Sydney eScholarship
Not a member yet
31878 research outputs found
Sort by
Efficient Prompt Engineering for Large Foundation Models
No full textThe rapid advancements in large foundation models, particularly GPT-4V, have unlocked significant potential in various applications, including visual recognition tasks. However, the high computational and financial costs associated with GPT-4V’s inference remain substantial barriers to its widespread use. In response to these challenges, this thesis introduces ``Collage Prompting'', a novel and budget-friendly prompt engineering technique that concatenates multiple images into a single visual input. By allowing GPT-4V to process several images simultaneously, this approach not only reduces inference costs but also opens new avenues for more efficient utilization of large-scale models in real-world scenarios.
The thesis further investigates the influence of image arrangement within the collage prompt on recognition accuracy. We present a framework that uses a graph-based predictor to optimize the placement of images, improving the model’s performance by ensuring the most favorable configuration. To facilitate future research in this area, we introduce CollagePrompt, a comprehensive benchmark designed to evaluate the cost-effectiveness and recognition performance of collage prompts. This benchmark provides a platform for testing various image arrangements and includes a baseline optimization technique derived from genetic algorithms.
Through extensive experimentation across diverse datasets, we demonstrate that collage prompts with optimized image layouts significantly outperform randomly arranged ones in terms of both accuracy and cost-efficiency. Moreover, two key metrics are proposed to measure the effectiveness of different collage configurations. This research contributes to the emerging field of prompt engineering by offering a practical solution that enhances the economic viability of large foundation models like GPT-4V, without compromising their visual recognition capabilities
Employability preparations for children and young people with blindness or low vision in Australia.
No full textThis research investigated employment preparations for children and young people with blindness or low vision (BLV). Vision is our primary learning sense. Children and young people with BLV are unlikely to automatically learn concepts, knowledge and skills about employment because their ability to learn incidentally through vision is compromised. Without deliberate intervention to mitigate for the functional impacts of BLV, children and young people are at risk of not developing skills which would prepare them for future employment.
This research project comprises two studies. Study One used a qualitative descriptive research design to explore the current scenario for young people with BLV as they prepare to transition into employment. Three groups were recruited – young people (aged 15-20 years; n=9) with BLV; parents of young people with BLV (n=10); and vision support teachers (n=11). Data were collected via interview.
Study Two focused on the development and evaluation of the STAR Kit employability home program for children with BLV and their parents. Findings from Study One, and tenets from the Theory of Planned Behaviour (Ajzen, 2008) guided the STAR Kit’s design principles and inclusions. Ten children (5-11 years) with BLV and their parents participated in the study. A mixed methods approach was used including an experimental cross-over design and a qualitative descriptive design. Data were collected via interviews to understand the impact of the STAR Kit on beliefs and aspirations about future employment; investigate behavioural drivers for participating in employability-building tasks; and gather participants’ perspectives about the STAR Kit. Findings emphasised the importance of starting employment preparations at an early age; providing parents with resources to support their child or young person to prepare for future employment; designing programs to fit into family life; and taking measures to address family readiness for employability programs
2D and the Sacred: Exploring the Reorientation of Desire and Love
No full textThis thesis explores the potentiality for sacred, spiritual, and religious experiences of two-dimensional love; specifically, romantic love for fictional characters from Japanese media, such as anime, manga, and video games. Analysing the phenomenon in both Japanese and Western contexts illuminates the unique features of loving two-dimensional characters, demonstrating the sacred and religious qualities in each socio-cultural context, and probing how and why these individuals choose to commit themselves to fictional others. This thesis argues that this phenomenon is a profound and meaningful
endeavour which allows individuals to find purpose and meaning in their lives. Loving and seeking committed relationships with two-dimensional characters may have started as a response to unfavourable socio-economic conditions for some, but nowadays it is an enchanting and compelling choice for those disillusioned with reality. As such, these sacred, spiritual, and religious elements are evident within the beliefs and practices in which a human individual engages with two-dimensional
love. To demonstrate this, I explore the notions of moe and love: moe provides a variety of deep, profound encounters shaped by concepts of animism, transcendence, and salvation, and love has inherently religious and mythical qualities. This is also exemplified in the growing presence of the fictosexual movement, which has elevated two-dimensional love to be an intrinsic human desire - as legitimate as other romantic and sexual orientation, which encompass religious functions. This thesis opines that two-dimensional love is representative of broader trends in the evolution of, and alteration of, the experience of intimacy and the nature of human relationships. Moreover, it argues that the committed choice/practice of two-dimensional love is inherently sacred, using religious frameworks and spiritual qualities to respond and answer the ultimate concerns of one’s identity, and of meaning in life
Optimised Resource-Constrained and Heterogeneity-Aware Federated Edge Learning
