SACM - United Kingdom
Permanent URI for this collectionhttps://drepo.sdl.edu.sa/handle/20.500.14154/9667
Browse
15 results
Search Results
Item Restricted Video Synthesis of a Talking Head(University of Leeds, 2023-07) Alghamdi, Mohammed Mesfer A; Hogg, David CThe ability to synthesise a video talking head from speech audio has many potential applications, such as video conferencing, video animation production and virtual assistants. Although there has been considerable prior work on this task, the quality of generated videos is typically limited in terms of overall realism and resolution. In this thesis, we propose a novel approach for synthesis of a high-resolution talking head video (1024x1024 in our experiments) from speech audio and a single identity image. The approach is built on top of a pre-trained StyleGAN image generator. We model trajectories in the latent space of the generator conditioned on speech utterances. To train this model, we use a dataset of talking-head videos, which are mapped into the latent space of the image generator using an image encoder that is also pre-trained. We train a recurrent neural network to map from speech utterances to sequences of displacements in the latent space of the image generator. These displacements are applied to the back-projection into the latent space of a single identity frame chosen from a target video in the training dataset. The thesis begins by reporting on an experimental evaluation of existing GAN inversion methods that map video frames into the latent space of a pre-trained StyleGAN image generator. We apply one such inversion method to train an unconditional video generator that requires an identity image and a random seed for the dynamical process that generates a trajectory through the latent space of the image generator. We evaluate our method for talking head synthesis from speech audio with standard measures and show that it significantly outperforms recent state-of-the-art methods on commonly used audio-visual talking-head datasets (GRID and TCD-TIMIT). We perform the evaluation with two versions of StyleGAN; one trained on video frames depicting talking heads and the other on faces with static expressions (i.e., not talking). The quality of the results is shown to be better when using StyleGAN pre-trained on talking heads. However, the range of possible identities is narrower due to the much smaller set of identities in the talking head dataset. The videos from experiments can be found at https://mohammedalghamdi.github.io/phd-thesis-website/23 0Item Restricted A Comparison of Automated Tracing Using Artificial Intelligence (AI) Software and Manual Digital Tracing Software.(Saudi Digital Library, 2023) Almalki, Malik; McGuinness, Niall J. P; Ulhaq, AmanBackground: Various fields of dentistry have been significantly influenced by Artificial Intelligence. One of the most prominent uses of artificial intelligence in orthodontics is automated cephalometric analysis. There have been many different automated cephalometric software developed recently, and they claim to be as effective as digital cephalometric analysis. Aim: To assess whether or not there is a statistically significant difference in time taken to establish the cephalometric analysis using three methods: Dolphin software, AI-generated cephalometric landmarks on WebCeph, and manually-modified cephalometric landmarks on WebCeph, also to assess whether or not there are statistically significant differences in the cephalometric analysis measurements between the same three methods. Methods: Thirty lateral cephalometric radiographs of patients were consecutively selected, and cephalometric analyses were done with three methods: digital tracing using Dolphin, automated tracing using WebCeph, and automated tracing using WebCeph but with landmark modification. Twenty-one measurements were obtained. The duration of each method was measured in seconds, and the results were tallied. Values were registered in a spreadsheet. Statistical analysis One-way ANOVA and The Kruskal–Wallis test were performed. The intraclass correlation coefficient (ICC) was utilised to determine the level of agreement between the measurements obtained from all three groups. Results: There is a statistically significant difference in the tracing time between the three groups (p-value = 0.0001). On the other hand, no statistically significant difference was found between the groups when comparing lateral cephalometric tracing measurement values (P> 0.05). Moreover, a high level of agreement is evident between the measurements from each group. Conclusions: Compared to Dolphin tracing, WebCeph cephalometric values are relatively accurate. It is economical, practical, and effective for routine orthodontic practices.18 0Item Restricted the evaluation of AI-generated subtitles(Saudi Digital Library., 2023) Alharbi, Amjad; Tilmann AltenbergAI in translation studies is a relatively new field that has piqued the interest of researchers due to its potentials on the translation as practice and a field. Following a review of subtitle evaluation and AI within the field of translation studies, this dissertation explores the domain of AI-generated subtitles and assesses their quality in comparison to human-generated subtitles. Through manual comparative analyses and utilizing quality assessment models FAR and MQM, the research uncovers the strengths and weaknesses of