69238 research outputs found
Sort by
Examining the Structure of Directed Motivational Currents (DMCs) Among Secondary and Tertiary English as a Second Language Learners
No full textMotivation remains a central concern in second language (L2) and English as a foreign language (EFL) education, yet its underlying mechanisms are insufficiently understood. This study employs the theory of Directed Motivational Currents (DMCs) to explore periods of intense, sustained L2 motivation among Chinese adolescent EFL learners across secondary and tertiary levels. Through in-depth interviews with ten participants, this research identified the conditions (e.g., collaborative peer dynamics, vivid goal visualization) that triggered their DMC experiences. The data also highlighted how facilitative elements—such as clear starting points, personalized goal alignment, behavioral routines, and timely feedback—played a crucial role in initiating and sustaining these motivational currents. These findings contribute to DMC theory by revealing how intrinsic and extrinsic factors jointly foster and maintain high levels of motivation over time, offering valuable insights for designing targeted interventions to enhance EFL motivation and learning among Chinese adolescents
The impact of first and second formant variations on vowel identification among elderly Japanese listeners
No full textCorrect vowel identification is important for effective speech communication. While vowel perception largely requires accurate processing and detection of spectral information, spectral processing abilities (suprathreshold deficits such as difference limens for frequency) have been shown to degrade with age, and therefore potentially affecting elderly listeners’ vowel categorisation. The current study examined a group of near-normal hearing elderly listeners matched in age, hearing thresholds, auditory filter bandwidths and cognitive performance on their difference limens for frequency (DLFs) and vowel identification. Japanese has a five-vowel system, where/o/ and/e/ approximately differ in second formant (F2) frequencies and/a/ and/u/ approximately differ in first formant (F1) frequencies. We created two continua differing in F1 and F2, respectively, and examined elderly listeners’ identification of the vowels. We found differences in DLFs within the elderly listeners’ group, where the group with lower difference limens (LDL) to have similar DLFs to a control group of young listeners. Grouping the listeners according to their DLFs via k-means clustering with the young listeners’ results, we found the two groups to differ in their/o/ -/e/ (F2) vowel perception, where we observed a significant shift in the boundary between the two groups, but not when F1 was manipulated. This suggests that for elderly listeners, vowel identification that depends on higher formant cues may be affected by frequency discrimination abilities. However, the results did not suggest a difference in accuracy, but instead, a difference in how categories are assigned
Liquidity Commonality in Cryptocurrencies
No full textThis study investigates the cryptocurrency market’s microstructure, focusing on liquidity, liquidity commonality, and seasonality patterns. We find an inverted U-shaped pattern in the daily bid–ask spread, trading volume, and volatility within a week. Furthermore, a strong liquidity commonality exists across cryptocurrencies, indicating co-movement of a coin’s liquidity with the overall market liquidity. Moreover, the liquidity commonality peaks midweek and declines over the weekend. These seasonal patterns persist for the largest cryptocurrency after controlling for additional liquidity determinants
Generalised Defences Against Adversarial Attacks
No full textMachine learning (ML) models have revolutionised numerous fields, such as healthcare, finance, and transportation. However, this success comes with a cost, as concerns have emerged about the vulnerabilities of systems that rely on ML models. Adversarial attacks, where malicious actors intentionally manipulate input data to cause misclassification or system failure, have become a significant threat to the security and reliability of ML systems. Therefore, it is essential to understand the underlying mechanisms of these vulnerabilities and to develop effective mitigation strategies.
This thesis investigates the vulnerabilities of ML models to adversarial attacks, with a primary focus on the vulnerabilities in classification tasks. By analysing the vulnerabilities at both training and inference time, this study aims to identify effective strategies for minimising the risks associated with adversarial attacks. Furthermore, we demonstrate that the insights gained from adversarial learning and vice versa, such as Quantitative Structure-Activity Relationship (QSAR) modelling in cheminformatics.
To better understand the motivations behind data poisoning attacks during training, this thesis proposes a novel attack and detection framework that are based on the attackers' objectives. Previous defences against these attacks have assumed the availability of a clean validation set, which is difficult to obtain when the training and validation sets come from the same source. To address this challenge, this thesis presents a detection framework to identify whether a dataset has been poisoned without access to clean validation data. The proposed framework directly targets the attackers' goal to cause misclassification during inference time. We train a meta-learner on Complexity Measures derived from synthetic datasets and use it to estimate the model's performance on clean data without a clean validation set. Our framework then detects whether the training data has been poisoned by comparing this predicted clean data accuracy to the observed accuracy of the training data.
