21793 research outputs found
Sort by
“The system is stacked against them”: Investigating issues of fairness in the IELTS writing task for test takers
Get PDFThis research investigated the perceived fairness of the IELTS writing assessment from the perspectives of educators/examiners and test takers. Using a mixed methods approach, the study gathered quantitative data from questionnaires completed by 30 participants and qualitative information from interviews. Quantitative findings revealed dissatisfaction among educators, examiners, and test-takers with the IELTS writing assessment, emphasizing its lack of clarity of assessment criteria and fairness. Qualitative data identified four critical themes for both groups: Unclear Scoring Criteria, Cultural Bias, Life-Changing Consequences, and Technological Impact. Themes showed limitations within the scoring system and their impacts on test takers’ futures. Notably, the Life-Changing Consequences theme highlighted deep social impacts of writing scores which delayed university admissions, hindered job opportunities and complicated immigration processes for some candidates. The findings called for a re-evaluation of IELTS scoring criteria and advocated for clearer guidelines to mitigate subjectivity and bias. By tackling these issues, the study sought to improve fairness and lessen the unfair challenges for test takers, aligning the test more accurately with their true language abilities
Detecting Persuasion Techniques in Memes
Get PDFMemes, which are user-generated content in the form of images and text, have become a powerful medium for shaping public discourse. Given their increasing influence, detecting persuasive techniques embedded within these multimodal forms of communication is crucial for identifying propaganda and combating online disinformation. Persuasion techniques in memes often combine rhetorical elements from both text and image, creating unique challenges for computational models.
This thesis seeks to determine the impact of multimodal integration on the detection of persuasion techniques in memes and to evaluate how well multimodal models perform compared to single-modality models in this classification task. To achieve this, we developed and fine-tuned several models for text-based and multimodal persuasion detection using both pre-trained language models (BERT, XLM-RoBERTa, mBERT) and image-based models (CLIP, ResNET, VisualBERT).
A key contribution of this work is the implementation of paraphrase-based data augmentation, which helped address class imbalance and improved the performance of text-only models. For multimodal approaches, we explored both early fusion and cross-modal alignment strategies. Surprisingly, cross-modal alignment underperformed, likely due to challenges in aligning abstract textual and visual cues. In contrast, the early fusion approach of combining text and image embeddings showed the highest performance, significantly outperforming text-only and image-only models.
We also conducted zero-shot experiments with GPT-4 to benchmark its effectiveness in multimodal persuasion detection. Although GPT-4 demonstrated potential in zero-shot settings, the fine-tuned models still outperformed it, particularly when leveraging multimodal integration.
This research advances the understanding of multimodal learning for detecting persuasion techniques, with broader implications for disinformation detection in online content
A hybrid NFV/In-Network Computing MANO Architecture for provisioning Holographic Applications
Get PDFThe emergence of innovative holographic applications, such as holographic concerts, have requirements on network infrastructure such as high bandwidth and ultra-low latency. As demonstrated in recent research, for efficient provisioning of these types of applications, we require a Hybrid Network Function Virtualization (NFV) / IN-Network Computing (INC) network infrastructures.
INC, a novel technology, distributes computational workloads across the network by deploying computational tasks on programmable devices like routers and switches. However, integrating INC into existing infrastructure presents significant challenges in terms of life-cycle management and orchestration of network functions and components. The network functions in these hybrid environments consist of two types of components, including VNFs and INCs. While the ETSI Management and Orchestration (MANO) framework enables application provisioning in NFV-enabled networks, it does not provide support for INC-enabled environments. Therefore it is necessary to develop a new MANO architecture capable of provisioning applications that incorporate both VNF and INC components.
The main contribution of this thesis is twofold. First, we propose a novel hybrid NFV-INC MANO architecture which is specifically designed for provisioning holographic applications within a hybrid NFV/INC environment. This architecture builds upon the foundation established by the ETSI NFV MANO framework. Second, the proposed architecture is prototyped and measurements are made to evaluate the deployment time of the hybrid Network Services and showing that deploying INC components takes less time than deploying VNFs.