AI-generated subtitles, shedding light on their struggles with nuanced language use and cultural references. Furthermore, the research underscores the paramount importance of cultural and contextual sensitivity in subtitling, an area where human subtitlers excel. It highlights the practical implications of these findings for translation practices and education, advocating for a balanced approach that harnesses the strengths of both AI and human capabilities. Despite certain limitations, including a limited sample size and Netflix-specific focus, the distraction illuminates the dynamic landscape of subtitling, emphasizing the potential for AI-human collaboration to optimize practices and ensure the delivery of high-quality subtitles.75 0Item Restricted Artificial Intelligence: Voice Assistants(University of Manchester, 2023) Almirabi, Alaa; Mehandjiev, NikolayThe increasing amount of data and the need to rapidly retrieve information has prompted individuals to rely on technology. Voice assistants have been developed as an innovative tool that provides a natural interface to retrieving information from the digital world and even instructing simple transactions such as purchasing items off the Web. A voice assistant is a software application driven by AI that operates as a personal information manager, such as Siri, Amazon Alexa and Google Now. However, a deeper understanding of users' perceptions regarding voice assistants is necessary to increase user engagement and ensure users continue using them. This need motivates the exploratory study presented in this thesis. It aims to investigate the relationship between individuals and their home's voice assistants to identify the main characteristics that facilitate establishing a significant and engaging connection between them, leading to their continuous use of the voice assistant. This research thus aims to enhance the post-adoption stage by providing a valuable contribution to improving the relationship between users and their home voice assistants. The main findings of this study indicate that both perceived anthropomorphism and perceived intelligence significantly impact the continuous intention to use and engage with the voice assistant. Nevertheless, the importance of these relationships comes from specific crucial mediators such as engagement, rapport, engagement motivations, perceived value and satisfaction. These factors significantly influence the user's intention to use and engage with voice assistant technology.24 0Item Restricted To what extent is contemporary copyright law prepared to regulate and respond to handle AI generated content ?(Saudi Digital Library, 2023-11-23) Alshamrani, Maram Saleh; Alshamrani, MaramIn the current technological age, intellectual property law faces an increasingly difficult task in terms of adapting to rapidly evolving technology. This new reality has triggered a profound legal debate and necessitated legislative modifications to accommodate the changes brought about by technology. The growing popularity of AI in recent decades has also shed light on the fact that human beings are not the sole source of creativity. AI has demonstrated its ability to generate original and innovative ideas, challenging the traditional notion of human exclusivity in this domain. Over the years, AI has resulted in the creation of works that are generated without human authorship. This occurrence raised concerns within the Copyright Registry regarding the ambiguous status of works produced with the assistance of computers1 .Hence, this research undertakes a rigorous examination of diverse contextual scenarios and furnishes a comprehensive evaluation of the complexities inherent in conferring copyright entitlement to works engendered by artificial intelligence (AI) within the parameters delineated by extant copyright criteria tailored for human-originated creativity. Additionally, the study furnishes cogent insights into the domain of algorithmic creativity. To this end, it systematically scrutinizes pertinent treaties, doctrinal tenets, codified copyright statutes, and judicial precedents that delineate the contours of authorship within the purview of the European Union (EU), the United States (US) and the UK. This analysis serves to uncover the existing regulatory interstices and inadequacies.It concludes that creativity is a trait not exclusive to humans. Algorithmic creativity, as harnessed by artificial intelligence (AI) to generate novel creations, embodies a societal shift that necessitates legal accommodations for safeguarding legal integrity. Moreover, the constituent components integral to legislation governing protected works can be suitably modified to encompass autonomic productions engendered by AI. Nevertheless, it is essential to note that the principal objective of copyright protection, which centers on fostering inventive endeavors by human authors, constitutes a pivotal facet surpassing the confines of technical and ontological dimensions of creativity. Given that AI's creative impetus lacks the incentive for innovation through acknowledgment and considering programmers as the primary architects of the software and, by extension, the indirect architects of the resultant output, a judicious resolution entails acknowledgement and endorsement of the exerted effort33 0Item Restricted Challenges posed by Artificial Intelligence to traditional Copyright law: Can Machine Learning meet and disrupt the test of originality?