One major challenge in detecting adversarial attacks at inference time is the lack of information about potential unknown attacks. Attackers can bypass defence by reverse engineering it once they know the mechanism of the defence. Rather than focusing on attacks, this thesis proposes a robust triple-stage data-driven framework to detect adversarial attacks at inference time by focusing on what is known, the training data. Inspired by the concept of the Applicability Domain in cheminformatics, this framework defines a domain based on training data and rejects any unknown samples that fall outside the domain. We demonstrate the effectiveness of this framework under the worst scenario, where the attacker has full access to both the model and the defence, by designing an adaptive white-box targeted attack that uses a heuristic approach to actively try to bypass each stage of the framework.
Inspired by the success of adopting the Applicability Domain in adversarial detection, we hypothesise that there is a connection between adversarial detection in adversarial learning and Applicability Domain in cheminformatics. To demonstrate that our hypothesis holds, we conduct extensive experiments to benchmark a wide range of adversarial detection and Applicability Domain methods. Our experiments cover both directions, including using adversarial detection methods as Applicability Domain measures, and applying Applicability Domain methods to detect adversarial examples.
Overall, this thesis aims to advance both the field of adversarial learning and cheminformatics by providing a deeper understanding of the vulnerabilities and limitations in ML models, as well as proposing effective strategies for mitigating these risks. By demonstrating cross-domain applications, we hope that this research will inspire future studies to build on our findings and contribute to the development of robust ML models
Nature-Inspired Delivery Systems for Enhancing the Biocidal Efficacy of Bacteriophages
No full textThesis embargoed until 07/202
Free-Space Continuous Variable Quantum Key Distribution: Towards a Global Quantum Network
No full textThe aim of this thesis is to develop continuous variable quantum key distribution (CVQKD) for the goal of securing a global quantum network.
A numerical tool was developed for comparing different CVQKD protocols and identifying parameter spaces for positive and negative secret key rate (SKR) while taking into account the simultaneous effects of CVQKD parameters. Using the transmittance, excess noise, and amplitude modulation, the tool shows that the M-QAM protocol outperforms M-APSK which outperforms M-PSK. In addition, increasing the number of coherent states for a protocol non-linearly increases the capability of a protocol to produce positive SKRs. The tool also identifies parameters spaces, such as small excess noise regimes, where protocols have the same performance. The tool helps identify appropriate protocols for channels where parameters such as the transmittance and excess noise fluctuate, and can be extended to include more CVQKD parameters.
The improvement of CVQKD in turbulent channels, through the use of adaptive optics (AO), was studied by investigating the interferometric visibility between a quantum signal and a local local oscillator in turbulent channels. The results showed that the adaptive optics system was able to increase the interferometric visibility and decrease its fluctuations. In the context of free-space CVQKD, these lead to larger positive and more precise SKRs.
The structures of existing QKD networks were analysed and shown how they can be fit into a general three-layer QKD network architecture for the endeavour of a global QKD network architecture. The asymptotic and finite size limit SKRs were calculated for different links CVQKD can be used. The results show that CVQKD achieves longer distances in inter-satellite, satellite-to-ground, fibre, and underwater links in order. In addition, the Gaussian modulated (GM) CVQKD protocol with multidimensional (MD) reconciliation outperforms all other protocols studied by producing positive SKRs for longer link distances. The link capacity, a routing metric for secret key distribution, which considers a dynamic SKR based on dynamic links was developed. Its feasibility in simulated CVQKD networks was presented for the aim of spatiotemporal secret key distribution through a CVQKD network
Assessment of Financial Management Strategies and Challenges Faced by Food Service Industries in Nsukka Urban
No full textThis study assessed the financial management strategies adopted by food service industries and the challenges they face in Nsukka Urban, Enugu State, Nigeria. A descriptive-comparative research design was employed to answer three research questions based on data collected from a sample of 250 food service industry staff. Two null hypotheses were tested at p < 0.05 level of significance. A structured questionnaire, validated by three experts with a reliability index of 0.79, was used for data collection. Descriptive statistics such as mean and standard deviation were used to analyse the data on the research questions while ANOVA was used to test the hypotheses. Findings revealed that the key financial management strategies adopted by the food service industry included cost management, revenue enhancement, budgeting and risk management. However, investment strategies like technology adoption and securing loans