No full textThe rapid integration of Internet of Things (IoT) technologies and smart devices at the network edge has driven a shift from centralized cloud architectures to edge computing to meet growing demands for efficient, low-latency, and secure data processing. Traditional centralized methods are increasingly unsustainable due to concerns over latency, bandwidth, privacy, and cost, prompting the rise of decentralized learning paradigms. Among them, Federated Edge Learning (FEL) has emerged as a key solution for highly distributed, resource-constrained edge environments. It enables multiple devices to collaboratively train a shared model without transmitting local data, enhancing privacy, scalability, and efficiency. This thesis proposes a series of FEL frameworks addressing key challenges in edge intelligence. Firstly, FeDEQ integrates consensus optimization with deep equilibrium learning via the alternating directions method of multipliers, significantly reducing communication and memory usage while maintaining strong performance, validated both theoretically and experimentally. Secondly, WAFL employs Wasserstein distributionally robust optimization to enhance model generalization under adversarial and non-i.i.d. data conditions, consistently outperforming baseline FL methods. Thirdly, FedKO combines Koopman operator theory and Reservoir Computing to process multivariate IoT time series data, enabling efficient, privacy-preserving anomaly detection through a bi-level optimization formulation. Finally, iREPO introduces an edge-friendly framework for aligning large language models via empirical preference optimization and implicit reward pairwise difference regression, achieving strong improvements in alignment benchmarks. Collectively, these contributions advance FEL by offering scalable, robust, and resource-efficient solutions, setting new directions for deploying machine learning in dynamic, heterogeneous, and resource-limited environments
A Code of Conduct for Best Practice in Nuclear Command, Control, and Communication (NC3): The Role of International Law and Lessons from Informal International Law-Making
No full textThis study examines the current situation regarding State nuclear command, control and communication (NC3). Nuclear command, control, and communication is the framework that governs the process that, as its ultimate end product, results in the launch of a nuclear weapon. This study explores how each nuclear weapons State structures its NC3 (if such structures are publicly known) and what international law governs nuclear operations. Given the lack of international law regarding NC3, this study explores whether it would be beneficial to adopt a code of conduct or other non-binding instrument outlining any lex lata or regarding nuclear weapons operations, including any best recommended practice regarding how States should structure their NC3 systems, given the strategic and policy objectives of NC3 systems. In proposing a model code of conduct, this study will examine the extant literature and practice on the adoption and implementation of non-binding instruments in the international law of armed conflict, to better understand how and why non-binding instruments gain traction in State practice. This study will then apply those conclusions to a draft code of conduct for State NC3
Characterisation and Prediction of Checkpoint Inhibitor Associated Autoimmune Diabetes Mellitus
No full textWith the expanding use of immune checkpoint inhibitors in cancer therapy, there is growing interest in characterising and predicting the severe immune related adverse events that can arise as a complication of this therapy. Checkpoint inhibitor associated autoimmune diabetes mellitus (CIADM) is one such immune related adverse event and has been reported in case series to date as a novel form of fulminant type 1 diabetes. This thesis aims to characterise the clinical and immune phenotype of CIADM to improve identification and management of CIADM and investigate potential avenues for prevention of CIADM.
In a multicentre case series (Chapter 2) the clinical phenotype of patients with CIADM is detailed, along with the key hallmarks of absolute insulin deficiency, rapid pancreatic volume loss, exocrine insufficiency and variable degree of type 1 diabetes autoantibody positivity. A systematic review (Chapter 3) was subsequently conducted which demonstrated what features were most prominent in CIADM to allow for formation of clinical diagnostic criteria. Finally in Chapter 6, clinical guidance is provided based on multidisciplinary consensus to improve the identification and care of patients treated with immune checkpoint inhibitors who develop hyperglycaemia on treatment.
Using longitudinal blood, serum samples and radiological data in a case-control study, potential biomarkers for CIADM are identified in Chapter 4. Immune cell subsets, pancreatic volume, and Type 1 diabetes autoantibodies at baseline were found to be predictive of CIADM with a ROC curve AUC of 0.96. Whilst data on CIADM in humans has largely been limited to systemic markers, using a mouse model of CIADM, the islet immune infiltrate and endocrine cells are further profiled in Chapter 5 and potential key local immune cell types in CIADM are identified. Together, these studies provide a foundation for future work in prediction and prevention of CIADM
High-Speed Implementations of Spectral Correlation Analysis Techniques
No full textSpectral correlation density (SCD) is widely used to characterize cyclostationary signals, but its high computational requirements pose significant challenges. Although the fast Fourier transform (FFT) enables efficient methods such as the FFT accumulation method (FAM) and strip spectral correlation analyzer (SSCA), their real-time adoption remains limited. Therefore, optimizing these methods for computational efficiency is essential to balance accuracy and efficiency. This work focuses on reducing wordlength, leveraging parallel hardware architectures, and minimizing computational complexity.
This study first analyzes the relationship between wordlength and signal-to-quantization-noise ratio (SQNR) in fixed-point SCD implementation, using a canonical FAM-based estimator with both fixed- and mixed-point arithmetic. High performance is achieved on the AMD/Xilinx Zynq UltraScale+ RFSoC ZCU111 platform by exploiting spatial parallelism, pipelining, I/O optimization, and algorithmic symmetry.
