer-vision; generative-ai; medical-ai; gan; medical-imaging", "keywords": "", "abstract": "This research project addresses the significant challenge of class imbalance in medical deep learning classification models, particularly when the abnormal class is underrepresented. Class imbalance can severely affect the classification performance of deep learning models, potentially leading to patient misdiagnosis and impeding clinician trust in AI adoption. In healthcare, class imbalance arises due to various factors, including the rarity of diseases or medical scenarios and data insufficiency for specific classes within large datasets. Existing solutions, such as traditional data augmentation and Generative Adversarial Networks (GANs) like DCGAN, as well as Conditional GANs like Pix2Pix and Stable Diffusion, have attempted to mitigate this issue, but with limitations.", "source_url": "https:\/\/openalex.org\/W7131907591", "pdf_url": "", "doi": "10.25916\/sut.31368169", "handle": "", "repository_name": "Swinburne figshare (Swinburne University of Technology)", "source_record_id": "W7131907591", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:51", "updated_at": "2026-11-15 16:52:51" }, { "id": 648, "title": "Safety of Machine Learning Systems in Autonomous Driving", "title_ar": "", "title_en": "Safety of Machine Learning Systems in Autonomous Driving", "author": "Fadi Al-Khoury", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "KTH Royal Institute of Technology", "university_source": "repository_host", "college": "", "department": "", "country": "Sweden", "governorate": "", "publication_year": 2017, "language": "en", "research_domain": "nlp", "research_subdomain": "speech-recognition", "categories": "machine-learning; deep-learning; computer-vision; nlp; robotics; responsible-ai; ai-applications; speech-recognition; fairness; ai-industry", "keywords": "", "abstract": "Machine Learning, and in particular Deep Learning, are extremely capable tools for solving problems which are difficult, or intractable to tackle analytically. Application areas include pattern recognition, computer vision, speech and natural language processing. With the automotive industry aiming for increasing amount of automation in driving, the problems to solve become increasingly complex, which appeals to the use of supervised learning methods from Machine Learning and Deep Learning. With this approach, solutions to the problems are learned implicitly from training data, and inspecting their correctness is not possible directly. This presents concerns when the resulting systems are used to support safety-critical functions, as is the case with autonomous driving of automotive vehicles. This thesis studies the safety concerns related to learning systems within autonomous driving and applies a safety monitoring approach to a collision avoidance scenario. Experiments are performed using a simulated environment, with a deep learning system supporting perception for vehicle control, and a safety monitor for collision avoidance. The related operational situations and safety constraints are studied for an autonomous driving function, with potential faults in the learning system introduced and examined. Also, an example is considered for a measure that indicates trustworthiness of the learning system during operation.", "source_url": "https:\/\/openalex.org\/W2770758792", "pdf_url": "", "doi": "", "handle": "", "repository_name": "KTH Publication Database DiVA (KTH Royal Institute of Technology)", "source_record_id": "W2770758792", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:51", "updated_at": "2026-11-15 16:52:51" }, { "id": 675, "title": "Deep Learning for Skeleton-Based Human Action Recognition", "title_ar": "", "title_en": "Deep Learning for Skeleton-Based Human Action Recognition", "author": "Sohaïb Laraba", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "", "university_source": "", "college": "", "department": "", "country": "Belgium", "governorate": "", "publication_year": 2020, "language": "en", "research_domain": "medical-ai", "research_subdomain": "video-analysis", "categories": "machine-learning; deep-learning; computer-vision; medical-ai; robotics; responsible-ai; cnn; video-analysis; xai", "keywords": "", "abstract": "Human action recognition from videos has a wide range of applications, including video surveillance and security, human-computer interaction, robotics, health care, etc. Nowadays, 3D skeleton-based action recognition has drawn increasing attention thanks to the availability of low-cost motion capture devices, and accessibility of large-scale 3D skeleton datasets, in addition to real-time skeleton estimation algorithms. In the first part of this thesis, we present a novel representation of motion capture sequences for 3D skeleton-based action recognition. The proposed approach consists of representing the 3D skeleton sequences into RGB image-like data and leveraging recent convolutional neural networks (CNNs) to model the long-term temporal and spatial structural information for action recognition. Extensive experiments have shown the superiority of the proposed approach over the state-of-the-art methods for 3D skeleton-based action recognition. In order to extract skeleton sequences, different devices extract first the depth information using multiple technologies (stereo, time-of-flight, etc.), then 3D skeleton poses are extracted using different algorithms. In the very late years, new researches proposed to extract skeleton sequences directly from RGB videos. The most precise methods extract 2D skeletons in real-time and with high accuracy. In the second part of this thesis, we leverage these tools to extend the use of our proposed approach to RGB videos. We first extract 2D skeleton sequences from RGB videos, and then, following approximately the same process as in the first part, we use CNNs for human action recognition. Different experiments showed that the proposed method outperforms different state-of-the-art methods on a large benchmark dataset. Another contribution of this thesis relates to the interpretability of deep learning models. Deep learning models are still considered alchemy due to the lack of understanding of their internal operations. Interpretability is a crucial task to understand and trust the decisions made by the machine learning model. Thus, we propose in the third part of this thesis to use CNN interpretation methods to understand the behavior of our classifier and extract the most informative joints during the execution of a particular action. This method allows us to see from the CNN point of view the most important joints, and understand why certain actions are confused by the proposed classifier.", "source_url": "https:\/\/openalex.org\/W3093549486", "pdf_url": "", "doi": "10.13140\/rg.2.2.30723.53283", "handle": "", "repository_name": "ORBi UMONS", "source_record_id": "W3093549486", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:52", "updated_at": "2026-11-15 16:52:52" }, { "id": 686, "title": "Trustworthiness in Vertical Federated Learning: Privacy attacks, Robustness Vulnerabilities, and Secure Protocols", "title_ar": "", "title_en": "Trustworthiness in Vertical Federated Learning: Privacy attacks, Robustness Vulnerabilities, and Secure Protocols", "author": "Duanyi Yao", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "", "university_source": "", "college": "", "department": "", "country": "", "governorate": "", "publication_year": 2025, "language": "en", "research_domain": "medical-ai", "research_subdomain": "federated-learning", "categories": "machine-learning; medical-ai; responsible-ai; ai-applications; federated-learning; fairness; ai-cybersecurity", "keywords": "", "abstract": "Vertical Federated Learning (VFL) enables organizations with distinct features of the same data samples to collaboratively train machine learning models without directly sharing raw data. While VFL offers strong potential for privacy-preserving analytics in domains such as finance and healthcare, its trustworthiness remains insufficiently understood. This thesis provides a systematic study of VFL trustworthiness along three dimensions: privacy, robustness, and secure protocol design. First, we investigate privacy vulnerabilities by introducing URVFL, an undetectable data reconstruction attack. By generating malicious gradients through a discriminator with auxiliary classifier, URVFL achieves state-of-the-art reconstruction fidelity while evading existing detection mechanisms, revealing the fragility of current defenses. Second, we analyze robustness vulnerabilities by examining adversarial example attacks in VFL inference. We formulate adaptive corruption of client-server channels as an online optimization problem and propose the E-TS algorithm, which leverages empirical Thompson sampling to identify optimal corruption patterns with provable regret bounds. Our results demonstrate that adaptive adversaries can severely compromise VFL robustness. Finally, we design FedVS, a synchronous split VFL protocol that simultaneously provides straggler resilience and information-theoretic privacy. Through coding-based secret sharing, FedVS losslessly aggregates embeddings even with stragglers while preventing leakage to curious servers and colluding clients. This thesis demonstrates that trustworthy VFL requires a holistic approach addressing privacy, robustness, and protocol-level vulnerabilities simultaneously. The proposed attacks expose critical weaknesses that must be addressed before large-scale deployment, while our secure protocol provides a constructive foundation for privacy-preserving collaborative learning. These contributions advance understanding of trustworthiness in distributed machine learning and establish principled design criteria for secure VFL systems.", "source_url": "https:\/\/openalex.org\/W7124459682", "pdf_url": "", "doi": "10.14711\/thesis-hdl167715", "handle": "", "repository_name": "", "source_record_id": "W7124459682", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:52", "updated_at": "2026-11-15 16:52:52" }, { "id": 690, "title": "Privacy-preserving federated learning framework", "title_ar": "", "title_en": "Privacy-preserving federated learning framework", "author": "Wenxuan Huang", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "University of Southampton", "university_source": "repository_host", "college": "", "department": "", "country": "United Kingdom", "governorate": "", "publication_year": 2025, "language": "en", "research_domain": "responsible-ai", "research_subdomain": "federated-learning", "categories": "responsible-ai; ai-applications; federated-learning; fairness; ai-cybersecurity", "keywords": "", "abstract": "Federated Learning (FL) enables multiple parties to collaboratively train a shared model without exposing raw data. However, FL often suffers from degraded performance due to three core challenges: non-IID data distributions, residual privacy risks, and limited data volume per node. This work explores the privacy–performance trade-offs arising from these challenges and proposes a unified framework combining partial data sharing, anonymisation, and data augmentation. We first investigate how non-IID distributions impair standard FL convergence, and propose a Cluster-of-Trust (CoT) mechanism that groups mutually trusted clients for hierarchical aggregation. Experiments on the IoT-23 intrusion detection dataset show that CoT significantly improves model accuracy and convergence speed, closing most of the gap to centralised training. Next, we compare combinations of privacy-preserving techniques—including differential privacy, data anonymisation, and CoT-based grouping—highlighting how these impact model utility. Our results demonstrate that strong privacy guarantees can be achieved with only modest accuracy loss and reduced communication overhead. To address the limited data volume problem, we introduce data augmentation strategies and test their effectiveness in a privacy-sensitive Human Activity Recognition (HAR) task. We find that combining augmentation with trust-aware aggregation improves generalisation under strict privacy constraints. Across two domains—cybersecurity and HAR—we systematically evaluate this framework and demonstrate its effectiveness and generalisability. Our results show that strong privacy can be reconciled with competitive performance, offering practical insights for deploying FL in real-world, heterogeneous environments.", "source_url": "https:\/\/openalex.org\/W7140959527", "pdf_url": "https:\/\/eprints.soton.ac.uk\/510212\/1\/Privacy_Preserving_Federated_Learning_Framework_2_.pdf", "doi": "", "handle": "", "repository_name": "ePrints Soton (University of Southampton)", "source_record_id": "W7140959527", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:52", "updated_at": "2026-11-15 16:52:52" }, { "id": 693, "title": "ACHIEVING PARTICIPATIVE AND PRIVACY-PRESERVING FEDERATED LEARNING", "title_ar": "", "title_en": "ACHIEVING PARTICIPATIVE AND PRIVACY-PRESERVING FEDERATED LEARNING", "author": "Jianan Chen", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "", "university_source": "", "college": "", "department": "", "country": "", "governorate": "", "publication_year": 2025, "language": "en", "research_domain": "responsible-ai", "research_subdomain": "federated-learning", "categories": "machine-learning; responsible-ai; federated-learning; fairness", "keywords": "", "abstract": "The proliferation of data-driven intelligence has reshaped the foundations of artificial intelligence and machine learning, enabling unprecedented collaboration across distributed entities. Yet, as organizations increasingly rely on shared computation and collective learning, concerns over privacy, fairness, and accountability have become central. Federated Learning (FL) has emerged as a promising paradigm that allows multiple institutions to collaboratively train models without disclosing raw data. However, despite its decentralized structure, FL is not inherently trustworthy. Rational participants may contribute unevenly or strategically, privacy guarantees often degrade model utility, and once data contributions are incorporated, their removal becomes nearly impossible. These challenges reveal that building trustworthy federated intelligence requires more than algorithmic optimizationit demands principled mechanisms that ensure participation, protect privacy, and enable verifiable unlearning. This dissertation addresses these intertwined challenges through a unified theoretical and algorithmic framework for participative, privacy-preserving, and accountable federated learning. The first part investigates participation incentives in cross-silo FL, where multiple organizations jointly train a global model. The interactions among organizations are modeled as a public goods game, and theoretical analysis proves that a social dilemma arises in which the Nash equilibrium fails to achieve maximum social welfare. To resolve this inefficiency, the Multi-player Multi-action Zero-Determinant (MMZD) strategy and its cooperative extension, the MMZD Alliance (MMZDA), are developed to maximize social welfare. These strategies enable each organization, or a coalition of them, to unilaterally control and enhance the global welfare without incurring additional cost. Experimental results verify that MMZD effectively achieves the theoretical social optimum, and that MMZDA further improves the maximum attainable welfare. The second part focuses on privacy preservation in Hierarchical Federated Learning (HFL), where a client-edge-cloud structure accelerates training and reduces communication cost but increases the risk of privacy leakage. To address this issue, a novel differential privacy definition tailored for HFL, Group Local Differential Privacy (GLDP), is proposed, along with the Sampling-Randomizing-Shuffling (SRS) mechanism to implement it. SRS combines randomized response and shuffling with a sampling process that achieves stronger privacy protection while adding less noise. Comprehensive theoretical analyses provide guarantees for both privacy and convergence, and experiments on real-world datasets demonstrate that SRS effectively balances privacy protection and model performance in HFL. The final part explores federated unlearning (FUI) under Differentially Private Federated Learning (DPFL), enabling the removal of a clients data influence while maintaining privacy guarantees. Unlike existing federated unlearning methods that fail under DP noise, the proposed Federated Unlearning with Indistinguishability (FUI) framework performs local model retraction followed by global noise calibration, yielding an unlearned model that is statistically indistinguishable from the retrained counterpart. A two-stage Stackelberg game is formulated to derive optimal unlearning strategies for both the server and the target client. Privacy