85000 research outputs found
Sort by
Robust Fairness in Spatial-Temporal Resource Allocation
No full textSmart cities aim to optimise city operations, enhance the quality of life, and minimise ecological footprints by integrating computer science, information, and communication technologies. The motivations behind developing smart cities include improved efficiency in public services to reduce costs and enhance service delivery. In this thesis, titled 'Robust Fairness in Spatial-Temporal Resource Allocation', we address the challenges in spatial-temporal resource allocation for various urban mobility and public transportation tasks.
Addressing robust fairness while optimising utility is vital in spatial-temporal resource allocation as unfairness potentially lead to negative economic (e.g., service deliverers with lower wages abandoning their jobs, leading to a shortage) and societal (e.g., unethical issues) consequences. Platforms must balance optimising utility with ensuring fairness, as focusing solely on fairness may inadvertently create a scenario where no one benefits, paradoxically resulting in a form of extreme fairness. The main challenges in strike a balance between robust fairness and utility in real-time optimisation platforms are i) the dynamic changing correlation between the two objectives, ii) the difficulties of achieving fairness increase along with the number of agents in the system, iii) the instability of the fairness achieved. Regarding these challenges we made several contributions in this thesis toward constructing a robust and multi-party fair spatial-temporal optimisation platform which will be essential towards constructing a smart city.
First, we introduced Multi-Level Fair Genetic Algorithm (MLFA) to tackle the challenge of increment complexity of promoting fairness among customers in spatial-temporal optimisation platforms (e.g., customers in service delivery systems, disaster locations in disaster management) along with the increment of the number of customers. The proposed approach incorporates a multi-level fairness scheme, with three distinctive characteristics: i) targeting fairness with a hierarchical structure to reduce the complexity of the problem, ii) incorporating the proposed multi-level fairness scheme into a Genetic Algorithm-based matching algorithm to allow the algorithm to consider fairness and utility at the same time, and iii) allowing the algorithm to adjust the weight of fairness by adjusting the number of fair iterations run in the algorithm.
Second, we take a further step toward constructing a robust and multi-party fair spatial-temporal optimisation platform. We introduced Two-sided Fairness-aware Genetic Algorithm (2FairGA), which expands the genetic algorithm from the original objective solely focusing on utility to multiple objectives that incorporate two-sided fairness. Most existing studies focusing on promoting fairness in a spatial-temporal optimisation system focus on improving fairness from a single side. There has been limited research addressing two-sided fairness that simultaneously considers the needs of both parties in an optimisation system—service deliverers and customers—along with utility optimisation. Subsequently, the impact of injecting two fairness definitions into the utility-focused model and the correlation between any pair of the three objectives are explored.
Third, we put the focus onto improving robustness in a fair spatial-temporal optimisation platform. We introduced a dynamic Markov Decision Process (MDP)-based model to address long-term fairness among agents. Existing approaches to promote fairness in spatial-temporal optimisation fall short in maintaining fairness in a long time horizon. The challenges in maintaining long-term fairness are due to i) short sight, ii) concept drift, and iii) disparity between utility and fairness increase across time. To address these challenges, we propose a dynamic Markov Decision Process (MDP)-based model that incorporates a prediction module to foresee the future requests (e.g., future demands change in a delivery service platforms or frequency of disasters at areas) and a scalarisation function to dynamically balance between efficiency and fairness.
Fourth, recognising that event prediction plays a crucial role in our third proposed approach, we introduced STEMO - a multi-objective Reinforcement Learning-based approach towards building an adaptive early spatial-temporal predictor. Early prediction allows for more proactive decisions in resource allocation. By anticipating events or demands before they occur, a system can prepare and respond more effectively (e.g., allow the MDP-based scheduling platform change decision making pattern early by modifying the transition probabilities). This capability not only forecasts demands but also identifies shifts in patterns that could impact fairness, allowing for timely and equitable adjustments. With the conflict between timeliness and accuracy in mind, the adaptive characteristic in STEMO guarantees accurate predictions for different events due to their spatial-temporal characteristics. We proposed the method to address the three primary challenges in adaptive early forecasting: i) balancing robustness and accuracy due to their often negative correlation, ii) varying negative correlations based on spatial-temporal characteristics of events, and iii) spatial-temporal data alignment.