(Saudi Digital Library, 2023-11-23) Aldawsari, Norah; Guadamuz, AndresThe dissertation first discusses the Test of Originality in Copyright law. While Section 9(1) of the Copyrights, Designs, and Patents Act 1988 lists effort, skill, and labour as being central for creating the new work as core elements for determining the copyrightable aspects of the work, caselaw in the UK has also emphasised on the element of creative aspect of the work. In the context of AI works, it is possible to argue that AI is capable of creativity in terms of traits or behaviours that are generally related to the concept of ‘intelligence’ and are also increasingly seen in the AI technology. This has potential to disrupt the copyright law as it is positioned now. The test of originality, which is an essential criterion for assessment of the copyright protection, is then discussed in the context of works that can be produced with Machine Learning technology. The test of originality requires the assessment of the ‘author’s own intellectual creation’, which has implications for AI works. For example, originality is explained in terms of the personality of the creator in the EU law. On the other hand, in the UK, the Copyright, Designs and Patent Act 1988 creates a different perspective to machine generated content even if there is no human author of the same because the UK law allows person arranging the database to be considered as the author. The UK law also applies a de minimis rule so that the threshold or standard is low with regard to the extent of the effort, skill and labour required for assessing originality of the work. This dissertation argues that the minimum threshold of originality may depend on the type of works, so that the quantitative and qualitative labour becomes important for considering the applicability of copyright. It may be noted that qualitative requirements where applicable, may be difficult to prove for AI works since it is easy to identify the creative choices made by AI because of the process of data analysis and processing. This is not the same as human authors whose creative choices can be individualistic and unique. With regard to Section 9, the dissertation argues that it does not answer to all the range of works that can be created by Machine Learning since it is possible to develop works via Machine Learning without any explicit programming since it can learn from their past experiences. This may obliterate the need for a person who makes the arrangements for the work who is only active during initial programming. In such a situation, the test of originality may be relevant to determining the extent to which copyright may be applicable even if the person who made the arrangements was only active at the beginning of the programming. The nature of the work may also be relevant since while the works of literary nature can be done by the AI in a way that is original in the sense of skill and labour and investment but the same cannot be said of visual arts, where a question may be raised about the artistic choices that the AI may not be able to meet. However, it is important to identify the range of works that can be done by the AI and then also provide framework under copyright law that addresses the creativity of the Machine Learning since suitability of the traditional copyright law may be raised in future when the commercial interests of the companies or investors in AI technologies come in conflict with the legal gaps in copyright law.55 0Item Restricted Pattern Recognition & Predictive Analysis of Cardiovascular Diseases: A Machine Learning Approach(Saudi Digital Library, 2023-11-23) Alseraihi, Faisal Fahad; Naich, AmmarCardiovascular disease (CVD) is a predominant global health concern, with its impact becoming increasingly pronounced in low- and middle- income countries due to challenges like limited healthcare access, inadequate public awareness, and lifestyle-related risks. Addressing CVD's multifactorial origins, which span genetic, environmental, and behavioral domains, requires advanced diagnostic techniques. This research leverages the UCI Heart Disease dataset to develop a deep learning predictive model for CVD, incorporating 14 vital heart health parameters. The models performance is critically assessed against conventional machine learning approaches, shedding light on its efficiency and areas of refinement. Utilizing sophisticated Neural Network structures, this study strives to enhance predictive health analytics, aiming for timely CVD identification and intervention. As the integration of machine learning into healthcare deepens, it's crucial to ensure that these tools are robust, thoroughly evaluated, and augment clinical insights to reduce misdiagnosis risks.77 0Item Restricted DESIGN FOR ADDITIVE MANUFACTURING POST-PROCESSING: DEVELOPING GUIDELINES AND BEST PRACTICES FOR DIGITAL MANUFACTURING CHAINS(Saudi Digital Library, 2023-09-29) Alsalman, Ahmed; Tammas-Williams, SamAdditive manufacturing, a transformative paradigm within modern manufacturing, holds the promise of intricate, efficient, and cost-effective production processes. This dissertation embarks on an in-depth exploration of additive manufacturing, addressing its multifaceted dimensions, challenges, and potential. The overarching objective is to synthesize pragmatic guidelines that optimize additive manufacturing