and convergence analyses provide formal guarantees, and experiments on four benchmark datasets show that FUI achieves superior accuracy and efficiency compared with existing unlearning approaches. Extensive empirical evaluations on datasets such as Adult, Purchase, MNIST, and CIFAR10 demonstrate that the proposed mechanisms-MMZDA for incentive alignment, GLDP with SRS for hierarchical privacy preservation, and FUI for accountable deletionjointly enhance the reliability and practicality of federated learning. Together, these contributions establish a cohesive framework that advances the development of participative, privacy-preserving, and trustworthy federated intelligence.", "source_url": "https:\/\/openalex.org\/W7114782219", "pdf_url": "", "doi": "10.25394\/pgs.30804161", "handle": "", "repository_name": "Figshare", "source_record_id": "W7114782219", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:52", "updated_at": "2026-11-15 16:52:52" }, { "id": 697, "title": "On the limits of privacy in federated learning : personalization and adversarial analysis", "title_ar": "", "title_en": "On the limits of privacy in federated learning : personalization and adversarial analysis", "author": "Sargam Gupta", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "KTH Royal Institute of Technology", "university_source": "repository_host", "college": "", "department": "", "country": "Sweden", "governorate": "", "publication_year": 2026, "language": "en", "research_domain": "deep-learning", "research_subdomain": "transformers", "categories": "machine-learning; deep-learning; computer-vision; responsible-ai; ai-applications; predictive-analytics; transformers; face-recognition; federated-learning; fairness; ai-cybersecurity; ai-industry; ai-smart-cities", "keywords": "", "abstract": "With the rapid adoption of Artificial Intelligence across diverse applications, concerns around data privacy have become increasingly prominent. Traditional centralized machine learning approaches require aggregating large volumes of user data at a central server, posing risks related to privacy, scalability, and regulatory compliance. Federated Learning has emerged as a decentralized alternative, enabling clients to train models locally while only sharing model updates with a central server. It addresses some of the concerns by enabling multiple participants to collaboratively train a shared model while keeping their raw data locally stored. However, data locality alone does not guarantee privacy. Information about local training records may still be revealed through model parameters, gradients, updates, or predictions.This makes FL vulnerable to privacy attacks such as model poisoning, data reconstruction and membership inference attacks. This thesis investigates these limitations in FL through two complementary perspectives: adversarial evaluation of the framework and personalization of privacy protection. In this thesis, we first examine the validity of various data anonymization techniques for facial images through the re-identification attacks. A weight-optimized record linkage method is developed to determine whether anonymized image records can be linked to their corresponding original records. The evaluation considers different adversarial settings, including cases in which an attacker possesses only partial information. The findings demonstrate that visually modifying an image or separating identity-related information does not necessarily prevent re-identification. We also experiment with Shapley values to check whether they could be useful in detecting model poisoning attacks.Then, we develop two privacy-enhancing mechanisms for federated time-series forecasting. The first framework is centered around using differential privacy with transformers specifically for traffic flow data whereas the other framework is a microaggregation-based model agnostic k-anonymous federated forecasting framework. The results demonstrate that useful forecasting performance can be maintained while raw data remain decentralized and additional protection is applied at the client level. One of the major challenges that exist in FL is heterogeneity. There have been several studies in the literature on data and device heterogeneity in FL. However, the heterogeneity that exists in the privacy requirements of different clients is widely unexplored. Building on these findings, this thesis introduces a personalized privacy-preserving federated learning framework that allows individual clients to select their preferred privacy mechanism and protection level. The proposed FedPUP framework supports the simultaneous use of differential privacy and k-anonymity within a federated round. We also used Membership Inference attacks to empirically evaluate the information leakage associated with different privacy choices and to identify combinations of ε and k that provide comparable levels of protection. We also propose a privacy-aware server-side aggregation algorithm that incorporates client-level privacy characteristics during the aggregation process in each training round. Our empirical contributions are also complemented by an exploratory study investigating the practical requirements for deploying privacy-preserving federated learning systems in the industry. Our study highlights the importance of transparency, accountability, trust, costs, and privacy–utility trade-offs. Overall, this thesis demonstrates that federated learning should not be considered inherently private merely because data remain decentralized. By combining adversarial evaluation with personalized privacy-preserving mechanisms, it provides a more comprehensive understanding of privacy risks and safeguards in FL. The findings offer practical guidance for developing federated learning systems that are more privacy-aware, adaptable, and suitable for real-world deployment.", "source_url": "https:\/\/openalex.org\/W7214512574", "pdf_url": "", "doi": "", "handle": "", "repository_name": "KTH Publication Database DiVA (KTH Royal Institute of Technology)", "source_record_id": "W7214512574", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:52", "updated_at": "2026-11-15 16:52:52" }, { "id": 701, "title": "Mitigating Poisoning Threats and Their Privacy Implications in Federated Learning", "title_ar": "", "title_en": "Mitigating Poisoning Threats and Their Privacy Implications in Federated Learning", "author": "Chamara Sandeepa", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "University College Dublin", "university_source": "repository_host", "college": "", "department": "", "country": "Ireland", "governorate": "", "publication_year": 2026, "language": "en", "research_domain": "medical-ai", "research_subdomain": "gan", "categories": "machine-learning; generative-ai; medical-ai; responsible-ai; ai-applications; data-mining; gan; xai; federated-learning; fairness; ai-cybersecurity; ai-industry", "keywords": "", "abstract": "Advancements in Machine Learning (ML) and distributed intelligence have accelerated the adoption of Federated Learning (FL) as a key enabler of privacy-preserving applications in domains such as IoT, healthcare, and Beyond 5G networks. FL enables collaborative model training without centralising data, but its distributed and heterogeneous nature introduces new security and privacy challenges. In addition to conventional adversarial threats, FL faces unique vulnerabilities including poisoning attacks, and privacy leakage through model updates via threats like inference attacks. Addressing these challenges is critical for ensuring trust in next-generation distributed AI systems. This thesis investigates such threats in FL with a particular emphasis on poisoning attacks and the interplay between poisoning and privacy leakage. The work also investigates how poisoning attacks can be mitigated, essentially preventing any privacy vulnerabilities induced via poisoning. To this end, four major contributions are presented: Rec-Def, SHERPA, X-SHIELD, and ROAM. First, Rec-Def is proposed as a defence against GAN-based inference attacks, highlighting the connection between poisoning and privacy risks. Second, SHERPA introduces an interpretable aggregation algorithm that employs Explainable AI (XAI) technique called SHAP feature attributions to detect poisoned model updates, improving the transparency and reliability of poisoning detection. Third, X-SHIELD extends this approach to hierarchical FL, combining explanation-guided detection with scalable clustering to address robustness in large-scale FL deployments. Finally, the thesis explores the extension of defences and its limitations on fully Decentralised FL (DFL) architectures. For this, the thesis proposes an offensive use of XAI via an new stealthy poisoning attack called ROAM, based on Layerwise Relevance Propagation (LRP) and designs a countermeasure that perturbs sensitive neurons to mitigate privacy leakage via poisoning. The effectiveness of the proposed approaches is validated through extensive experiments across multiple datasets, learning scenarios, and adversarial settings. Results demonstrate that the solutions enhance robustness and privacy while remaining scalable and adaptable to diverse FL environments. Overall, this thesis contributes to advancing the safety and trustworthiness of FL by investigating novel vulnerabilities introducing and solutions that balance privacy preservation, poisoning robustness, and justifiability. These contributions provide both theoretical insights and design mechanisms for deploying FL securely in critical infrastructures such as 6G networks, IoT ecosystems, and industrial applications.", "source_url": "https:\/\/openalex.org\/W7165993495", "pdf_url": "", "doi": "", "handle": "", "repository_name": "Research Repository UCD (University College Dublin)", "source_record_id": "W7165993495", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:52", "updated_at": "2026-11-15 16:52:52" }, { "id": 705, "title": "Privacy and Security Enhanced Federated Learning Framework Design", "title_ar": "", "title_en": "Privacy and Security Enhanced Federated Learning Framework Design", "author": "Yasaman Pakdel", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "", "university_source": "", "college": "", "department": "", "country": "", "governorate": "", "publication_year": 2024, "language": "en", "research_domain": "medical-ai", "research_subdomain": "federated-learning", "categories": "machine-learning; medical-ai; responsible-ai; federated-learning; fairness", "keywords": "", "abstract": "Federated learning, while promising for its ability to train models across decentralized data sources, presents both security challenges and solutions.Its security depends on how well these challenges are addressed.On one hand, federated learning can be secure when proper privacy-preserving techniques are employed, such as encryption, secure aggregation, and diferential privacy, to protect data and model updates.Participant authentication, secure communication, and compliance with data protection regulations further contribute to its security.However, federated learning may also be vulnerable if not carefully implemented, potentially exposing sensitive data or risking model integrity.By encrypting and processing data using Crypten, federated learning participants can collaborate on model training without exposing their raw data, ensuring data privacy and confdentiality.Crypten's features, such as secure aggregation and encrypted computation, further bolster the trustworthiness of federated learning, making it an indispensable tool for applications in healthcare, fnance, and other sectors where sensitive data protection is paramount.In an era of growing data privacy concerns, Crypten empowers federated learning to thrive in a secure and privacy-preserving manner.In this project, we embarked on a comprehensive exploration of machine learning methodologies, starting with traditional plaintext machine learning techniques.Subsequently, we", "source_url": "https:\/\/openalex.org\/W4400228779", "pdf_url": "http:\/\/scholarworks.calstate.edu\/downloads\/70795g815", "doi": "10.46569\/9s161f47k", "handle": "", "repository_name": "", "source_record_id": "W4400228779", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:52", "updated_at": "2026-11-15 16:52:52" }, { "id": 706, "title": "Towards Robust and Privacy-Preserving Federated Learning", "title_ar": "", "title_en": "Towards Robust and Privacy-Preserving Federated Learning", "author": "Sana Awan", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "The University of Kansas", "university_source": "repository_host", "college": "", "department": "", "country": "United States", "governorate": "", "publication_year": 2024, "language": "en", "research_domain": "medical-ai", "research_subdomain": "federated-learning", "categories": "machine-learning; medical-ai; responsible-ai; ai-applications; predictive-analytics; federated-learning; fairness; ai-cybersecurity", "keywords": "", "abstract": "Machine Learning (ML) has revolutionized various fields, from disease prediction to credit risk evaluation, by harnessing abundant data scattered across diverse sources. However, transporting data to a trusted server for centralized ML model training is not only costly but also raises privacy concerns, particularly with legislative standards like Health Insurance Portability and Accountability Act (HIPAA) in place. In response to these challenges, Federated Learning (FL) has emerged as a promising solution. FL involves training a collaborative model across a network of clients, each retaining its private data. By conducting training locally on the participating clients, this approach eliminates the need to transfer entire training datasets while harnessing their computation capabilities. However, FL introduces unique privacy risks, security concerns, and robustness challenges. Firstly, due to the statistical and system heterogeneity within FL, substantial differences exist between the distribution of each local dataset and the global distribution, causing clients’ local objectives to deviate greatly from the global optima, resulting in a drift in local updates. Secondly, FL is susceptible to malicious actors who may tamper with local data, manipulate the local training process, or intercept the shared model or gradients to implant backdoors that affect the robustness of the joint model. Addressing such vulnerabilities and challenges is crucial before deploying FL systems in critical infrastructures. In this dissertation, we present a multi-pronged approach to address the privacy, security, and robustness challenges in FL. This involves designing innovative robust aggregation schemes and privacy protection mechanisms to counter attacks during the training process. We studied the challenges in FL due to heterogeneous data distributions and found that existing FL algorithms often suffer from slow and unstable convergence and are vulnerable to poisoning attacks, particularly in extreme non-independent and identically distributed (non-IID) settings. We proposed a robust aggregation scheme, named CONTRA, to mitigate data poisoning attacks and ensure an accuracy guarantee even under attack. This defense strategy identifies malicious clients by evaluating the cosine similarity of their gradient contributions and subsequently removes them from FL training. Finally, we introduced FL-GMM, an algorithm designed to tackle data heterogeneity while prioritizing privacy. It iteratively constructs a personalized classifier for each client while aligning local-global feature representations. By aligning local distributions with global semantic information, FL-GMM minimizes the impact of data diversity. Moreover, FL-GMM enhances security by transmitting derived model parameters via secure multiparty computation, thereby avoiding vulnerabilities to reconstruction attacks observed in other approaches. To address the privacy risk due to model or gradient interception, we present the design of a reliable and accountable blockchain-enabled Privacy-Preserving Federated Learning (PPFL) framework that leverages homomorphic encryption to protect individual client updates. The blockchain is adopted to support the provenance of model updates during training so that malformed or malicious updates can be identified and traced back to the source.", "source_url": "https:\/\/openalex.org\/W7155578449", "pdf_url": "", "doi": "", "handle": "", "repository_name": "KU ScholarWorks (The University of Kansas)", "source_record_id": "W7155578449", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:52", "updated_at": "2026-11-15 16:52:52" }, { "id": 709, "title": "Privacy-preserving detection of poisoning attacks in federated learning", "title_ar": "", "title_en": "Privacy-preserving detection of poisoning attacks in federated learning", "author": "Trent