The essential NFV-INC MANO functional entities, such as INC management module, NFV-INC orchestration, and INC infrastructure manager are identified. In addition, a set of RESTful interfaces is proposed to enable interaction with components.
To validate the proposed concept, the prototype is constructed utilizing Open Source MANO (OSM) for orchestration and management network services. Microk8s is employed for container orchestration, automating the deployment of Kubernetes Network Functions (KNFs). Also, the Mininet emulator prepares the programmable switches which serve as the INC infrastructure. The performance of the proposed architecture is evaluated through comprehensive measurements. Additionally, the architecture is validated by concrete measurements on the deployment latency for five different test cases
Detection, Identification and Isolation of Cyber-Attacks using Enhanced Long Short-Term Memory in Single and Network of Quadcopters
Get PDFThe cybersecurity of cyber-physical systems (CPS), particularly quadcopters, is critical due to their reliance on communication networks, which makes them vulnerable to cyber-attacks. This thesis addresses the security of quadcopters by introducing a novel framework for the simultaneous detection, identification, and isolation of cyber-attacks using Long Short-Term Memory (LSTM) networks. Unlike previous research that primarily focuses on detection, this work advances the field by integrating attack type identification and target isolation, enhancing overall security capabilities.
A contribution of this thesis is the emphasis on sequence generation as a pre-processing step for time-series data in LSTM models. By optimizing sequence length, overlap, and labeling methods, the proposed approach ensures the effective capture of temporal dependencies, substantially improving model performance for attack detection, identification, and isolation.
The study introduces a novel multi-output (MO) model for single quadcopters, utilizing a shared LSTM backbone with three output heads. This framework is extended to a network of quadcopters through a Multi-Input, Multi-Output (MIMO) architecture, which incorporates a flexible number of input heads for each quadcopter, enhancing scalability. The model supports both centralized and decentralized topologies, accommodating networks of varying sizes, ranging from 2 to 5 quadcopters.
Simulation results for Denial of Service (DoS), False Data Injection (FDI), and Replay attacks demonstrate the robustness of the proposed framework. The single quadcopter model achieved over 95% accuracy in attack detection, along with high precision in identifying attack types and locations. In networked setups, the centralized MIMO model delivered superior performance, while the decentralized approach also yielded promising results. These findings highlight the adaptability and effectiveness of the proposed approaches, paving the way for broader CPS applications and further advancements in sequence generation techniques
Leak Localization in Water Distribution Networks with a Hybrid CNN-LSTM Model and Novel Distance-Based Loss Function
No full textWater loss in distribution networks represents a critical challenge in urban water management, with significant environmental, economic, and public health implications. Around 126 million cubic meters of treated water are globally lost annually through distribution network leakages. This thesis presents a hybrid deep-learning approach for leakage localization that minimizes the distance between predicted and actual leak locations in water distribution networks. The developed algorithm integrates hydraulic modelling with deep learning techniques through three main phases: (1) Hydraulic Model Modification, (2) Water Demand Adjustment and Simulations, and (3) Leak Localization. A hybrid CNN-LSTM model was developed, combining CNN for spatial feature extraction and LSTM for temporal pattern recognition. Additionally, a hybrid loss function was designed to optimize both classification accuracy and distance in leak localization. The algorithm was validated using the Battle of L-Town Water Distribution Network benchmark. Results demonstrated significant improvements in localization accuracy, achieving average errors of 120.15 meters for background leakages and 98.76 meters for bursts when using the hybrid loss function - representing 29% and 39% reductions, respectively, compared to standard loss functions. The main contributions of this thesis are two-fold: the accuracy of leakage localization has been significantly improved through the development of a hybrid CNN-LSTM model with a customized hybrid loss function, and the search area for leak identification has been reduced, with the distance-based loss function helping to minimize the physical distance between predicted and actual leak locations. This reduction in search area translates directly to decreased inspection time and costs for water utilities
Artificial Intelligence-Driven Recommender Solutions for E-Commerce: A Multidisciplinary Approach for Enhancing Collaborative Filtering Quality.