Next, an SSCA implementation is proposed for large datasets on the AMD/Xilinx Versal VCK5000 platform. It utilizes parallelism across the programmable logic (PL), double data rate memory controller (DDRMC), and AI engine (AIE), and combines very-long instruction word (VLIW) and single-instruction multiple-data (SIMD) architectures. The PL handles data transfer between the DDRMC and AIE, ensuring seamless communication. This architecture maximizes hardware efficiency and throughput for large-scale processing.
Lastly, the sparse strip spectral correlation analyzer (S3CA) is introduced, employing the sparse fast Fourier transform (SFFT) to leverage the sparsity of the cyclic spectrum in practical signals. It computes a sparse, downsampled channel-data product (CDP) and applies a modified SFFT to estimate the SCD efficiently. This approach reduces computation and enhances scalability, enabling real-time spectral analysis of large datasets
Growing up, ill: Uncovering the experiences of young people living with chronic illness
No full textChronic illness in young people is routinely underacknowledged and underreported. In general, pain and illness are represented as the exception rather than the rule for young people, however experiencing chronic illness when young is not rare. Rather, it is misunderstood, unrecognised, and hidden. This speaks to how the experience of chronic illness when young is entwined with social expectations, histories of discrimination and silencing, and current institutional and political failures. I examine this complexity in order to interrogate the ways in which current understandings of illness fall short, and consider what social, cultural, and healthcare changes are needed going forward.
This thesis is based on 33 in-depth interviews with adults aged 19-29 years who live with a range of diagnosed and un-diagnosed physical health conditions. The breadth of this participant criteria is an attempt to decentre institutional categorisations of disease and instead focus closely on how people construct and negotiate chronic illness in their lives. With reference to emancipatory sociological theory and critical disability studies, I seek to explore how individuals’ experiences of chronic illness interact with the social and institutional judgement and neglect they face.
I argue that the marginalisation experienced by my participants is often caused by persistent failures to listen to young people, the enduring stigmatisation of chronic illness, and a socio-cultural environment that does not accommodate uncertainty, complexity, and vulnerability. I further argue for an understanding of chronic living while young that better incorporates the shifting meanings and expectations of youth, the multivalent tensions around visibility and recognition, the ambivalences and potentials of online worlds, and the fluctuating salience of diagnosis. This provides new and adapted conceptual frameworks to better understand the complex experiences of young people living with chronic illness
Building Mentally Healthy Futures: A Call to Action
No full textTwo years ago, Australia’s Mental Health Think Tank (2022) released the policy paper Building Mentally Healthy Futures: Australian Youth Recovery Plan. The plan put forward eight evidence-based policy recommendations addressing three primary drivers of the disproportionate mental health burden on young Australians: economic inequalities, social disconnection and lack of access to quality mental health services. In light of the compounding effects of additional economic and social crises in recent years, Australia’s Mental Health Think Tank have identified the critical need to take stock of progress made against the original policy paper recommendations, re-examine the evidence, and issue a renewed call for policy action. As the 2025 federal election approaches, this timely policy report focuses on actionable, evidence-based steps a newly elected government can take to address the drivers of escalating mental health issues in the Australian population
Exploring Changing Treatment Paradigms in Medullary Thyroid Carcinoma, the Immune Milieu of the Tumour Microenvironment and a Novel Targeted Therapy for Advanced Disease
No full textThe immune milieu of the tumour microenvironment (TME) in Medullary Thyroid Cancer (MTC) and the potential role of immune therapies have not been extensively explored. This study aimed to define the current treatment landscape and changing management paradigms in MTC, describe the nature of the immune microenvironment of these tumours, and ultimately explore a novel targeted therapy for advanced disease.
Tumour-infiltrating lymphocytes (TILs) in the TME of MTC patients were assessed and correlated with clinicopathological prognostic variables and survival outcomes. All patients with MTC had low TILs (≤10%), and there was no significant association between TILs and local recurrence or disease-specific survival on multivariable analysis. These findings highlight that MTC is an immune-quiescent tumour.
A novel targeted therapy for advanced MTC was also investigated. The EDV™ (EnGeneIC Dream Vector) is a bacterially-derived construct loaded with cytotoxic drugs and conjugated with a bispecific antibody directed against specific overexpressed surface receptors on tumour cells. The EDV™ effectively killed human MTC cells in vitro and in a nude-mouse xenograft and syngeneic neuroendocrine tumour model. In addition to targeted delivery of the cytotoxin PNU-159682 mediated by antibody binding, EDV treatment triggered an innate and adaptive immune response against tumour cells, with upregulation of M1 macrophages, cytotoxic natural killer (NK) cells, and invariant natural killer T cells, followed by CD8 effector T cells. The shift to an immune-activated phenotype in the TME correlated with changes in the cytokine and chemokine profile, with upregulation of the key drivers of macrophage, NK cell and T cell activation and chemotaxis in the serum and TME. These results provide preclinical data demonstrating the efficacy of a novel targeted therapy for advanced MTC and form the evidence base to support a human clinical trial to confirm the translational relevance of the results