Muhr", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "", "university_source": "", "college": "", "department": "", "country": "", "governorate": "", "publication_year": 2024, "language": "en", "research_domain": "responsible-ai", "research_subdomain": "federated-learning", "categories": "responsible-ai; ai-applications; federated-learning; fairness; ai-cybersecurity", "keywords": "", "abstract": "With federated learning, local learners train a shared global model using their own data, and report model updates to a server to aggregate and then update the global model. Such a learning paradigm may suffer from two attacks: privacy attacks by the untrusted server; adversarial attacks (e.g., poisoning attacks) by malicious learners. There is extensive research on addressing each of the attacks separately, but there is no scheme that can address both of them. In this paper, we propose a scheme that enables both privacy-preserving aggregation and poisoning attack detection at the server, by utilizing additive homomorphic encryption and a trusted execution environment (TEE). Our evaluation based on an implemented prototype system demonstrates that our scheme can attain a similar level of detection accuracy as the state-of-the-art poisoning detection scheme, and that the increased computational workload can be parallelized and mostly executed outside of the TEE. A privacy analysis shows that the proposed scheme can protect individual learners' model updates from being exposed.", "source_url": "https:\/\/openalex.org\/W4410025063", "pdf_url": "", "doi": "10.31274\/td-20250502-190", "handle": "", "repository_name": "", "source_record_id": "W4410025063", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:52", "updated_at": "2026-11-15 16:52:52" }, { "id": 735, "title": "Adaptive Deep Learning for Vision and Multimodal Understanding with Heterogeneous Data", "title_ar": "", "title_en": "Adaptive Deep Learning for Vision and Multimodal Understanding with Heterogeneous Data", "author": "Youwei Liang", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "", "university_source": "", "college": "", "department": "", "country": "United States", "governorate": "", "publication_year": 2026, "language": "en", "research_domain": "nlp", "research_subdomain": "llm", "categories": "machine-learning; deep-learning; computer-vision; nlp; medical-ai; transfer-learning; few-zero-shot; predictive-analytics; transformers; multimodal; image-segmentation; llm", "keywords": "", "abstract": "Large-scale pretraining has produced powerful deep learning representations, but real-world data rarely match the clean, uniform assumptions of standard transfer learning. This dissertation uses heterogeneous data to describe settings in which information differs in modality, representation, relevance, quality, or frequency across inputs, outputs, tasks, or supervision. Examples include molecules represented as graphs, images, strings, and text; images containing both informative and redundant regions; noisy or incomplete labels; and rare but important classes. Applying pretrained models in these settings requires deciding which information to process, how to connect different forms of information, and how much to trust each supervision signal. This dissertation develops adaptive learning as a unifying approach through three paradigms: adaptive selection focuses computation on useful input components; adaptive alignment maps heterogeneous modalities, tasks, and outputs into a shared predictive space; and adaptive supervision adjusts imperfect or imbalanced training signals. The first contribution, EViT, demonstrates adaptive selection for efficient Vision Transformers. Standard Vision Transformers devote equal computation to all image patches despite substantial redundancy. EViT uses class-token attention to identify important tokens, preserves them in subsequent layers, and fuses inattentive tokens into a compact representation rather than discarding their context. This progressive token reorganization reduces computation while maintaining accurate visual recognition. The second contribution, DrugChat, demonstrates adaptive alignment for multimodal molecular prediction. Molecules can be represented as chemical graphs, rendered images, SMILES strings, and natural-language prompts, while predictions may require labels, probabilities, categorical decisions, or free-form descriptions. DrugChat encodes molecular graphs and images with pretrained encoders, projects them into the latent space of an instruction-following language model, and conditions generation on molecular structure and task prompts. It thereby aligns heterogeneous inputs in a shared representation space and expresses diverse prediction targets through a common language-generation interface. The third contribution, RobPicker, demonstrates adaptive supervision for cryo-electron tomography particle picking, where annotations are expensive, incomplete, noisy, and highly imbalanced across macromolecular species. Its meta-learning framework jointly trains a segmentation network, a label correction network, and a data reweighting network. Guided by cleaner validation data, these components refine imperfect labels and adapt each subtomogram's contribution to the loss, giving greater influence to rare or small particle classes. Together, EViT, DrugChat, and RobPicker apply adaptation at complementary levels of the learning pipeline: selecting what data to process, aligning heterogeneous representations and outputs, and learning from imperfect supervision.", "source_url": "https:\/\/openalex.org\/W7222391945", "pdf_url": "https:\/\/escholarship.org\/content\/qt3h17286t\/qt3h17286t.pdf", "doi": "", "handle": "", "repository_name": "eScholarship (California Digital Library)", "source_record_id": "W7222391945", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:52", "updated_at": "2026-11-15 16:52:52" }, { "id": 738, "title": "AI-Generated Image Detection Using Multimodal Deep Learning", "title_ar": "", "title_en": "AI-Generated Image Detection Using Multimodal Deep Learning", "author": "Mamadou Keïta", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "", "university_source": "", "college": "", "department": "", "country": "France", "governorate": "", "publication_year": 2026, "language": "en", "research_domain": "nlp", "research_subdomain": "llm", "categories": "machine-learning; deep-learning; computer-vision; nlp; generative-ai; responsible-ai; few-zero-shot; transformers; multimodal; llm; gan; diffusion; xai; fairness", "keywords": "", "abstract": "Generative artificial intelligence (AI) and its applications offer significant benefits but raise potentially critical societal and ethical concerns. This thesis explores the field of synthetic content generation and detection, with a particular focus on synthetic images. These are AI-generated images that can appear indistinguishable from real photographs, creating new challenges for digital forensics, media integrity, and public trust. While generative models such as generative adversarial networks (GANs) and diffusion models have evolved rapidly, most existing detection methods remain limited in terms of generalizability, robustness, and interpretability. To address these challenges, this research investigates new AI-driven frameworks to improve the generalizability, robustness, and interpretability of current image detection methods.The various aspects of synthetic image detection are covered in this thesis by four key contributions. The first contribution, Bi-LORA, presents an efficient vision-language approach that reformulates the detection problem into an image caption task, achieving strong zero-shot generalization across unseen generative models. The second, RAVID, proposes a retrieval-augmented visual detection framework that enhances robustness and interpretability by incorporating external visual context. The third, DeeCLIP, presents a lightweight transformer-based model that fuses shallow and deep features to improve resistance to image degradation and post-processing. Finally, FIDAVL unifies synthetic image detection and source attribution in a single multitasking framework based on the soft prompt tuning of vision-language models.This thesis provides an in-depth analysis of the current state of synthetic image research and its societal impacts. It emphasizes the urgent need for effective and robust detection methods. The approaches introduced advance the scientific community and could benefit practical applications in media authenticity and security. This research marks an important step forward in developing solutions to the ethical challenges posed by GenAI.", "source_url": "https:\/\/openalex.org\/W7126004607", "pdf_url": "", "doi": "", "handle": "", "repository_name": "HAL", "source_record_id": "W7126004607", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:52", "updated_at": "2026-11-15 16:52:52" }, { "id": 743, "title": "A Study on Robustness, Adaptation, and Generalization for Visual and Multimodal Learning", "title_ar": "", "title_en": "A Study on Robustness, Adaptation, and Generalization for Visual and Multimodal Learning", "author": "Xiasi Wang", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "", "university_source": "", "college": "", "department": "", "country": "", "governorate": "", "publication_year": 2026, "language": "en", "research_domain": "nlp", "research_subdomain": "llm", "categories": "machine-learning; deep-learning; computer-vision; nlp; responsible-ai; ai-applications; transfer-learning; self-semi-supervised; multimodal; llm; fairness; ai-cybersecurity", "keywords": "", "abstract": "Recent advances in visual and multimodal learning have demonstrated impressive performance on diverse tasks. However, their reliability in real-world scenarios remains challenged by issues of robustness, adaptation, and generalization. In this thesis, we tackle these challenges by establishing four integral pillars of reliable deep learning: (1) robustness against adversarial attacks, (2) adaptation to sequential data streams, (3) resilience to domain shift and unseen categories, and (4) enhanced generalization via self-supervised learning. First, we identify a critical vulnerability in vision-language models (VLMs): their susceptibility to adversarial manipulation of inference efficiency. We propose a black-box evaluation framework to assess this efficiency robustness, establishing a new dimension for VLM safety. Second, we find that VLMs exhibit limited adaptability to evolving data streams. We explore long-tailed class incremental learning within the VLM framework and introduce a method to address the dual challenges of catastrophic forgetting and data imbalance. Third, we examine vision models’ failures on out-of-distribution samples. We tackle the open-set domain adaptation challenge, where both distribution shift and unknown classes appear in the target domain, and develop a two-stage framework that adapts the model to classify known classes while detecting unknown ones. Fourth, we revisit self-supervised pretraining in computer vision and design a novel objective to enhance the generalizability of pre-trained models. Collectively, these contributions form a unified framework for advancing robustness, adaptation, and generalization, facilitating more trustworthy visual and multimodal intelligence.", "source_url": "https:\/\/openalex.org\/W7135015825", "pdf_url": "", "doi": "10.14711\/thesis-hdl169527", "handle": "", "repository_name": "", "source_record_id": "W7135015825", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:52", "updated_at": "2026-11-15 16:52:52" }, { "id": 777, "title": "GenVRAD: dataset generator for machine learning detection of vehicle-roadside attacks : a thesis in Computer Engineering", "title_ar": "", "title_en": "GenVRAD: dataset generator for machine learning detection of vehicle-roadside attacks : a thesis in Computer Engineering", "author": "Devaj Ramsamooj", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "", "university_source": "", "college": "", "department": "", "country": "", "governorate": "", "publication_year": 2024, "language": "en", "research_domain": "generative-ai", "research_subdomain": "fairness", "categories": "machine-learning; deep-learning; generative-ai; robotics; responsible-ai; ai-applications; fairness; ai-cybersecurity; ai-smart-cities", "keywords": "", "abstract": "Autonomous vehicles are gaining popularity rapidly for transportation safety and efficiency. However, to achieve fully automated driving without human intervention, cars need to be aware of traffic and environmental dynamics. One key component of their awareness is through communication. Connected and Automated Vehicles (CAV) leverage both automation and communication technologies to obtain information with nearby devices for decision making. Vehicular Ad-Hoc NETworks (VANETs) provide an architecture to support CAV with cars’ On-Board Units (OBUs) to communicate with each other and other devices like Roadside Units (RSUs). This communication can involve sharing information about traffic conditions, road hazards, and other data. Security of RSUs is crucial to ensure reliability and trustworthiness of its communication. Some security concerns with RSUs are authentication of RSUs, data integrity, etc. Addressing these concerns will require multiple types of mechanisms such as encryption and intrusion detection systems (IDS) to prevent misbehaving RSUs. With any type of communication, security of information is always at the forefront. Conventional security techniques have been tested, such as password protection and biometric security, but they do not meet the needs of the high dynamics of VANET. Machine and Deep have the ability to learn without human intervention, which is highly desirable to CAV. Since CAV is still new, there is little data to train and test Machine and Deep learning algorithms. Some attempts to synthesize attack datasets but only on vehicle-to-vehicle (V2V) communication. To the best of our knowledge, there is no data generator for vehicle-to-infrastructure (V2I). This research fills the gap by providing a dataset generator to inject misbehaving RSUs. Using simulation, data is generated to be able train and test several algorithms. Our work reveals machine learning algorithms are not sufficient to solve VANET security problems. The deep learning algorithm shows promise, but more analysis will be needed to be suitable for VANET security.", "source_url": "https:\/\/openalex.org\/W4409785539", "pdf_url": "", "doi": "10.62791\/20351", "handle": "", "repository_name": "", "source_record_id": "W4409785539", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:53", "updated_at": "2026-11-15 16:52:53" }, { "id": 778, "title": "Contrastive Temporal Explanations: A Novel Explanation Method for Machine Learning-Based Fault Prediction. Master Thesis University Utrecht", "title_ar": "", "title_en": "Contrastive Temporal Explanations: A Novel Explanation Method for Machine Learning-Based Fault Prediction. Master Thesis University Utrecht", "author": "Joris Graff", "supervisor": "", "degree_type": "master", "degree_evidence": "نص السجل: «Master»", "university": "", "university_source": "", "college": "", "department": "", "country": "Netherlands", "governorate": "", "publication_year": 2020, "language": "en", "research_domain": "medical-ai", "research_subdomain": "xai", "categories": "machine-learning; medical-ai; responsible-ai; ai-applications; self-semi-supervised; predictive-analytics; xai; fairness; ai-industry", "keywords": "", "abstract": "Digital twinning has enabled the real-time usage of sensor readings from a variety of systems, such as industrial machines and medical patients.These time-series data offer opportunities to accurately predict the system's future behaviour, e.g. by predicting whether faults are likely to occur within the near future on the basis of the most recent sensor data.Machine learning (ML) algorithms have shown an impressive accuracy in such fault prediction tasks, but suffer from a lack of explainability.This situation problematises the application of such algorithms to real-life problems, where user insight into the prediction model is generally a desideratum.This thesis aims to improve the explainability, and thereby trustworthiness, of fault prediction ML algorithms.In order to have a solution that addresses explainability, the main criteria for successful explanation algorithms have to be identified.A first challenge of this thesis is therefore the formulation of these citeria in a way that answers to theoretical and practical aspects of explainability.A further challenge is to meet these criteria within the domain of fault prediction.The result is a novel solution, called Contrastive Temporal Explanations (CTE), which is shown to outperform state-of-the-art methods on most of the key criteria.First of all, I wish to thank both of my supervisors, Paolo Pileggi from TNO and