Get PDFAI-driven recommender systems are transforming E-commerce by taking on tasks traditionally handled by human staff, such as sales associates, inventory clerks, and others. They apply machine learning methods to replace human decisions, enhance accuracy and scalability, and improve customer experiences. In this context, significant achievements have been witnessed in recent years in improving collaborative filtering-based recommender systems (CFRSs) through optimizing recall and normalized discounted cumulative gain (NDCG) metrics. Nonetheless, major issues remain that significantly limit the performance and generalization of these systems, such as diversity and novelty in recommendations, fairness, inclusion of long-tail items, the cold start problem, reproducibility, and evaluation overfitting. This study advocates the need for new approaches for addressing these problems comprehensively, moving beyond the traditional optimization metrics (recall and NDCG).
This work is novel in its multidisciplinary approach, integrating principles from systems engineering, software engineering, and TRIZ into the development and optimization of CFRSs. Since systems engineering takes a holistic standpoint, it allows for the reasoning and optimization of user-item interactions. Meanwhile, Software engineering provides several systematic ways and techniques to analyze and improve the functional parts of CFRSs. The TRIZ methodology facilitates the development of innovative solutions and AI tools to eliminate technical contradictions and enhance the performance of CFRSs.
To guide the optimization of CFRSs, the research also uses the ISO/IEC 25010:2011 standards to evaluate the CFRSs thoroughly. These standards evaluate the reliability, usability, performance efficiency, and privacy of the CFRSs against high-quality benchmarks. The evaluation of the results based on real-world data pertaining to E-commerce datasets demonstrates that the recommendation accuracy, diversity, and coverage were improved.
All in all, the current research improves CFRS technology by providing robust, innovative, and user-centric solutions. The proposed multidisciplinary approaches serve as a template for future research and development work. The findings achieved, peer-reviewed, and published in various publications have contributed to the discourse in academics along with practical implementation by creating high-quality CFRSs
Efficient Explainable AI And Adversarial Robustness using Formal Methods
Get PDFArtificial Intelligence (AI) systems are increasingly used in critical applications, but their lack of transparency often hinders trust and reliability. Explainable AI (XAI) addresses this by making machine learning models more understandable and interpretable. Current XAI approaches, however, lack consistency or theoretical guarantees. Formal methods, which are rigorous mathematical reasoning tools, could play a significant role in overcoming these limitations by providing sound and consistent explanations for model decisions. This thesis builds upon an existing formal XAI tool, XReason, which uses logical reasoning to generate explanations for individual predictions. The contributions of this work are threefold. First, the tool is extended to support a powerful tree based model, Light Gradient Boosting Machine (LighGBM), which offers improved scalability and performance for large datasets. Second, it introduces explanations at the class level, enabling the analysis of general patterns in model behavior across different prediction categories. This provides insights into the factors shaping model decisions for each class and helps identify biases or inconsistencies in predictions. Third, adversarial robustness is explored by integrating methods to generate and detect adversarial examples. These adversarial samples expose vulnerabilities in the model by identifying subtle input changes that lead to incorrect predictions. Detection mechanisms are then developed to identify such inputs, enhancing the model’s reliability. Experiments on a variety of datasets from different domains demonstrate that the extended framework produces consistent and robust explanations, both at the individual prediction level and across broader trends. By integrating formal methods, this work provides a practical formal XAI framework applicable to areas where trust in AI systems is essential
Business Owners’ Careers: Motives in the Longer Term
Get PDFThis thesis looks at the motives that affect the careers of business owner-managers. Using McClelland's Need Theory and Socioemotional Selectivity Theory, this research examines how achievement, power, and affiliation shift over time and affect decisions about whether to continue working or retire.