were not prioritised. Major challenges they faced were rising costs of raw materials, lack of access to loan facilities, lack of skilled labour, improper money management and fluctuating prices. The solutions to these challenges included improving energy efficiency, use of technology, monitoring cash flows and employee training among other things. The study concludes that the adoption of specific financial management strategy is dependent on the type of food service industry and their profit margins are influenced by the challenges they faced. Recommendations include prioritising energy-efficient practices, leveraging technology for invoicing, and fostering employee development. Policymakers should also facilitate access to loans and training programs to enhance sector resilience. These insights offer practical value to business owners, researchers, and stakeholders in optimising financial sustainability within Nigeria’s food service industry
Cost-Effective Miniaturised Sensor System for Sewer Flow and Seepage Detection
No full textThesis embargoed until 07/202
Sexism and Sex: Illuminating Possible Paths Between Benevolent Sexism and Heterosexual Couples’ Sexual Wellbeing
No full textMy thesis presents three novel dyadic studies investigating the associations between women’s and men’s benevolent sexism and sexual wellbeing. In Chapter One, I review literature that demonstrates sex is an important domain of intimate relationships and is crucial for women’s and men’s wellbeing. However, little research has addressed how benevolent sexism, an attitude that romanticises traditional gender roles, may impact sexual relationships between women and men. In Chapter Two, I draw on literature that demonstrates how benevolent sexism shapes behaviours and expectations within close relationships generally to develop a case for why benevolent sexism will likely be costly to women’s and men’s sexual wellbeing. I then present two initial dyadic studies that provide evidence that women’s benevolent sexism is associated with their men partner’s poorer sexual wellbeing. However, I find null associations between (1) women’s benevolent sexism and women’s own sexual wellbeing and (2) men’s benevolent sexism and men’s own and their women partner’s sexual wellbeing. In Chapter Three, I suggest that the mix of negative and null associations found in Studies 1 and 2 may be due to three key mediators: sexual destiny or growth beliefs, sexual entitlement, and sexual communal strength. I explore why women’s and men’s benevolent sexism may be associated with these constructs and whether they may present costs or benefits to actors’ and partners’ sexual wellbeing. In Study 3, I find that women’s benevolent sexism was associated with higher sexual destiny beliefs and sexual entitlement, although these variables did not explain the links between women’s benevolent sexism and men’s sexual wellbeing found in Studies 1 and 2 which were not replicated in Study 3. Moreover, two new associations emerged linking men’s benevolent sexism to men’s and women partner’s higher sexual wellbeing. However, although men’s benevolent sexism was associated with higher sexual destiny beliefs, sexual entitlement, and sexual communal strength, none of these variables mediated the association between men’s benevolent sexism and sexual wellbeing. Overall, my thesis generates insights into associations between women’s and men’s benevolent sexism and their sexual attitudes and expectations. While some of these attitudes or expectations are often associated with poorer sexual wellbeing, it is likely that benevolent sexism generates a more complex mix of costs and benefits, thereby explaining the mix of negative, positive, and null associations with overall evaluations of sexual wellbeing found across the studies presented. These findings underscore the need for future research to identify specific contexts that present greater risks to wellbeing in order to understand and ultimately preserve important intimate relationships
Discovering Angkor: Henri Mouhot and the European Rediscovery of Angkor Wat
No full textIn 1858, Henri Mouhot became the first “white man” to “find” Angkor Wat. Despite being widely credited with the rediscovery of Angkor Wat, Mouhot’s posthumous journal, published in 1864, was not the first description of Angkor. In fact, I have found ample pre-Mouhot correspondence with the Royal Geographical Society detailing Angkor. Moreover, texts developed from French sources, such as John Bowring’s famous account of Cambodia and Siam, also detail Angkor. Needless to say, Angkor was certainly known prior to its “rediscovery”— yet it is also true that in the modern era, we appear to have internalised his brother Charles’s wrongful proclamation of discovery.
Because of this myth, Mouhot (in the vein of Captain Cook) has been defined by historians as the ‘typical’ colonial explorer. In this thesis, I intend to move beyond these stereotyped interpretations and reanalyse Mouhot’s travelogue and the associated literature, such as letters and newspapers. In combination with other scholars’ readings of Mouhot, I will develop an account of how Mouhot both fulfils and contradicts our understanding of the nineteenth-century explorer. By examining how knowledge was produced, reproduced, and presented, I intend to show how the travelogue offers compelling evidence that offers mutual commentary on both past and present understandings of discovery