Hakim Qahtan from the University of Utrecht.Paolo has tirelessly assisted me during the duration of the internship of which this thesis is a product, even when the intervention of a pandemic made it impossible to meet in person.His many phone calls were filled with both encouragement and useful comments.Without his enthusiasm and insights this thesis would not have been possible in the current form.Hakim has helped give structure to the initial literature review stage of this project, and has given me many invaluable suggestions on the content, structure and stylistic aspects of the end product.Without his incidental adjustments I would have repeatedly steered decidedly off course.I also wish to thank Ron Snijders of TNO Groningen for commenting on earlier versions of parts of this thesis, and Sietse Drost and Lars-Uno Axelsson for providing the problem and the data that initially inspired this project, as well as for receiving me and other TNO interns and employees for a tour of the OPRA headquarters.Finally, I am indebted to the researchers, coordinators and administrative and HR officers at the ICT department of TNO The Hague, who have all been crucial in the success of my internship, both in their helpfulness in administrative issues and in creating a welcoming and intellectually stimulating environment that has helped shape many of the ideas presented in this thesis.", "source_url": "https:\/\/openalex.org\/W7132381722", "pdf_url": "https:\/\/publications.tno.nl\/publication\/34637482\/Dhw4Ki\/graff-2020-contrastive.pdf", "doi": "", "handle": "", "repository_name": "TNO Repository", "source_record_id": "W7132381722", "other_sources": "", "verification_status": "verified", "created_at": "2026-11-15 16:52:53", "updated_at": "2026-11-15 16:52:53" }, { "id": 781, "title": "Securing Federated Learning: A Comprehensive Defence Against Privacy Attacks", "title_ar": "", "title_en": "Securing Federated Learning: A Comprehensive Defence Against Privacy Attacks", "author": "Moyeen, M A", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "Concordia University", "university_source": "repository_host", "college": "", "department": "", "country": "Canada", "governorate": "", "publication_year": 2025, "language": "en", "research_domain": "medical-ai", "research_subdomain": "federated-learning", "categories": "machine-learning; medical-ai; reinforcement-learning; responsible-ai; ai-applications; federated-learning; fairness; ai-cybersecurity; ai-industry", "keywords": "", "abstract": "In this information age, machine learning (ML) applications drive smart living through innovations such as personalized healthcare, intelligent transportation, and smart homes. However, despite these advancements, businesses and industries continue to face significant challenges in safeguarding data privacy, as ML systems increasingly rely on vast amounts of user data. In this direction, Federated Learning (FL) has emerged as a promising solution, enabling collaborative model training while keeping user data on local premises without the need to share raw data. However, FL faces significant challenges also, including expensive communication costs, system heterogeneity, and vulnerability to various attacks. In particular, it is susceptible to poisoning attacks, where malicious participants corrupt models and data as well as inference attacks that exploit gradients to reveal sensitive information through membership inference or model inversion techniques. These attacks can extract sensitive information from shared gradients, undermining the fundamental privacy guarantees of federated systems. Thus, effective defence mechanisms are fundamental for fully leveraging the advantages of FL. Numerous defences, such as FoolsGold, Flod, Flad, MADDPG, and others, are in place to secure the FL systems. However, the majority of the defence mechanisms suffer from accuracy degradation, computational overhead, and inadequate attack prevention. Most client selection methods cannot reliably separate malicious and straggler clients, with even cutting-edge approaches struggling with herding and cold-start issues. Furthermore, recent state-of-the-art techniques frequently fail to defend against inference attacks adequately. These methods typically employ Secure Multiparty Computation (SMPC), Homomorphic Encryption (HE), or Differential Privacy (DP) as defensive measures. However, SMPC and HE suffer from high computational complexity, while DP often leads to degraded model accuracy. Thus, this research addresses these limitations by proposing five different defence mechanisms that ensure robust FL with protected gradients. The proposed mechanisms consist of FedChallenger, Fed-Reputed, SignDefence, Ada-Sign, and SignMPC. The proposed FedChallenger introduces a dual-layer defence mechanism that comprises the zero-trust challenge-response-based authentication at the first layer and a variant of Trimmed-Mean aggregation at the second layer that leverages pairwise cosine similarity and Median Absolute Deviation (MAD). Extensive evaluation on MNIST, FMNIST, EMNIST, and CIFAR-10 datasets demonstrates 3-10% accuracy improvement over state-of-the-art approaches with 1.1-2.2 times faster convergence and 2-3% higher F1-scores. Subsequently, the reputation-based client selection approach, Fed-Reputed, leverages device capability information and a modified Bellman equation within a hierarchical framework, integrated into a Deep Q-Learning Network (DQN)-based Imbalanced Classification Markov Decision Process (ICMDP) classifier for enhanced client selection. Testing on MNIST and FMNIST datasets demonstrates 9-50% accuracy gains and 1.3-1.7 times faster convergence while effectively detecting both malicious and straggler clients. Moreover, existing methods often suffer from the dying ReLU problem, where neurons permanently deactivate during training. To counter the dying ReLU problem, SignDefence implements a sophisticated aggregation scheme that utilizes sign direction and LeakyReLU-based aggregation, incorporating Jaccard similarity derived from binary-encoded model weights. This technique demonstrates consistent accuracy and F1-score improvements across different attack conditions. Despite its benefits, SignDefence remains vulnerable to inference attacks and suffers from limited generalization due to its fixed threshold across diverse benchmark datasets. To address these limitations, a lightweight strategy, Ada-Sign, employs adaptive threshold computation and incorporates DP mechanisms. This approach maintains comparable accuracy to SignDefence while providing enhanced gradient protection through adaptive DP. Extensive evaluation on MNIST and HAR datasets reveals 3-20% accuracy improvement for Ada-Sign over the majority of state-of-the-art techniques. Finally, to enhance the protection of both SignDefence and Ada-Sign against inference attacks, iv SignMPC integrates highly configurable SMPC, HE, and DP algorithms. This combined approach ensures comprehensive communication security and gradient privacy while avoiding significant performance bottlenecks. Comprehensive evaluation on MNIST and HAR datasets demonstrates 4-17% accuracy gains for SignMPC over established approaches while maintaining computational efficiency and robust privacy guarantees.", "source_url": "https:\/\/openalex.org\/W7112513152", "pdf_url": "", "doi": "", "handle": "", "repository_name": "Spectrum Research Repository (Concordia University)", "source_record_id": "W7112513152", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:53", "updated_at": "2026-11-15 16:52:53" }, { "id": 785, "title": "Federated Learning Platforms for Dynamic and Value-Driven Participation in Privacy-Preserving Machine Learning", "title_ar": "", "title_en": "Federated Learning Platforms for Dynamic and Value-Driven Participation in Privacy-Preserving Machine Learning", "author": "Christoph Düsing", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "Universität Bielefeld", "university_source": "repository_host", "college": "", "department": "", "country": "Germany", "governorate": "", "publication_year": 2025, "language": "en", "research_domain": "medical-ai", "research_subdomain": "federated-learning", "categories": "machine-learning; medical-ai; responsible-ai; predictive-analytics; xai; federated-learning; fairness", "keywords": "", "abstract": "In recent years, Federated Learning (FL) has emerged as a powerful solution for privacy-preserving machine learning, enabling multiple parties - so-called clients - to collaboratively train models without directly sharing their data. This paradigm is particularly relevant in data-sensitive domains such as healthcare and finance. Yet, traditional FL frameworks typically rely on a fixed set of participants and lack mechanisms to adapt to dynamic client participation. This thesis addresses this limitation by proposing the concept of Federated Learning Platforms (FLPs): A novel, platform-oriented framework designed to support the onboarding and offboarding of clients over time, thereby aligning more closely with real-world collaboration dynamics and strategic (business) objectives. To this end, the framework comprises modular components for technical infrastructure and platform governance that enable scalable model training, secure communication, incentive alignment, and long-term federation stability. A central contribution of this work is the development of an explainable input control mechanism that serves as a gate-keeper to the FLP. It evaluates prospective clients based on privacy-preserving data imbalance metrics. Through extensive empirical studies on multiple datasets, we uncover a systemic trade-off between clients' contribution and benefit in FL, where highly imbalanced clients often contribute most to the model but gain the least in return. The proposed input control mechanism addresses this tension by making client admission both data-driven and transparent. In addition, we systematically investigate the role of data sampling and client selection in mitigating the detrimental effect of data imbalance - one of the most pressing challenges in FL. We benchmark existing strategies for local data sampling, introduce a novel optimization-based sampling approach, and propose a distribution-aware client selection algorithm. Together, these contributions effectively manage local and global imbalances, as well as other unfavorable data properties, leading to improved model convergence, predictive performance, and overall training time. Finally, we explore the impact of concept drift in dynamic FLP environments, where the landscape of participating clients and data distributions evolves over time. Therefore, we evaluate both error-based and data-based drift detection methods with respect to their timeliness of intervention and propose a nuanced perspective that distinguishes between harmful and beneficial drift. This enables the development of adaptive strategies that preserve model quality while leveraging potential performance improvements. Together, these contributions lay the groundwork for FLPs - scalable, adaptive, and trustworthy systems that transform static FL into an open, continuously evolving digital platform for privacy‑critical machine learning.", "source_url": "https:\/\/openalex.org\/W7108321425", "pdf_url": "", "doi": "10.4119\/unibi\/3012109", "handle": "", "repository_name": "Publikationen an der Universität Bielefeld (Universität Bielefeld)", "source_record_id": "W7108321425", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:53", "updated_at": "2026-11-15 16:52:53" }, { "id": 787, "title": "Federated Learning of Fuzzy Rule-based Systems: Advances in Privacy, Security and Interpretability", "title_ar": "", "title_en": "Federated Learning of Fuzzy Rule-based Systems: Advances in Privacy, Security and Interpretability", "author": "Mattia Daole", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "University of Pisa", "university_source": "repository_host", "college": "", "department": "", "country": "Italy", "governorate": "", "publication_year": 2026, "language": "en", "research_domain": "responsible-ai", "research_subdomain": "federated-learning", "categories": "machine-learning; responsible-ai; expert-systems; ai-applications; predictive-analytics; optimization; xai; federated-learning; fairness; ai-cybersecurity", "keywords": "", "abstract": "This thesis addresses the growing need for Trustworthy AI, focusing on the transparency and privacy requirements. Transparency is commonly addressed through eXplainable AI (XAI), while privacy-preserving paradigms such as Federated Learning (FL) enable collaborative model training without sharing sensitive data. The work extends traditional Fuzzy Rule-Based Classifiers (FRBCs) to federated settings, systematically analyzing their learning behavior and performance under heterogeneous and non-IID data distributions. To further improve the trade-off between predictive accuracy and interpretability, a multi-objective federated framework based on Evolutionary Fuzzy Systems is introduced, using synthetic data generation techniques to support privacy-preserving server-side optimization. In addition to model design, the thesis investigates security aspects by analyzing classical adversarial threats in FL, such as data and model poisoning, and by proposing a taxonomy of attacks specifically targeting federated FRBCs. Finally, an edge-oriented application framework compliant with Multi-Access Edge Computing architectures is presented, demonstrating the practical feasibility and effectiveness of deploying trustworthy, transparent, and privacy-aware AI systems in real-world distributed environments.", "source_url": "https:\/\/openalex.org\/W7140251669", "pdf_url": "", "doi": "", "handle": "", "repository_name": "Electronic Theses and Dissertations Repository (University of Pisa)", "source_record_id": "W7140251669", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:53", "updated_at": "2026-11-15 16:52:53" }, { "id": 789, "title": "Secure and Privacy-preserving Federated Learning with Explainable Artificial Intelligence for Smart Healthcare System", "title_ar": "", "title_en": "Secure and Privacy-preserving Federated Learning with Explainable Artificial Intelligence for Smart Healthcare System", "author": "Ali Raza", "supervisor": "", "degree_type": "phd", "degree_evidence": "theses.fr: الفهرس الوطني الفرنسي لأطاريح الدكتوراه", "university": "", "university_source": "", "college": "", "department": "", "country": "France", "governorate": "", "publication_year": 2023, "language": "en", "research_domain": "medical-ai", "research_subdomain": "federated-learning", "categories": "machine-learning; deep-learning; medical-ai; responsible-ai; ai-applications; xai; federated-learning; fairness; ai-cybersecurity", "keywords": "", "abstract": "The growing population around the globe has a significant impact on various sectors including the labor force, healthcare, and the global economy. The healthcare sector is among the most affected sectors by the growing population due to the increasing demand for resources including doctors, nurses, equipment, and healthcare facilities. Intelligent systems have been incorporated to enhance decision-making, management, prognosis, and diagnosis in order to tackle such issues and offer improved healthcare to patients. Among such systems, those based on deep learning (DL), a subclass of machine learning (ML) have outperformed many traditional statistical and ML systems owing to their capability of automatically discovering and learning related features for a given task and robustness. Therefore, the use of DL has seen a steady increase in many applications. Nevertheless, usually, the training of DL models relies on a single centralized server, which brings many challenges: (1) except for some big enterprises most of the small enterprises have limited quality data, which is insufficient to support the training of data-hungry DL models, (2) access to data, which is vital for these systems, often raises privacy concerns. The collection and analysis of sensitive patient information must be done in a secure and ethical manner to ensure the protection of individual privacy rights, (3) high communication cost and computation resources required, (4) a large number of trainable parameters make the outcome of DL hard to explain, which is required in some applications, such as healthcare. Compared to centralized ML, federated learning (FL) improves both privacy and communication costs, where clients collaboratively train a joint model without sharing the raw data directly. FL minimizes privacy breaches and safeguards sensitive data by keeping it distributed locally. This enables collaborative model training while reducing the risk of unauthorized access and data breaches. Additionally, it promotes data diversity and scalability by involving multiple sources in joint model training and decreases