The study used a qualitative approach, conducting semi-structured in-depth interviews with twelve business owner-managers over fifty years old. These interviews provided valuable insights into how their motives have changed throughout their lives, the challenges they face regarding retirement, and the effects of their decisions on themselves and their businesses. The analysis explored six themes: Achievement, Power, Affiliation, Emotional Fulfillment, Perceived Change, and Retirement Plans. These themes help explain why business owner-managers stay involved in their work even past the traditional retirement age.
The findings show that early in their careers, many business owner-managers focus on financial success, innovation, and influence. As they age, their priorities shift toward sustaining their businesses, creating a legacy, and building relationships. By exploring these changes, this thesis helps deepen our understanding of how aging business owner-managers manage their careers and what keeps them engaged. These insights are essential for the business community as it considers the significant impact of aging business owner-managers on sustainability and the economy
Essays on Human Capital Management Disclosure
Get PDFThis dissertation consists of two essays on the newly mandated Human Capital Management (HCM) disclosure introduced by the Securities and Exchange Commission (SEC) in 2020. The SEC adopts a principles-based approach, allowing firms discretion over what to disclose, resulting in significant variation in HCM disclosures. The first essay investigates the role of board Human Resources (HR) governance on HCM disclosure transparency. More companies are renaming their compensation committees as HR committees to reflect a broader responsibility for HCM, and more companies are appointing directors with HR expertise. The results show that board HR governance is positively associated with HCM disclosure transparency, but only when both an HR committee and HR expertise are present. Conversely, when either mechanism exists in isolation, HCM disclosure transparency is lower. These findings suggest that both mechanisms are necessary to promote effective HCM transparency. The results are more pronounced in firms with a Chief Human Resources Officer (CHRO) and firms experiencing employment growth. These findings hold true when using an entropy-balanced sample. The second essay examines the usefulness of HCM disclosure transparency to financial analysts, who play a key role in capital markets. The results show that HCM transparency is associated with higher analyst forecast accuracy but has no significant association with forecast dispersion. This suggests that HCM transparency adds to individual analysts' private information, helping them assess firm value from HCM practices. Furthermore, the second essay identifies which HCM topics are most relevant to analysts. The results reveal that topics related to attraction, retention, development, turnover, compensation and benefits, diversity, equity, inclusion (DEI), and culture are associated with higher forecast accuracy, particularly when the numerical intensity of these topics is higher. This suggests that not all HCM topics are equally informative, with a clear emphasis on quantitative details that provide analysts with the information needed to evaluate firm value effectively
Managing E-Learning Projects for Workplace Learning: An Integrative Review of the Peer Reviewed and Professional Literature
Get PDFABSTRACT
Managing E-Learning Projects for Workplace Learning: An Integrative Review of the Peer Reviewed and Professional Literature
Tetiana Brandt
Research Problem: Workplace learning organizations face challenges in managing e-learning projects effectively as they strive to adapt to the digital age and meet modern learners’ needs. This study aims to explore theses management practices, focusing on how organizations handle e-learning projects and the factors that influence their approaches.
Research Question: How do workplace learning organizations manage e-learning projects?
Literature Review: The literature review focuses on two essential bodies of literature that are fundamental to the management of e-learning projects: instructional design models and project management theory. Together, these perspectives offer a comprehensive foundation for understanding how e-learning projects are managed in workplace settings.
Method: The study employs an integrative literature review methodology, with sources selected through replicable search parameters. The final dataset includes 57 sources published between 20023 and 2023. Content analysis was then applied to examine publications, a method well-suited for uncovering patterns and themes within the literature.
Results and Conclusions: The study reveals key insights into managing e-learning projects in workplace learning organizations, offering implications for practice, research and theory. Practitioners should understand the distinct roles of instructional design models and projects management methodologies, and how to integrate both effectively. Theoretical contributions highlight the need for hybrid frameworks addressing project complexity and emerging technology. Limitations include reliance of English-language sources and a focus on specific industries. Future research should explore diverse contexts, validate findings empirically, and assess the impact of AI and new technologies on e-learning project management