Putting all the steps together, we eventually contributed towards building future smart cities in terms of urban mobility and public transportation. To evaluate the applicability of the proposed methods, we adapted the models into scenarios such as ridesharing, officer patrolling problems, and dynamic arc routing problems.</p
Development and Enhancement of Data-Driven Methods for Fuel Loss Detection Using Deep Learning
No full textThis thesis focuses on developing and improving data-driven methods for detecting fuel loss issues, leveraging Deep Learning (DL) to automatically extract features from complex sequential data. Ensuring the integrity of petroleum product storage at service stations is important, as fuel loss may pose significant economic and environmental consequences at service stations. Despite the growing deployment of sensors at service stations that provide abundant data, these resources remain underutilised in advancing data-driven detection methods. These methods identify fuel loss issues by actively monitoring sensor data, offering the advantage of remote online operation and reduced costs. Manual analysis of large datasets is impractical, especially with challenges such as noise in real-world data and variability in tank conditions. Automated tools for feature extraction are therefore essential for enhancing efficiency and accuracy. Machine Learning (ML), particularly DL, is a key technology to overcome the above challenges and revolutionise fuel loss detection techniques, given its success in extracting valuable insights from complex and large-scale data.
The first contribution of this thesis is the development of the first data-driven method for meter error detection. Meter error is a critical issue in fuel loss, yet its detection traditionally relies on infrequent physical inspections, which can leave issues undetected for long periods. To overcome this, we propose a framework that uses data available at service stations to enable offsite detection that can be performed at any time, eliminating the need for onsite visits. Change Point Detection (CPD) is applied to analyse the sequences of key indicators related to meter error, with Long Short-Term Memory-Variational Autoencoder (LSTM-VAE) employed for representation learning to facilitate accurate detection. Experiments show the framework’s effectiveness in detecting meter errors and its potential for broader CPD applications. Overall, the proposed approach offers a more efficient and timely solution for identifying meter errors.
The second part of this research focuses on developing a real-time fuel leakage detection method. Fuel leakage is a significant concern for site operators, desiring early detection to minimise its consequent impacts. However, existing data-driven studies perform tests based on daily data collected over weeks, leading to detection delays. To address this, we introduce a novel memory-based online CPD method that continuously monitors and processes real-time fuel variance data for early detection. This method maintains a size-constrained memory of representative historical data and supports both statistical techniques and ML methods to measure shifts in fuel variance data over time. Experiments with simulated fuel leakage data and benchmark CPD datasets demonstrate the method's effectiveness in detection accuracy. It also achieves a much shorter detection delay compared to the general turnaround time of fuel leakage detection in practice. In summary, this real-time approach offers a more responsive solution for detecting fuel leaks, enhancing operational efficiency and safety.
The final contribution of this thesis is the development of an explainable fuel leakage detection method. Apart from real-time detection, explainability is another major limitation in current data-driven fuel leakage detection studies. False alarms are common in practice, so providing practitioners with explanations or interpretations can greatly assist in their result validation, especially when the systems are built with complex algorithms whose decision-making processes are opaque. Explainability helps build user trust and aids in verifying the system’s reliability. Therefore, we propose the first explainable fuel leakage detection approach in the field, which combines a high-performance DL model for accurate online detection with a transparent ML model with great interpretability to generate clear textual explanations. Through case studies, we demonstrate the method’s ability to generate intuitive explanations using linguistic terms to describe the changes in features indicative of fuel leakage. The approach also maintains accurate early detection, as shown through experiments with real-world fuel data with induced leakage. This work highlights the importance of ensuring both accuracy and explainability to improve fuel leakage detection and foster user confidence.</p
Algorithmic Price Personalisation and Consumer Protection in Australia
No full textThe Australian Consumer Law was designed to protect consumers when
conduct in trade or commerce was primarily made by human actors.