communication costs by sharing only model updates instead of the entire dataset. However, FL brings its own challenges. For example, heterogeneous local data among the clients makes it challenging to train a high-performing and robust global model. Sharing updates (hundreds of thousands of parameters) still has high communication costs. Additionally, the distributed nature and access control of local data in FL make it more vulnerable to malicious attacks. Moreover, the challenge of explaining the results of DL still remains challenging, and methods are needed to be developed to bring trust accountability, and transparency in sensitive applications, such as healthcare. Therefore, the aim of this thesis is to create robust frameworks that are secure, high-performing, and privacy-friendly within federated settings. These frameworks will be specifically designed for end-to-end (we train our frameworks using raw data without any manual feature extraction) healthcare applications, considering the presence of non-identically distributed data among clients in FL to bring robustness. By addressing these challenges, the objective is to enhance the overall system's resilience and effectiveness. We also propose a methodology for detecting anomalies within federated settings, particularly in applications with limited available data for the abnormal class. Furthermore, clients in FL are usually resource-constrained with limited computation and communication resources available. Therefore, to support efficient computation and communication in a federated setting we propose a lightweight framework (in terms of the trainable number of parameters). Additionally, to provide explanations of the DL models' outcomes, which are usually hard to explain because of the large number of parameters, we propose model-agnostic explainable AI modules to help explain the results of DL models. Moreover, in order to protect the proposed frameworks against cyber attacks, such as poisoning attacks, we propose a framework in federated settings, which makes the proposed healthcare frameworks more secure and trustworthy. Finally, with experimental analysis using baseline datasets for one of the most common health conditions i.e., cardiovascular diseases (arrhythmia detection, ECG anomaly detection) and human activity recognition (used for supplementing cardiovascular diseases detection), we show the superiority of the proposed frameworks over state-of-the-art work.", "source_url": "https:\/\/openalex.org\/W4399027046", "pdf_url": "http:\/\/www.theses.fr\/2023ULILB019\/document", "doi": "10.22024\/unikent\/01.02.102717", "handle": "", "repository_name": "theses.fr (ABES)", "source_record_id": "W4399027046", "other_sources": "", "verification_status": "verified", "created_at": "2026-11-15 16:52:53", "updated_at": "2026-11-15 16:52:53" }, { "id": 887, "title": "Interpretability of a Deep Learning Model for Semantic Segmentation : Example of Remote Sensing Application", "title_ar": "", "title_en": "Interpretability of a Deep Learning Model for Semantic Segmentation : Example of Remote Sensing Application", "author": "Adrianna Janik", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "KTH Royal Institute of Technology", "university_source": "repository_host", "college": "", "department": "", "country": "Sweden", "governorate": "", "publication_year": 2019, "language": "en", "research_domain": "computer-vision", "research_subdomain": "image-segmentation", "categories": "machine-learning; deep-learning; computer-vision; responsible-ai; ai-applications; predictive-analytics; image-segmentation; xai; fairness; ai-smart-cities", "keywords": "", "abstract": "Understanding a black-box model is a major problem in domains that relies on model predictions in critical tasks. If solved, can help to evaluate the trustworthiness of a model. This thesis proposes a user-centric approach to black-box interpretability. It addresses the problem in semantic segmentation setting with an example of humanitarian remote sensing application for building detection. The question that drives this work was, Can existing methods for explaining black-box classifiers be used for a deep learning semantic segmentation model? We approached this problem with exploratory qualitative research involving a case study and human evaluation. The study showed that it is possible to explain a segmentation model with adapted methods for classifiers but not without a cost. The specificity of the model is likely to be lost in the process. The sole process could include introducing artificial classes or fragmenting image into super-pixels. Other approaches are necessary to mitigate identified drawback. The main contribution of this work is an interactive visualisation approach for exploring learned latent space via a deep segmenter, named U-Net, evaluated with a user study involving 45 respondents. We developed an artefact (accessible online) to evaluate the approach with the survey. It presents an example of this approach with a real-world satellite image dataset. In the evaluation study, the majority of users had a computer science background (80%), including a large percentage of users with machine learning specialisation (44.4% of all respondents). The model distinguishes rurality vs urbanization (58% of users). External quantitative comparison of building densities of each city concerning the location in the latent space confirmed the later. The representation of the model was found faithful to the underlying model (62% of users). Preliminary results show the utility of the pursued approach in the application domain. Limited possibility to present complex model visually requires further investigation.", "source_url": "https:\/\/openalex.org\/W3011114404", "pdf_url": "", "doi": "", "handle": "", "repository_name": "KTH Publication Database DiVA (KTH Royal Institute of Technology)", "source_record_id": "W3011114404", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:54", "updated_at": "2026-11-15 16:52:54" }, { "id": 922, "title": "Investigating New Zealanders' Attitudes, Behaviours and Intentions toward Cybersecurity, Privacy, Authentication and Artificial Intelligence (AI)", "title_ar": "", "title_en": "Investigating New Zealanders' Attitudes, Behaviours and Intentions toward Cybersecurity, Privacy, Authentication and Artificial Intelligence (AI)", "author": "Lisa Patterson", "supervisor": "", "degree_type": "phd", "degree_evidence": "نص السجل: «PhD»", "university": "", "university_source": "", "college": "", "department": "", "country": "New Zealand", "governorate": "", "publication_year": 2025, "language": "en", "research_domain": "medical-ai", "research_subdomain": "fairness", "categories": "machine-learning; medical-ai; responsible-ai; ai-applications; predictive-analytics; fairness; ai-cybersecurity", "keywords": "", "abstract": "This PhD thesis presents a comprehensive investigation into the attitudes, behaviours, and security intentions of general population New Zealanders toward a variety of cybersecurity risks. It culminates in the development of a novel empirically derived model for understanding cybersecurity behaviour in the NZ context. Through systematic research progressing from qualitative exploration to quantitative validation, this thesis addresses critical gaps in understanding how New Zealanders conceptualise and respond to risks relating to cybersecurity, privacy, authentication and Artificial Intelligence (AI). The research began with a pilot study examining Internet of Things (IoT) cybersecurity behaviours, marking the first academic application of Protection Motivation Theory (PMT) to general population New Zealanders in this context. This initial phase revealed important patterns in how people evaluate security risks, particularly highlighting the roles of awareness, trust, and influence. A subsequent confirmatory study substantially validated these findings while expanding understanding through examination of remote working contexts during the COVID-19 pandemic, demonstrating both the applicability and limitations of existing theoretical frameworks in the NZ environment. Building on these qualitative insights, a nationally representative online survey was designed, focusing on four areas of vital importance in a contemporary Aotearoa New Zealand (NZ) context: Privacy, Authentication, AI and General Cybersecurity. Results from this survey directly informed construction of the PSCYBR Model, which attempted to extend PMT by incorporating Social Influence Theory (SIT) constructs. While this theory driven approach showed promise when restructured, empirical testing revealed the need for a more fundamental re-evaluation of how New Zealanders approach cybersecurity-related risks. During the development process, additional theoretical frameworks were examined but ultimately discounted as less suitable for explaining NZ cybersecurity behaviours compared to the empirically-derived approach. Through rigorous exploratory analysis, the DPER (Data, Password, Effort and Risk) Model was developed. The DPER Model, emerging directly from analysis of a nationally representative sample of 1,006 New Zealanders, identifies three distinct factors that shape cybersecurity behaviour: Protection of Data, Importance of Password Management, and Effort and Risk. These empirically derived factors reflect actual behavioural patterns rather than theoretical constructs, providing valuable insights for developing targeted interventions and support strategies. The model validation employed regression analysis, with findings showing that the Protection of Data factor significantly predicts variance in trust outcomes, while Importance of Password Management, and Effort and Risk, factors demonstrate weaker but meaningful associations. The model’s validation through robust statistical testing demonstrates its reliability and utility for understanding attitudes, behaviours, and intentions toward risks relating to cybersecurity in a modern Aotearoa NZ context. The research reveals unique characteristics in how New Zealanders evaluate and respond to cybersecurity-related risks, influenced by cultural values, social structures, and practical considerations specific to the NZ environment. The findings highlight the importance of developing culturally appropriate security initiatives that resonate with diverse communities, including Māori and Pasifika populations. The thesis makes substantial contributions to both theoretical understanding and practical application. Organisations working to improve cybersecurity in NZ will have access to a model that reflects how general population New Zealanders actually think about and respond to cybersecurity-related risks. The research provides evidence based guidance for developing effective security awareness programs and support strategies tailored to the NZ context. Future research directions emerging from this work include longitudinal studies to track behavioural changes over time, detailed investigation of cultural adaptations, and examination of the model’s applicability to emerging technologies such as artificial intelligence. The thesis concludes by discussing practical implications for improving cybersecurity posture across NZ society, emphasising the importance of considering both individual and community factors in cybersecurity initiatives. This research represents a significant advancement in understanding cybersecurity behaviour among general population New Zealanders, helping to enhance New Zealand’s cybersecurity posture. The empirically-derived DPER Model offers a robust framework for developing targeted interventions that align with how New Zealanders actually think about and respond to cybersecurity challenges.", "source_url": "https:\/\/openalex.org\/W4411463037", "pdf_url": "https:\/\/openaccess.wgtn.ac.nz\/articles\/thesis\/Investigating_New_Zealanders_Attitudes_Behaviours_and_Intentions_toward_Cybersecurity_Privacy_Authentication_and_Artificial_Intelligence_AI_\/29363615\/1\/files\/55491929.pdf", "doi": "10.26686\/wgtn.29363615", "handle": "", "repository_name": "Open Access Victoria University of Wellington | Te Herenga Waka (Figshare)", "source_record_id": "W4411463037", "other_sources": "", "verification_status": "verified", "created_at": "2026-11-15 16:52:54", "updated_at": "2026-11-15 16:52:54" }, { "id": 932, "title": "Blockchain for AI: Smarter Contracts to Secure Artificial Intelligence Algorithms", "title_ar": "", "title_en": "Blockchain for AI: Smarter Contracts to Secure Artificial Intelligence Algorithms", "author": "Syed Badruddoja", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "", "university_source": "", "college": "", "department": "", "country": "United States", "governorate": "", "publication_year": 2023, "language": "en", "research_domain": "nlp", "research_subdomain": "llm", "categories": "machine-learning; deep-learning; nlp; predictive-analytics; llm", "keywords": "", "abstract": "In this dissertation, I investigate the existing smart contract problems that limit cognitive abilities. I use Taylor's serious expansion, polynomial equation, and fraction-based computations to overcome the limitations of calculations in smart contracts. To prove the hypothesis, I use these mathematical models to compute complex operations of naive Bayes, linear regression, decision trees, and neural network algorithms on Ethereum public test networks. The smart contracts achieve 95\\% prediction accuracy compared to traditional programming language models, proving the soundness of the numerical derivations. Many non-real-time applications can use our solution for trusted and secure prediction services.", "source_url": "https:\/\/openalex.org\/W4389378466", "pdf_url": "https:\/\/digital.library.unt.edu\/ark:\/67531\/metadc2179338\/m2\/1\/high_res_d\/BADRUDDOJA-DISSERTATION-2023.pdf", "doi": "10.12794\/metadc2179338", "handle": "", "repository_name": "University of North Texas Digital Library (University of North Texas)", "source_record_id": "W4389378466", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:54", "updated_at": "2026-11-15 16:52:54" }, { "id": 933, "title": "On the Safety Alignment of Large Language Models", "title_ar": "", "title_en": "On the Safety Alignment of Large Language Models", "author": "Mengxuan Hu", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "", "university_source": "", "college": "", "department": "", "country": "United States", "governorate": "", "publication_year": 2026, "language": "en", "research_domain": "nlp", "research_subdomain": "llm", "categories": "machine-learning; nlp; medical-ai; responsible-ai; ai-applications; transfer-learning; llm; fairness; ai-education; ai-cybersecurity", "keywords": "", "abstract": "Large language models (LLMs) are increasingly deployed in high-stakes settings such as finance, healthcare, and education, where failures in safety and fairness can cause real harm. In this dissertation, we advance LLM trustworthiness by (i) benchmarking how LLMs introduce explicit risks within the RAG pipeline, (ii) developing red-teaming attacks that expose latent vulnerabilities even under seemingly “safe” conditions, and (iii) proposing robust defenses that strengthen alignment in practice. First, we study Retrieval-Augmented Generation (RAG), a widely used strategy to reduce hallucinations and improve domain grounding, and show that its benefits are not a free lunch. We propose a practical three-level threat model based on user awareness of fairness, which induces different degrees of “fairness censorship” in external dataset. Across uncensored, partially censored, and fully censored datasets, we find that fairness alignment can be undermined through RAG alone, without any fine-tuning, and that biased outputs can persist even when the external data is ostensibly unbiased. These results reveal fundamental limitations of current alignment when models are coupled with retrieval. Second, we uncover a complementary vulnerability in the fine-tuning stage: even benign training data can degrade safety. Building on this phenomenon, we develop a stronger red-teaming attack by identifying the most safety-degrading outliers within benign datasets. We introduce Self-Inf-N, an outlier-detection method that selects a small set of influential samples. Fine-tuning on as few as 100 selected outliers can severely compromise safety alignment across seven mainstream LLMs, with high transferability and poor coverage by existing mitigations. Third, motivated by these failures, we develop reasoning-aware alignment defenses. Using causal intervention, we show that many jailbreak vulnerabilities stem from shallow alignment behaviors that refuse without deeply understanding harmful intent. We address this by constructing a new Chain-of-Thought fine-tuning dataset spanning utility and safety-critical prompts to encourage principled, reasoned refusals. Finally, we propose Alignment-Weighted