practices from design to post-processing. The research commences with a comprehensive literature review that surveys the landscape of additive manufacturing, revealing gaps in understanding and knowledge. It proceeds to a series of experimental investigations, including case studies, designed to illuminate the complexities and advantages inherent in additive manufacturing. Guided by a mixed-method approach, the study delves into topics ranging from design considerations to post-processing methodologies, leveraging insights from industry collaboration. The culmination of this research is the formulation of comprehensive guidelines that navigate additive manufacturing's intricacies. These guidelines encompass design principles, printer selection, material utilization, and post-processing efficiency. While not universally exhaustive, they serve as robust decision-making tools, bridging the gap between theoretical knowledge and practical application. Contributions of this study extend to various industries, offering insights that foster additive manufacturing's effective integration. By addressing challenges specific to Design for Additive Manufacturing (DfAM) and Design for Additive Post-Processing (DfAPP), industries can navigate the complexities of additive manufacturing with greater precision. The guidelines proposed herein advocate a proactive approach, optimizing manufacturing practices for efficiency, cost-effectiveness, and product quality enhancement. Despite its contributions, this research acknowledges inherent limitations, including experimental constraints and the evolving nature of technology. Thus, future research avenues beckon, from advanced post-processing techniques to tailored guidelines for specific industries. As the additive manufacturing landscape evolves, ongoing research endeavors will be critical to ensuring that guidelines remain adaptive and effective. In essence, this dissertation encapsulates a journey through additive manufacturing's landscape, resulting in a set of guidelines that empower industries to leverage this transformative technology. As the manufacturing realm evolves into a digital era, characterized by precision and innovation, the guidelines proposed here stand as an essential compass, guiding industries toward optimal additive manufacturing practices.30 0Item Restricted EFL Learners’ Perceptions of AI Tools in Enhancing their Target Language Use: A Quantitative Study of Saudi International Students in the UK(Saudi Digital Library, 2023-12-05) Alsubaei, Mouneerah; Thompson, PaulThe integration of AI tools into education has gained considerable attention in this era of technological advancement. In light of the increasing interest in the use of AI in language learning, it is essential to understand how EFL learners perceive the potential of these tools to improve their usage of the target language. Despite the growing prevalence of AI tools, there is limited research concerning Saudi international students' perceptions of these tools in the context of language learning, specifically in the UK. This study attempts to address the consequential gap in the literature by investigating Saudi international students' perspectives on AI tools for improving EFL use, particularly in overall language performance and four skills, especially in the UK. Moreover, it attempts to identify significant differences in students' responses to the questionnaire based on gender, age, current educational courses and years of language learning experience. Data collection questionnaires were distributed to targeted respondents (N= 237) and analysed statistically by SPSS. This study revealed interesting findings. Participants generally believe AI improves overall language performance and writing skills, particularly grammar, spelling and punctuation. However, AI's effect on critical thinking, reading, listening and speaking could not be proven by the research findings, except in terms of vocabulary use. There is no significant difference between participants' responses based on gender, age, current academic courses or years of language learning experience. However, there is a significant difference based on current academic courses in relation to the number of times the research participants use AI tools to improve their English Language use. Most comments from students were positive and they expressed the effectiveness of AI in improving writing skills.172 0Item Restricted Examining Adversarial Examples as Defensive Approach Against Web Fingerprinting Attacks(Saudi Digital Library, 2023) Alzamil, Layla; Elahi, TariqIn the age of online surveillance, and the growth in privacy and security concerns for individuals activities over the internet. Tor browser is a widely used anonymisation network offering security and privacy-enhanced features to protect users online. However, web fingerprinting attacks (WF) have been a challenging threat that aims to deanonymise users browsing activities over Tor. This interdisciplinary project contributes to defending against WF attacks by employing the “attack-on-attack” approach, where Adversarial Examples (AEs) attacks are launched to exploit existing vulnerabilities in the neural network architecture. The FGSM and DeepFool construction methods are implemented to introduce perturbed data to these models and lead them to misclassify, significantly decreasing the classifier prediction accuracy.18 0