However, as the world of business and consumer interactions undergoes a
transformative phase, new phenomena have emerged. One such
transformation is the widespread adoption of algorithmic price
personalisation in business, where pricing algorithms are used to determine
individualised prices based on consumer willingness to pay. Powered by
emerging technologies such as big data, machine learning and artificial
intelligence, algorithmic price personalisation has gained prominence in the
current business landscape, raising various concerns about consumer
protection. This article examines whether the key provisions of the Australian
Consumer Law — particularly the provisions on misleading or deceptive
conduct, unconscionable conduct, and unfair contract terms — are capable
of safeguarding consumers amidst the rise of personalised pricing
algorithms. This article finds that while the Australian Consumer Law can
provide a basic protection to consumers, various challenges exist and need
to be addressed to ensure consumer rights in the face of algorithmic price
personalisation.</p
A national survey on nature connection: Infographic summary
No full textThese infographics summarise the key findings of our national survey of >4000 Australians from diverse backgrounds to understand their nature connection and how it relates to wellbeing and pro-environmental behaviours. We summarise the most connected categories of Australians, the barriers to connecting with nature, and the dimensions of nature connection. We find that people with the highest levels of nature connection are 64 times more likely to undertake pro-environmental behaviours than people with the lowest levels of connection. Similarly, people with the highest levels of nature connection are 4.3 times more likely to have the highest levels of life satisfaction (a measure of wellbeing) compared to people with the lowest levels of connection and that nature connectedness is 8 times more important for life satisfaction than socioeconomic status.</p
Heterojunction of Bi2Se3 and Epitaxially Grown GaN Nanostructures on Oxide-Based Substrates for Self-Powered Broadband Photodetectors
No full textOver the past decades, in addition to conventional Si-based devices, significant efforts have been made to explore stable compound semiconductor materials (SiC, III-V compounds, and various oxides) and their applications. Among these, GaN-based devices have been widely adopted and commercialized successfully, as GaN possesses chemical inertness, high electron mobility, a wide energy band gap (3.4 eV), etc. However, previous research on GaN UV photodetectors has primarily focused on GaN grown in the polar direction, which faces performance limitations due to spontaneous and piezoelectric fields. Further, while significant progress has been made in examining the photodetection properties of GaN grown on conventional dielectric materials like SiO2 and Al2O3, there remains an unexplored area in examining these properties on unconventional oxide substrates such as MgO, LiAlO3, SrTiO3 (STO), etc. Due to their incompatibility at high GaN growth temperature (800-1000 ℃) required by conventional techniques (MOCVD, MBE, HVPE, etc.). Further, the optical radiation detection range of GaN is confined to the ultraviolet region. For practical optoelectronic applications, photodetectors capable of detecting light across a wide wavelength range (300-1100 nm) with autonomous light-detection capabilities are essential.
To address these issues, growing GaN along semi-polar and non-polar directions could improve the performance of GaN-based photodetector devices. Further, the photodetection properties of GaN grown on other lattice-matched oxide substrates can be explored using the LMBE technique, which can grow GaN at relatively low temperatures. Further, heterojunction semiconductors comprising a GaN in conjunction with a narrow bandgap material play a crucial role in developing self-powered multi-wavelength photodetectors. Recent advances in topological insulators offer promising prospects for quantum, electronic, and optoelectronic devices. Integrating these materials with GaN nanostructures paves the way for high-efficiency innovations. That spans nearly the entire spectrum of interest in photodetector exploration.
Laying a foundation, this study first addresses the growth of GaN nanostructure on various plane orientations of sapphire using the LMBE technique. Further, we demonstrated the photodetection capability of single crystalline epitaxial GaN grown on STO at ≤ 600℃ using the LMBE technique. On the other side of the coin, we aim to make the heterojunction of the Bi2Se3 a topological insulator with LMBE-grown GaN nanostructures on sapphire and STO to fabricate highly responsive self-powered UV-Vis-NIR broadband photodetectors.</p
Ultrasound‐stimulated microbubbles to enhance radiotherapy: A scoping review
No full textINTRODUCTION: Primarily used as ultrasound contrast agents, microbubbles have recently emerged as a versatile therapeutic vector that can be 'burst' to deliver payloads in the presence of suitably optimised ultrasound fields. Ultrasound-stimulated microbubbles (USMB) have recently demonstrated improvements in treatment outcomes across a variety of clinical applications. This scoping review investigates whether this potential translates into the context of radiation therapy by evaluating the application of this technology across all three phases of radiation action. METHODS: Primary research articles, excluding poster presentations and conference proceedings, were identified through systematic searches of the PubMed NCBI/Medline, Embase/OVID, Web of Science and CINAHL/EBSCOhost databases, with additional articles identified via manual Google Scholar searching. Articles were dual screened for inclusion using the Covidence systematic review platform and classified against all three phases of radiation action. RESULTS: Overall, 57 eligible publications from a total of 1389 identified articles were included in the review, with studies dating back to 2012. Study heterogeneity prevented formal statistical analysis; however, most articles reported improved outcomes using USMB in the presence of radiation compared to that of radiation alone. These improvements appear to result from the use of USMB as either a biovascular disruptor causing tumour cell damage via indirect mechanisms, or as a localised treatment vector that directly increases tumour cell uptake of other therapeutic and physical agents designed to enhance radiation action. CONCLUSIONS: USMB demonstrate exciting potential to enhance the effects of radiation treatments due to their versatility and capacity to target all three phases of radiation action.</p