DPO, which assigns different preference weights to reasoning versus response segments, enabling targeted updates that improve jailbreak robustness while preserving utility. Together, these studies provide a unified framework for diagnosing and mitigating trustworthiness failures across RAG, fine-tuning, offering actionable guidance for deploying safer and fairer LLM systems.", "source_url": "https:\/\/openalex.org\/W7151325055", "pdf_url": "", "doi": "10.18130\/3vqg-ng96", "handle": "", "repository_name": "Libra", "source_record_id": "W7151325055", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:54", "updated_at": "2026-11-15 16:52:54" }, { "id": 961, "title": "Uncertainty in large language models", "title_ar": "", "title_en": "Uncertainty in large language models", "author": "Lukas Aichberger", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "Johannes Kepler Universitat Linz", "university_source": "repository_host", "college": "", "department": "", "country": "Austria", "governorate": "", "publication_year": 2026, "language": "en", "research_domain": "nlp", "research_subdomain": "llm", "categories": "machine-learning; deep-learning; computer-vision; nlp; predictive-analytics; multimodal; llm", "keywords": "", "abstract": "Large language models (LLMs) are increasingly serving as a primary interface for accessing information and interacting with digital systems. Nonetheless, LLMs can generate text that includes incorrect or unsupported claims while being fluent, coherent, and presented as factual. This creates a fundamental reliability issue: the generated text alone often does not reveal whether it can be trusted. Such unsupported factual claims are commonly referred to as LLM hallucinations. Detecting hallucinations is essential for deciding when a generated text can be trusted and when the system should instead abstain from answering, retrieve further evidence, or defer to a human. A principled approach to hallucination detection is uncertainty quantification, as elevated uncertainty is associated with a higher risk of hallucinations. However, estimating uncertainty in LLMs is challenging because uncertainty depends on the model’s predictive distribution over possible output sequences. Since LLMs stochastically generate output sequences token by token, this predictive distribution is high-dimensional and generally intractable to compute exactly. Thus, uncertainty estimation must rely on tractable approximations of this high-dimensional distribution. Developing effective approximations is a key enabler for the reliable deployment of LLM-based systems. In this cumulative thesis, we address the challenges posed by high-dimensional distributions to advance uncertainty estimation. We first establish foundational principles in the context of classical classification: our method QUAM demonstrates that uncertainty estimation benefits from targeted exploration, explicitly searching for samples that most influence the estimate. We then translate this insight to LLMs through SDLG, a method that improves uncertainty estimation by sampling output sequences that are most informative about the underlying distribution. Furthermore, we show that competitive uncertainty estimates can also be obtained without sampling multiple output sequences by leveraging a single-sequence measure: the negative log-likelihood of the most likely output sequence, approximated through our method G-NLL. This provides a highly efficient and practical alternative to sampling-based uncertainty estimation in LLMs. While this thesis primarily focuses on the internal reliability of LLMs through uncertainty estimation, it also investigates the external reliability of LLM-based systems deployed as multimodal agents that observe and act within their environment. To this end, we introduce Malicious Image Patches (MIPs), demonstrating that multimodal agents can be compromised into executing malicious actions when such patches appear in their visual observations. By exposing this vulnerability, this work highlights the need for safeguards against external failure modes and broadens the thesis from the internal to the external reliability of systems built around LLMs. Overall, the core contributions of this thesis advance uncertainty estimation in LLMs and thereby improve uncertainty-based hallucination detection. Complemented by insights into malicious failure modes, these contributions provide a principled foundation for improving the internal and external reliability of LLM-based systems.", "source_url": "https:\/\/openalex.org\/W7169636052", "pdf_url": "", "doi": "", "handle": "", "repository_name": "University Library Linz repository (Johannes Kepler Universitat Linz)", "source_record_id": "W7169636052", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:55", "updated_at": "2026-11-15 16:52:55" }, { "id": 972, "title": "Quality Assurance in Large Language Models", "title_ar": "", "title_en": "Quality Assurance in Large Language Models", "author": "Song, Da", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "University of Alberta", "university_source": "repository_host", "college": "", "department": "", "country": "Canada", "governorate": "", "publication_year": 2025, "language": "en", "research_domain": "nlp", "research_subdomain": "llm", "categories": "machine-learning; nlp; medical-ai; responsible-ai; data-mining; predictive-analytics; llm; xai; fairness", "keywords": "", "abstract": "Large Language Models (LLMs) have demonstrated impressive capabilities across a wide range of tasks in natural language processing (NLP) and software engineering (SE). However, their increasing adoption has raised critical concerns about trustworthiness, including robustness under distribution shifts, vulnerability to hallucinations, and misalignment between confidence and accuracy. These issues are exacerbated by the complex and probabilistic nature of LLMs, which makes it difficult for developers to observe, interpret, and address their failures. To bridge this gap, this thesis presents a series of tools and frameworks that enable a more transparent and actionable understanding of LLM trustworthiness. Together, these works follow a methodological arc that moves from interactive exploration to internal structure modeling and finally to external behavioral correction. The first contribution, DeepLens, introduces a human-centered system that supports interactive diagnosis of out-of-distribution (OOD) data in large text corpora. By integrating clustering-based analysis and neuron-level highlighting, {\\deeplens} empowers developers to detect, categorize, and interpret OOD issues that threaten model reliability. Building on this foundation, the second contribution, LUNA, proposes a white-box framework that models LLM behavior using abstract probabilistic structures. By mapping hidden states to interpretable semantic concepts, LUNA enables structured evaluation of trustworthiness dimensions such as hallucination and adversarial robustness, providing insights grounded in both semantic and model-level perspectives. The final contribution addresses the limitations of white-box methods in black-box scenarios. It introduces a confidence calibration framework that quantifies and corrects model confidence based on semantic-preserving behavioral variations. By leveraging agreement ratios, entropy, and stability features from multiple LLM outputs, this method learns a regression-based correction factor to improve alignment between model confidence and factual correctness, without requiring model internals. Together, these three works provide a practical toolchain to enhance the transparency and trustworthiness of LLMs, spanning interactive user interfaces, interpretable model-based analysis, and scalable confidence calibration.", "source_url": "https:\/\/openalex.org\/W7105991970", "pdf_url": "", "doi": "10.7939\/83296", "handle": "", "repository_name": "ERA: Education and Research Archive (University of Alberta)", "source_record_id": "W7105991970", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:56", "updated_at": "2026-11-15 16:52:56" }, { "id": 981, "title": "On the capabilities and risks of large language models", "title_ar": "", "title_en": "On the capabilities and risks of large language models", "author": "Jie Huang", "supervisor": "", "degree_type": "phd", "degree_evidence": "نص السجل: «Ph.D»", "university": "University of Illinois at Urbana-Champaign", "university_source": "repository_host", "college": "", "department": "", "country": "United States", "governorate": "", "publication_year": 2024, "language": "en", "research_domain": "nlp", "research_subdomain": "llm", "categories": "nlp; generative-ai; responsible-ai; llm; fairness", "keywords": "", "abstract": "The advent of Large Language Models (LLMs) has significantly influenced the field of artificial intelligence, offering remarkable text generation capabilities through their vast number of parameters.These advancements have established new benchmarks across various domains.However, despite the impressive capabilities of LLMs, there exist critical limitations and ethical challenges.This dissertation critically examines the capabilities of LLMs, including their reasoning abilities, and explores potential risks, such as privacy leakage.Through this analysis, we underscore the crucial need to improve the capabilities of LLMs while mitigating the associated risks.Based on this understanding, we propose methodologies to augment and safeguard LLMs.To enhance their functionality, we develop techniques to integrate LLMs with external knowledge and design an innovative data structure for knowledge representation.Additionally, we advocate for incorporating citation mechanisms within LLMs to promote transparency, accountability, and respect for intellectual property.Through rigorous research and the introduction of cutting-edge techniques, this dissertation aims to advance the capabilities of LLMs while ensuring their responsible and ethical use, ultimately contributing to the development of powerful and trustworthy artificial intelligence systems.I am deeply grateful for the support and guidance I have received throughout my Ph.D. journey.First, I would like to express my sincere gratitude to my advisor, Kevin Chen-Chuan Chang.Together, we initiated and led the NLP research in our group.Kevin's support and guidance have been crucial in shaping", "source_url": "https:\/\/openalex.org\/W7220806755", "pdf_url": "https:\/\/www.ideals.illinois.edu\/items\/132635\/bitstreams\/439958\/data.pdf", "doi": "", "handle": "", "repository_name": "Illinois Digital Environment for Access to Learning and Scholarship (University of Illinois at Urbana-Champaign)", "source_record_id": "W7220806755", "other_sources": "", "verification_status": "verified", "created_at": "2026-11-15 16:52:56", "updated_at": "2026-11-15 16:52:56" }, { "id": 1011, "title": "Reconciling with Noise in Machine Learning for Healthcare", "title_ar": "", "title_en": "Reconciling with Noise in Machine Learning for Healthcare", "author": "Sujay Nagaraj", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "University of Toronto", "university_source": "repository_host", "college": "", "department": "", "country": "Canada", "governorate": "", "publication_year": 2025, "language": "en", "research_domain": "medical-ai", "research_subdomain": "fairness", "categories": "machine-learning; medical-ai; responsible-ai; ai-applications; predictive-analytics; fairness; ai-industry", "keywords": "", "abstract": "This thesis explores the challenges presented by label noise in machine learning (ML) for health. Label noise is unavoidable in real-world data; it can arise from variability in measurement, human annotation errors, and aleatoric uncertainty in the ground truth. Ignoring label noise when training classifiers with standard risk-minimization algorithms can result in faulty models that predict noise instead of signal; this is particularly problematic in healthcare where predictions can representmission-critical decisions. This thesis addresses the problem of label noise from several perspectives: exploring how and why label noise arises in health data, proposing robust algorithms to handle label noise, and studying the impact of label noise on individualized predictions while proposing methods for safer inference. Through a real-world study of stress detection in sensor data, the thesis first explores the significant inter- and intra-individual variability in label representations, undermining the reliability of classifiers trained on these labels. Motivated by this, the thesis introduces the concept of Temporal Label Noise, where noise levels vary over time, and proposes robust algorithms to handle this type of noise. It then examines the consequences of label noise on individual-level predictions, a gap in existing label noise research, while providing tools to mitigate the potential harm in high-stakes healthcare predictions. Finally, the work delves into a clinical case study in the Critical Care Unit, demonstrating how noisy sensor data can be leveraged to generate higher-quality annotations and improve patient safety initiatives. By identifying clinical challenges and proposing novel algorithmic solutions, this thesis intends to help advance the development of safe, robust, and trustworthy machine learning for healthcare.", "source_url": "https:\/\/openalex.org\/W7132904433", "pdf_url": "https:\/\/utoronto.scholaris.ca\/bitstreams\/9822ef19-c180-4bf4-96cb-b050f266a01b\/download", "doi": "", "handle": "", "repository_name": "TSpace (University of Toronto)", "source_record_id": "W7132904433", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:57", "updated_at": "2026-11-15 16:52:57" }, { "id": 1141, "title": "Data mining of classification for sybil user detection", "title_ar": "", "title_en": "Data mining of classification for sybil user detection", "author": "Anand Chinchore", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "University of Technology Sydney", "university_source": "repository_host", "college": "", "department": "", "country": "Australia", "governorate": "", "publication_year": 2016, "language": "en", "research_domain": "machine-learning", "research_subdomain": "data-mining", "categories": "machine-learning; ai-applications; data-mining; predictive-analytics; ai-cybersecurity", "keywords": "", "abstract": "Data analytics and Big Data application research, along with new structures in complex data, are reveling the secrets of their own complexity and patterns with valuable and critical, but challenging, issues through newly designed tools, techniques and models in data science technology. A common example concerns the interconnectivity of social network users on mobiles, involving content and information sharing through mobile social networks. There have been a large number of studies on mobile networks. Many focus on a variety of secured applications that attempt to exploit social connections, impersonate users or attack social groups. Such applications are often created with the intention of collecting confidential information, laundering money, blackmail or to perform other criminal activities. Existing methods for identifying such activity, such as distributed systems, social graph-based sybil detection, behaviour classification, and local ranking systems that estimate the trust level between users, rely on the dependencies between random nodes of connection on mobile social networks. These models aim to detect suspicious connections and have the advantage of learning the relationships between nodes and data. However, their detection patterns tend to impose the behavioural patterns typically associated with community-based and external networks. In data mining, the graph-based and classification models used for pattern collection can accurately predict patterns in data in targeted categories. Decision trees, commonly used for classification, are trees in which each branch represents a choice between a number of alternatives, and each leaf represents a classification, or decision. For example, a decision tree may help an institution decide whether a node in a dataset is suspicious, or considered to be sybil, if a decision tree can be induced from a set of data about its instances and the - classifications of those instances. It could also provide the flexibility to demonstrate data distribution. Thus, researchers have tried to combine different techniques and methods into network-based models to detect various patterns generated by sybil nodes within a network. The purpose of this thesis is to abridge existing classification and regression techniques to identify sybil nodes, and the correlation of those nodes with time, to address these research limitations. Classification and regression techniques predict behaviour based on continuous or categorical responses. For example if the predicted response is continuous, then it is called a regression tree. If the response is categorical, it is called a classification tree. At each node of the tree, the value of one the connected input nodes is checked and a binary answer – yes or no – determines whether one continues to the left or right sub-branch. When a leaf is reached, a prediction follows from a series of entropy calculations and graphing techniques. This thesis introduces a novel classification model for sybil detection in mobile social behaviour that identifies dependencies using connection duration and other attributes. Roger Quinlan’s C4.5 algorithm, its resulting decision tree and a random forest simplify the step-by-step identification process, while maintaining its merits. Partial correlations between nodes are simplified using Rattle programming, and the dataset is divided into majority nodes to assist processing. This research also includes a behavioural survey of the nodes and an extended analysis using a classification system for sybil detection, with a particular focus on sybil attacks in mobile social network environments. Each sybil node is tracked and identified based on the frequency and duration of its connections with other nodes. An outline of how the classified model identifies behaviour is also included, along with an explanation of the flow of the decision tree and the C4.5 algorithm process, which press-gangs identified sybil nodes based on the results of entropy calculations and information gain. The calculated entropy for each node connection across the all datasets informs the information gain. The maxGain calculations for individual node bring the final stage of draw decision tree and helped to predict the sybil nodes, compare and justify the sybil attackers. These processes and new models applied to sybil detection provide insight into the behaviour of connections, through deep analytics and entropy gain. The evidence gleaned from this research brings significant knowledge to data analytics and data science in the identification of threats on mobile social networks.", "source_url": "https:\/\/openalex.org\/W2604611620", "pdf_url": "", "doi": "", "handle": "", "repository_name": "UTS ePRESS (University of Technology Sydney)", "source_record_id": "W2604611620", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:58", "updated_at": "2026-11-15 16:52:58" }, { "id": 1161, "title": "Building Trust in Diagnostic Artificial Intelligence: An Explainable Approach to Medical Decision Support; Interpreting Trust: The Social Construction of IBM Watson for Oncology", "title_ar": "", "title_en": "Building Trust in Diagnostic Artificial Intelligence: An Explainable Approach to Medical Decision Support; Interpreting Trust: The Social Construction of IBM Watson for Oncology", "author": "Kenny Tran", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "", "university_source": "", "college": "", "department": "", "country": "United States", "governorate": "", "publication_year": 2026, "language": "en", "research_domain": "deep-learning", "research_subdomain": "transformers", "categories": "machine-learning; deep-learning; computer-vision; medical-ai; responsible-ai; predictive-analytics; transformers; xai", "keywords": "", "abstract": "Artificial Intelligence is increasingly used in clinical diagnostics; yet many systems fail to capture physician trust because their decision-making processes are unclear and difficult to interpret. To address this challenge, I propose the design of an explainable diagnostic AI system that balances predictive performance with clinically meaningful explanations. The proposed solution leverages transformer-based architectures augmented with weighting techniques and natural-language generation to communicate diagnostic reasoning in human-readable terms. Rather than relying solely on statistical feature visualizations, the system emphasizes explanations aligned with established clinical reasoning practices. Publicly available electronic health record data will be used to evaluate diagnostic accuracy and interpretability. The anticipated outcome is a diagnostic model that improves transparency without sacrificing performance, thereby supporting shared decision-making between clinicians and patients. Future work includes clinician-centered usability studies, refinement of explanation interfaces, and evaluation of the system’s effectiveness across different clinical settings.", "source_url": "https:\/\/openalex.org\/W7160598486", "pdf_url": "", "doi": "10.18130\/b87h-7367", "handle": "", "repository_name": "Libra", "source_record_id": "W7160598486", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:59", "updated_at": "2026-11-15 16:52:59" }, { "id": 1163, "title": "The Relevance of Explainable Artificial Intelligence (xAI) in High-Risk Decisions", "title_ar": "", "title_en": "The Relevance of Explainable Artificial Intelligence (xAI) in High-Risk Decisions", "author": "Julie Gerlings", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "Copenhagen Business School", "university_source": "repository_host", "college": "", "department": "", "country": "Germany", "governorate": "", "publication_year": 2025, "language": "en", "research_domain": "responsible-ai", "research_subdomain": "xai", "categories": "machine-learning; responsible-ai; xai", "keywords": "Artificial Intelligence (AI), Explainable artificial intelligence (xAI), machine learning (ML), stakeholder theory, human-AI collaboration, high-risk decision making, trust, organizational context, transaction monitoring (TM), anti-money laundering (AML)", "abstract": "The increasing adoption of machine learning (ML) and artificial intelligence (AI) in critical decision-making has intensified demands for explainability and transparency, particularly in high-risk domains such as finance. This dissertation investigates how explainable AI (xAI) can be designed and adapted to meet the diverse explainability needs of stakeholders in high-risk organizations through a longitudinal Action Design Research (ADR) study conducted in collaboration with a European financial institution implementing ML in transaction monitoring (TM) for anti-money laundering (AML). The ADR study addresses key gaps in the explainable AI literature by providing empirical evidence of how stakeholders interact with xAI in organizational contexts. The systematic literature review reveals four central debates in xAI research and identifies a critical need for empirical studies examining stakeholder needs. The empirical investigation demonstrates how implementation modes (automation versus augmentation) fundamentally alter stakeholder information requirements and liability structures, challenging assumptions about universal explainability needs. The research introduces a novel proximity-responsibility framework that systematically explains how stakeholders' proximity to AI systems influences both their trust in these systems and their utilization of xAI. High-proximity stakeholders (data scientists, investigators) develop trust through operational familiarity and use xAI for optimization and decision support, while low-proximity stakeholders (compliance managers, auditors) rely on xAI primarily for oversight, documentation, and regulatory compliance. Significantly, this dissertation challenges the prevailing explainability-trust hypothesis that dominates xAI literature by providing empirical evidence that trust influences explainable AI use rather than being produced by it. The research reconceptualizes trust and control as complementary coordination mechanisms and proposes a hierarchical conceptual framework positioning interpretability as methodology for developing xAI serving a specific purpose rather than generic trust generation. The findings contribute theoretical frameworks for understanding human-AI collaboration in high-stakes environments, practical design principles for stakeholder-specific xAI implementation, and methodological guidance for conducting rigorous xAI research in organizational contexts. The study demonstrates that effective xAI design requires moving beyond one-size-fits-all approaches toward nuanced, stakeholder-specific solutions that recognize the complex interplay between organizational context, implementation choices, and human factors in AI adoption. Keywords: Artificial Intelligence (AI), Explainable artificial intelligence (xAI), machine learning (ML), stakeholder theory, human-AI collaboration, high-risk decision making, trust, organizational context, transaction monitoring (TM), anti-money laundering (AML)", "source_url": "https:\/\/openalex.org\/W4414292580", "pdf_url": "", "doi": "10.22439\/phd.35.2025", "handle": "", "repository_name": "OpenArchive@CBS (Copenhagen Business School)", "source_record_id": "W4414292580", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:59", "updated_at": "2026-11-15 16:52:59" }, { "id": 1164, "title": "Uncertainty in Explainable Artificial Intelligence", "title_ar": "", "title_en": "Uncertainty in Explainable Artificial Intelligence", "author": "Teodor Chiaburu", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "", "university_source": "", "college": "", "department": "", "country": "", "governorate": "", "publication_year": 2026, "language": "en", "research_domain": "responsible-ai", "research_subdomain": "xai", "categories": "deep-learning; responsible-ai; ai-applications; multimodal; xai; fairness; ai-industry", "keywords": "", "abstract": "Artificial intelligence (AI) has become deeply embedded in modern decision-making systems, driving advances across science, industry and society. Yet, as models grow in scale and complexity, they increasingly operate as opaque systems whose internal reasoning remains difficult to understand. Explainable AI (XAI) seeks to remedy this by providing interpretable insights into model behavior, supporting model verification, accountability and calibrated human-AI trust. However, a growing body of work shows that explanations themselves are oftentimes unstable, noisy or sensitive to slight perturbations - a variation that could be attributed to datasets, models or explanation techniques. Without a thorough understanding of where this uncertainty comes from and a reliable way to communicate it, explanations risk being misleading, unreliable and ultimately detrimental to user trust. This threatens to turn XAI from a transparency tool into a further source of ambiguity. In this dissertation, we address these challenges by developing a principled framework for Uncertainty in Explainable AI (UXAI) that conceptualizes and operationalizes uncertainty within the explanation process. First, we introduce a mathematical formulation of explainers as functions $e_{\\theta}(x, f)$, enabling a propagation-based view of how uncertainty originating in the data, model and explainer hyperparameters flows into the final explanation. Building on this formulation, we derive a test for checking how reliably uncertainty is propagated through the explanation pipeline, comparing first-order uncertainty approximations against empirical perturbation-based estimates. We then demonstrate how uncertainty quantification can enhance transparency in practice by (i) constructing a post-hoc stochastic concept-based explainer and (ii) conformalizing a multimodal, multitask model used in real-world scientific workflows. As far as the human dimension is concerned, we develop a human-in-the-loop experimental framework to study how explanations communicated together with model uncertainty influence human-AI collaboration, user performance and learning in unfamiliar tasks. Finally, we conclude by discussing what is still missing in UXAI and future directions that could advance the field further.", "source_url": "https:\/\/openalex.org\/W7140237592", "pdf_url": "", "doi": "10.14279\/depositonce-25521", "handle": "", "repository_name": "", "source_record_id": "W7140237592", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:59", "updated_at": "2026-11-15 16:52:59" }, { "id": 1165, "title": "A Framework for Explainable Artificial Intelligence in Geospatial Contexts", "title_ar": "", "title_en": "A Framework for Explainable Artificial Intelligence in Geospatial Contexts", "author": "Cédric Roussel", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "", "university_source": "", "college": "", "department": "", "country": "Germany", "governorate": "", "publication_year": 2026, "language": "en", "research_domain": "computer-vision", "research_subdomain": "xai", "categories": "machine-learning; computer-vision; responsible-ai; ai-applications; predictive-analytics; xai; ai-smart-cities", "keywords": "", "abstract": "Artificial intelligence has become a key driver of innovation in geospatial data analysis. It enables new prediction, classification and decision-making methods in areas such as urban planning, environmental monitoring and mobility services. However, many artificial intelligence models, particularly complex machine learning algorithms, lack transparency. This makes it difficult for users to understand, validate or trust the outcomes of these models. This is particularly problematic in geospatial contexts, where decisions often have real-world implications and depend on local knowledge and geospatial reasoning. While existing explainable artificial intelligence methods offer insights into model behavior, they often disregard the geospatial context, such as the location of phenomena and the relationships between nearby areas. This limits their effectiveness in understanding and interpreting the geospatial behavior of the model. This dissertation addresses this challenge by advancing the field of geospatial explainable artificial intelligence, an emerging area that redefines how explanations relating to geospatial data should be constructed and communicated. It proposes a novel framework that works with a new type of explanation which can be tailored to the geospatial context. Glocal explanations are computed through the aggregation of local explanations and defined conditions. This enables explanations to be created that range between local and global explainable artificial intelligence. As for explanations, the model quality is adapted to these glocal units, enabling analysis of where and when the model performs well or poorly. The framework also provides interactive visualizations that enable users to explore, interpret, and question model outputs. One visualization is an interactive matrix, while a second uses multi-level glyphs to enable geospatial analysis of the explanations. Both visualizations integrate model quality to reliably interpret the reasoning behind the model's decisions. The novel contributions are supported by real-world experiments and prototype implementations. Three distinct use cases are considered: predicting rental bike station bookings, predicting car park occupancy, and classifying accident severity. An additional evaluation of the usability of the visualizations shows that representations such as the developed matrix or geovisualization significantly improve efficiency, effectiveness, and satisfaction. The final framework can be used by different types of users. Machine learning experts can use it to identify and understand where and why the model performs poorly. Domain experts can use the framework to gain a better understanding of the actions and causes within a use case, potentially enabling them to take actions that could have economic and social benefits.", "source_url": "https:\/\/openalex.org\/W7163335674", "pdf_url": "", "doi": "10.13140\/rg.2.2.17271.79525", "handle": "", "repository_name": "Fraunhofer-Publica (Fraunhofer-Gesellschaft)", "source_record_id": "W7163335674", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:59", "updated_at": "2026-11-15 16:52:59" }, { "id": 1171, "title": "eXAI: an innovative ensemble technique with explainable artificial intelligence applied to computer vision in critical contexts", "title_ar": "", "title_en": "eXAI: an innovative ensemble technique with explainable artificial intelligence applied to computer vision in critical contexts", "author": "Miguel Jose das [UNESP] Neves", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "", "university_source": "", "college": "", "department": "", "country": "", "governorate": "", "publication_year": 2025, "language": "en", "research_domain": "medical-ai", "research_subdomain": "cnn", "categories": "machine-learning; deep-learning; computer-vision; medical-ai; responsible-ai; predictive-analytics; cnn; multimodal; xai; fairness", "keywords": "", "abstract": "The growing complexity of Deep Learning models, particularly Convolutional Neural Networks (CNNs), presents a fundamental challenge in critical applications: the opacity of their decisionmaking processes. This thesis addresses this interpretability gap by proposing and validating a methodological framework termed eXAI (Ensemble Explainable Artificial