A Configurational Approach to CSP Selection and Rejection
No full textSelecting the appropriate cloud service provider (CSP) is crucial for organizations, significantly impacting business performance and growth. However, the multitude of available providers can make this decision daunting. Existing studies focus on technical and operational CSP attributes but often overlook how these attributes should be configured, especially their interdependencies. To address this gap, we explore parsimonious configurations for CSP acceptance and rejection. Using a configurational approach and fuzzy-set qualitative comparative analysis (fsQCA), we uncover complex nonlinear relationships among key attributes. The fsQCA provides the combinations of causal recipes associated with the acceptance and rejection of a CSP, supporting the conjunction, equifinality, and asymmetry perspectives. Our results reveal that no single attribute is pivotal; instead, four configurations predict CSP selection, while five foresee rejection. Notably, we identified configurations tailored for small vs. medium-sized enterprises. This study enriches both theory and practical approaches in CSP selection, offering new insights for choosing CSPs.</p
Composition for Mnemosyne
No full textBackground: This practice-led research investigates how synthetic systems mediate perception and embodiment. It engages debates around algorithmic visual culture and its entanglements with social systems, drawing on Hito Steyerl (Duty Free Art), Trevor Paglen (Clouds), and Holly Herndon (PROTO). It contributes to critical moving-image practice by questioning how AI reshapes aesthetic and social relations. Contribution: Developed within Newcastle’s layered technological and military landscape, Composition for Mnemosyne engaged local youth and sites to explore how emerging technologies reshape both media and the body as sites of mediation. Created during The Lock-Up’s Artist in Residence program, the work comprises a two-channel video installation with five accompanying wall works. The first channel documents The Hunter Singers youth choir interpreting an AI-assisted score through voice and gesture, transforming algorithmic patterns into an embodied choral experience. The second channel follows teenagers moving through Newcastle’s technological and military environments — including disused defence sites and the F-35 fighter jet base — with their movements interwoven with AI-generated imagery, creating a speculative layer on how military, technological, and synthetic systems shape contemporary experience. Significance: Exhibited as part of DATA MINDS from November 2024 to February 2025 and supported by Creative Victoria’s Creators Fund, the work contributed to current debates on AI and digital culture within screen and digital media arts. Shown alongside leading practitioners including Jon Rafman, Brie Trenerry, and Roy Ananda, its collaborative approach broadened audience engagement and encouraged dialogue on technology, community, and culture. The work was reviewed in Memo Review and selected for further presentations at Hayden's Gallery (VIC) and in Dublin (aemi and Pallas Projects).</p
A leadless power transfer and wireless telemetry solutions for an endovascular electrocorticography
No full textOBJECTIVE: Endovascular brain-computer interfaces (eBCIs) offer a minimally invasive way to connect the brain to external devices, merging neuroscience, engineering, and medical technology. Currently, solutions for endovascular electrocorticography (ECoG) include a stent in the brain with sensing electrodes, a chest implant to accommodate electronic components to provide power and data telemetry, and a long (tens of centimeters) cable travel through vessels with a set of wires in between. Removing this long cable is the key to the clinical viability of eBCIS as it carries risks and limitations, especially for patients with fragile vasculature. APPROACH: This work introduces a wireless and leadless telemetry and power transfer solution for ECoG. The proposed solution includes an optical telemetry module and a focused ultrasound (FUS) power transfer system. The proposed system can be miniaturised to fit in an endovascular stent, removing the need for long, intrusive cables. MAIN RESULTS: The optical telemetry achieves data transmission speeds of over 2 Mbit/s, capable of supporting 41 ECoG channels at a 2 kHz sampling rate with 24-bit resolution. The FUS power transfer system delivers up to 10 mW of power to the implant through the scalp(6mm), skull(10mm), and subdural space(5mm), adhering to safety limits. Testing on bovine tissue (10 mm thick bone, 7 mm thick skin) confirmed the system's efficacy. SIGNIFICANCE: This leadless and wireless solution eliminates the need for long cables and auxiliary implants, potentially reducing complications and enhancing the clinical applicability of eBCIs. The proposed system represents a step forward in enabling safer and more effective ECoG for a broader range of patients.</p
The perceived impact of the agile development and project management method scrum on information systems and software development productivity
No full textThis research contributes to the body of knowledge in information systems development (ISD) with an empirical investigation in form of a case study that demonstrates the positive impact of the agile development and project management method Scrum on information systems and software development productivity and it provides a useful operationalization of the concept through seven identified indicators for productivity. Despite the fact that the case unit had challenges with the use of Scrum, the indicators identified the areas where the company had managed to exploit the potential of Scrum and its practices with regard to increasing productivity. The research results are discussed both with regard to the existing Scrum literature as well as to complex adaptive systems (CAS) as a foundation for ISD and agile development.</p