Intelligence), designed to create image analysis systems that merge high predictive performance with transparency. Conceptually inspired by Ensemble Learning, the framework’s novelty lies not in combining predictive models, but in synergistically integrating an ensemble of explainability methods specifically combining SHAP for pixel-level attribution with Grad-CAM for region-level localization to provide a multi-faceted interpretation of a single model’s decisions. The framework’s flexibility and efficacy are demonstrated through two case studies in distinct domains. The first, focused on Diabetic Retinopathy diagnosis, implements a stacking architecture that integrates deep visual features from a CNN with structured clinical data, classified by a LightGBM (LGBM) meta-model. This multimodal approach achieved outstanding accuracy, while the XAI analysis validated the clinical relevance of its decision factors. The second case study, in digital forensics, applies a direct classification CNN to detect Seam Carving image manipulations. Here, XAI was crucial in revealing that the model learned to identify subtle structural distortion artifacts rather than relying on semantic content. The collective results demonstrate that the eXAI framework is a robust and adaptable approach for building trustworthy AI models for critical contexts. By delivering comprehensive and understandable explanations, this methodology enhances the confidence in, and practical utility of, artificial intelligence in both healthcare and information security.", "source_url": "https:\/\/openalex.org\/W7120300663", "pdf_url": "", "doi": "", "handle": "", "repository_name": "LA Referencia (Red Federada de Repositorios Institucionales de Publicaciones Científicas)", "source_record_id": "W7120300663", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:59", "updated_at": "2026-11-15 16:52:59" }, { "id": 1175, "title": "An Explainable Artificial Intelligence Approach to Forecasting Monthly Electricity Consumption", "title_ar": "", "title_en": "An Explainable Artificial Intelligence Approach to Forecasting Monthly Electricity Consumption", "author": "Tyra Wodén", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "KTH Royal Institute of Technology", "university_source": "repository_host", "college": "", "department": "", "country": "Sweden", "governorate": "", "publication_year": 2026, "language": "en", "research_domain": "responsible-ai", "research_subdomain": "xai", "categories": "machine-learning; responsible-ai; ai-applications; predictive-analytics; xai; ai-smart-cities", "keywords": "", "abstract": "Recent years have seen instability in the electricity market and increasing energy demands in the world. Reliable electricity consumption forecasting is a helpful tool for consumers and other stakeholders to manage energy consumption more efficiently. This thesis forecasts the monthly electricity consumption of individual residential households in Sweden using five robust ensemble learning models. To address a gap in previous research, the interpretability of the monthly residential forecasts is enhanced using insights produced by the explainable artificial intelligence (XAI) method SHAP. A diverse set of features, including historical consumption, household attributes, weather data, and calendar information, is investigated for their contributions to the model output. Results show that the ensemble models are able to forecast the consumption with around 15% prediction error, although performance is highly impacted by extreme values for some households. The XAI analysis reveals that recent consumption is the most determining factor for the forecasts, while external factors have a more varied impact depending on model. Two external features that are influential across all models are sunshine amount and wind direction, while several other features show complex or weak relationships to model output. Furthering the research on interpretable forecasts is important for increasing end-users' trust in the forecasts, and subsequently promoting more efficient energy management.", "source_url": "https:\/\/openalex.org\/W7165082060", "pdf_url": "", "doi": "", "handle": "", "repository_name": "KTH Publication Database DiVA (KTH Royal Institute of Technology)", "source_record_id": "W7165082060", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:59", "updated_at": "2026-11-15 16:52:59" }, { "id": 1216, "title": "Analisis dampak penggunaan ChatGPT dalam kebiasaan pencarian informasi mahasiswa Universitas Brawijaya", "title_ar": "", "title_en": "", "author": "Uswatun Hasanah", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "Maulana Malik Ibrahim State Islamic University", "university_source": "repository_host", "college": "", "department": "", "country": "Indonesia", "governorate": "", "publication_year": 2025, "language": "id", "research_domain": "nlp", "research_subdomain": "llm", "categories": "deep-learning; nlp; ai-applications; transformers; llm; arabic-nlp; ai-education", "keywords": "", "abstract": "INDONESSIA Perkembangan teknologi kecerdasan buatan (AI) seperti ChatGPT telah mempengaruhi kebiasaan mahasiswa dalam mencari informasi akademik, sehingga semakin sering menggunakan ChatGPT dalam pemenuhan kebutuhan informasi dan akademik mereka. Penggunaan ChatGPT yang terlalu sering berpotensi dalam mengubah kebiasaan pencarian informasi, pola penggunaan perpustakaan, preferensi terhadap sumber informasi, serta tingkat ketelitian mahasiswa dalam memilah informasi. Penelitian ini bertujuan menganalisis dampak penggunaan ChatGPT terhadap kebiasaan pencarian informasi mahasiswa Universitas Brawijaya. Metode penelitian menggunakan pendekatan kuantitatif dengan penyebaran kuesioner kepada 106 responden. Penelitian ini beracuan pada penelitian lain yang telah dilakukan oleh Karunaratne & Adesina (2023) dengan topik serupa menggunakan lima indikator, yaitu awareness, use and effectiveness, challenges, trust in the function, dan user competence. Hasil penelitian menunjukkan bahwa mahasiswa telah merasakan dampak dari penggunaan ChatGPT dalam proses pencarian informasi. Mahasiswa lebih sering menggunakan ChatGPT sebagai titik awal pencarian informasi karena kemudahan akses dan kecepatan respons, kemudian memverifikasi hasil yang diperoleh melalui sumber ilmiah lain. Sikap kritis ini membantu mahasiswa menghindari kesalahan akibat keterbatasan akurasi ChatGPT sekaligus mencegah ketergantungan berlebihan. Selain itu, pada indikator awareness diketahui bahwa mahasiswa Universitas Brawijaya menyadari manfaat ChatGPT sebagai sarana pencarian informasi cepat, namun pada saat yang sama juga memahami keterbatasannya sehingga menggunakannya secara bijak dan selektif. Pada indikator user competence menunjukkan bahwa penggunaan ChatGPT telah mendorong mahasiswa terbiasa merumuskan pertanyaan lebih spesifik dan meningkatkan kemampuan mengevaluasi kredibilitas informasi. ChatGPT berperan sebagai alat bantu yang mempercepat pencarian informasi, tetapi tetap ditempatkan secara proporsional sebagai pendukung proses belajar, bukan sumber rujukan utama. ENGLISH The development of artificial intelligence (AI) technology such as ChatGPT has influenced students' habits in searching for academic information, so that they increasingly use ChatGPT to fulfill their information and academic needs. Excessive use of ChatGPT has the potential to change information search habits, library usage patterns, preferences for information sources, and the level of accuracy of students in sorting information. This study aims to analyze the impact of ChatGPT usage on the information search habits of students at Brawijaya University. The research method uses a quantitative approach with a questionnaire distributed to 106 respondents. This study refers to another study conducted by Karunaratne & Adesina (2023) on a similar topic using five indicators, namely awareness, use and effectiveness, challenges, trust in the function, and user competence. The results show that students have felt the impact of using ChatGPT in the information search process. Students more often use ChatGPT as a starting point for information search because of its ease of access and speed of response, then verify the results obtained through other scientific sources. This critical attitude helps students avoid mistakes due to ChatGPT's accuracy limitations while preventing excessive dependence. Additionally, the awareness indicator shows that Brawijaya University students recognize the benefits of ChatGPT as a means of quick information retrieval, but at the same time understand its limitations, so they use it wisely and selectively. The user competence indicator shows that the use of ChatGPT has encouraged students to formulate more specific questions and improve their ability to evaluate the credibility of information. ChatGPT serves as a tool that accelerates information retrieval, but it is still placed proportionally as a support for the learning process, not the primary reference source. ARABIC أثر تطور تكنولوجيا الذكاء الاصطناعي مثل شات جي بي تي على عادات الطلاب في البحث عن المعلومات الأكاديمية. يتزايد استخدام الطلاب لـ شات جي بي تي لتلبية احتياجاتهم المعلوماتية والأكاديمية. الاستخدام المتكرر لـ شات جي بي تي من شأنه أن يغير عادات البحث عن المعلومات، وأنماط استخدام المكتبات، وتفضيلات مصادر المعلومات، ومستوى دقة الطلاب في فرز المعلومات. تهدف هذه الدراسة إلى تحليل تأثير استخدام شات جي بي تي على عادات البحث عن المعلومات لدى الطلاب في جامعة براويجايا. تستخدم طريقة البحث نهجًا كميًا مع استبيان تم توزيعه على ١٠٦ مستجيبًا. تشير هذه الدراسة إلى دراسة أخرى أجراها كاروناراتنه وأديسينا (٢٠٢٣) تناقش موضوعًا مشابهًا باستخدام خمسة مؤشرات تم تحليلها، وهي الوعي والاستخدام والفعالية والتحديات والثقة في الوظيفة وكفاءة المستخدم. تظهر النتائج أن الطلاب قد شعروا بالتأثير الحقيقي لاستخدام شات جي بي تي في عملية البحث عن المعلومات.", "source_url": "https:\/\/openalex.org\/W7125093218", "pdf_url": "", "doi": "", "handle": "", "repository_name": "eTheses of Maulana Malik Ibrahim State Islamic University (Maulana Malik Ibrahim State Islamic University)", "source_record_id": "W7125093218", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:52:59", "updated_at": "2026-11-15 16:52:59" }, { "id": 1246, "title": "Mutual Reinforcement Learning", "title_ar": "", "title_en": "Mutual Reinforcement Learning", "author": "Cameron Reid", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "Purdue University System", "university_source": "repository_host", "college": "", "department": "", "country": "United States", "governorate": "", "publication_year": 2021, "language": "en", "research_domain": "reinforcement-learning", "research_subdomain": "", "categories": "reinforcement-learning", "keywords": "", "abstract": "Mutual learning is an emerging field in intelligent systems which takes inspiration from naturally intelligent agents and attempts to explore how agents can communicate and cooperate to share information and learn more quickly. While agents in many biological systems have little trouble learning from one another, it is not immediately obvious how artificial agents would achieve similar learning. In this thesis, I explore how agents learn to interact with complex systems. I further explore how these complex learning agents may be able to transfer knowledge to one another to improve their learning performance when they are learning together and have the power of communication. While significant research has been done to explore the problem of knowledge transfer, the existing literature is concerned either with supervised learning tasks or relatively simple discrete reinforcement learning. The work presented here is, to my knowledge, the first which admits continuous state spaces and deep reinforcement learning techniques. The first contribution of this thesis, presented in Chapter 2, is a modified version of deep Q-learning which demonstrates improved learning performance due to the addition of a mutual learning term which penalizes disagreement between mutually learning agents. The second contribution, in Chapter 3, is a presentation work which describes effective communication of agents which use fundamentally different knowledge representations and systems of learning (model-free deep Q learning and model-based adaptive dynamic programming), and I discuss how the agents can mathematically negotiate their trust in one another to achieve superior learning performance. I conclude with a discussion of the promise shown by this area of research and a discussion of problems which I believe are exciting directions for future research.", "source_url": "https:\/\/openalex.org\/W3177417054", "pdf_url": "", "doi": "10.25394\/pgs.14439575", "handle": "", "repository_name": "Purdue e-Pubs (Purdue University System)", "source_record_id": "W3177417054", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:53:00", "updated_at": "2026-11-15 16:53:00" }, { "id": 1370, "title": "تأثير تقنيات الذكاء الاصطناعي على جودة علاقات العملاء \"دراسة تطبيقية على عملاء المتاجر الإلكترونية\"", "title_ar": "تأثير تقنيات الذكاء الاصطناعي على جودة علاقات العملاء \"دراسة تطبيقية على عملاء المتاجر الإلكترونية\"", "title_en": "", "author": "Talaat Asaad Abdelhamid ; Ahmed Ghoneim ; Mayan Magdy", "supervisor": "", "degree_type": "unspecified", "degree_evidence": "", "university": "", "university_source": "", "college": "", "department": "", "country": "", "governorate": "", "publication_year": 2024, "language": "ar", "research_domain": "nlp", "research_subdomain": "predictive-analytics", "categories": "machine-learning; nlp; predictive-analytics", "keywords": "", "abstract": "هدفت هذه الدراسة إلى بحث تأثير تقنيات الذكاء الاصطناعي على جودة علاقات العملاء، وفى ضوء مراجعة الدراسات السابقة تم استخلاص إطار مفاهيمي لأبعاد هذه الدراسة يتضمن متغيرين أساسيين، وهما المتغير المستقل ويتمثل في تقنيات الذكاء الاصطناعي (روبوتات الدردشة، التسعير الديناميكي، العروض الموجهة، تحليل البيانات، فهم العملاء، توصية المحتوى) والمتغير التابع جودة علاقات العملاء (الرضا، الثقة، الالتزام، الولاء)، وتم الاعتماد في جميع البيانات الأولية لهذه الدراسة على عينة النهر (عينة الإنترنت الاعتراضية) وبلغ عددها (412) مفردة من عملاء المتاجر الالكترونية، وتم تحليل البيانات باستخدام مصفوفة بيرسون وثبت وجود علاقة ارتباط (متوسطة القوة) ذات معنوية إيجابية بين أبعاد الدراسة ( روبوتات الدردشة، التسعير الديناميكي، العروض الموجهة، تحليل البيانات، فهم العملاء، توصية المحتوى، الرضا، الثقة، الالتزام، الولاء)، حيث جاءت دالة إحصائيًا عند مستوى دلالة (0.01)، وأيضاً تم استخدام معامل الانحدار الخطى المتعدد وثبت وجود تأثير معنوي جزئي بين تقنيات الذكاء الاصطناعي (روبوتات الدردشة، التسعير الديناميكي، العروض الموجهة، تحليل البيانات، فهم العملاء، توصية المحتوى) وجودة علاقات العملاء (الرضا، الثقة، الالتزام، الولاء)، وذلك باستخدام البرنامج الاحصائي IBM SPSS V.29، كما قدمت الدراسة مجموعة من التوصيات التي تلقى الضوء على أهمية تقنيات الذكاء الاصطناعي. This study aimed to determine The Effect of Artificial Intelligence Techniques on Customers' Relationship Quality, Based on a review of previous studies, a conceptual framework was extracted for the dimensions of this study that included two main variables: The independent variable, Represented by Artificial intelligence techniques (Chatbots, dynamic pricing, Promotional Personalization, data analysis, understanding customers, content recommendation) and the dependent variable, customer relationship quality (satisfaction, commitment, trust, loyalty). The primary data for this study was collected through a river sample (Accidental sample) of 412 E-Stores customers, The researcher also used a set of secondary data, obtained from books, articles, and periodicals and scientific dissertations. The results by using the Pearson correlation matrix showed that there is a statistically significant positive correlation (Medium-strength) between the dimensions of the study (Chatbots, dynamic pricing, Promotional Personalization, data analysis, understanding customers, content recommendation, satisfaction, commitment, trust, loyalty), The statistical significance was at a level of (0.01). The study also used the multiple linear regression coefficient and found that there is a partial significant effect between artificial intelligence technologies (Chatbots, dynamic pricing, Promotional Personalization, data analysis, understanding customers, content recommendation) and Customers' Relationship Quality (satisfaction, trust, commitment, loyalty). The study was conducted using the IBM SPSS statistical program V29. The study also presented a set of recommendations that highlight the importance of artificial intelligence Techniques.", "source_url": "https:\/\/openalex.org\/W7171508316", "pdf_url": "", "doi": "10.5281\/zenodo.21593077", "handle": "", "repository_name": "Zenodo (CERN European Organization for Nuclear Research)", "source_record_id": "W7171508316", "other_sources": "", "verification_status": "needs_review", "created_at": "2026-11-15 16:53:01", "updated_at": "2026-11-15 16:53:01" } ] }