85000 research outputs found
Sort by
Target-Driven Data Curation for Tabular Data Lakes: A Three-Stage Pipeline for Discovery, Assemblage, and Selection
No full textIn the era of data-centric AI, transforming raw data into high-quality, task-aligned datasets has become increasingly critical for enabling effective downstream applications. While significant advances have been made in model design, much less progress has been achieved in systematically optimizing the data itself, particularly in large-scale, heterogeneous data lakes dominated by structured tabular data. This thesis addresses this gap by introducing a principled, target-driven data curation pipeline that comprises three interrelated stages: datasets discovery, dataset collection assemblage, and data points selection. Each stage is guided by a distinct optimization target and contributes to building curated datasets that better serve downstream tasks. The target of the first stage is to retrieve the most relevant datasets from a large data lake in response to coarse-grained user queries, such as keywords, exemplar tables, or natural language (NL) statements. Existing approaches are often limited to a single query modality and intent type and require extensive labeled supervision. To address these challenges, we propose a novel cross-modal graph learning framework that constructs a heterogeneous graph of tables and NL statements, connected via context-aware edges. Through dual-view neighbor aggregation and a joint optimization objective, our method learns robust, semantically aligned representations without relying on largescale labeled data. Experiments on real-world datasets, covering short texts, questions, and claims, demonstrate our method’s strong performance across both NL- and table-based queries, as well as its benefits for downstream tasks such as multi-table question answering and table-based fact verification.
The target of the second stage is to assemble a subset of the discovered datasets that maximizes distinctiveness, defined as the total number of distinct tuples covered, under a user-specified query set and budget constraint. We prove this objective is NP-hard. While a greedy algorithm offers a theoretical approximation guarantee, it incurs prohibitive computation costs due
to repeated scans over large datasets. To overcome this, we develop a machine learning-based estimator for marginal distinctiveness gain and formulate a novel multi-dataset-query cardinality estimation task, significantly extending prior work on single-query cardinality estimation. Extensive experiments show that our proposed solution outperforms all relevant baselines in both effectiveness and scalability. Case studies on downstream ML tasks further demonstrate its potential to identify more useful datasets and enhance predictive performance. The target of the third stage is to select the most informative data points from the assembled dataset to maximize model performance. Prior methods suffer from inefficiencies caused by repeated model retraining and reduced selection quality due to rigid criteria and sampling constraints. To address these issues, we adopt an online learning approach for incremental model updates and design an adaptive scoring function that dynamically balances exploration and exploitation based on selection history. We further improve effectiveness by removing the single-cluster sampling bias and enabling multicluster
batch selection. Experiments demonstrate that our method achieves
superior effectiveness and efficiency compared to state-of-the-art baselines.
Together, these three target-driven stages form a robust and scalable
pipeline for structured data curation in data-centric AI. This thesis contributes
novel problem formulations, principled and efficient methods, and
extensive empirical validations across multiple datasets and downstream
tasks. It lays a strong foundation for future research on end-to-end data
curation pipelines, cross-modal representation learning, and responsible data-centric AI at scale.</p
Interactions Between Lactose-Protein-Mineral and their Implication on the Production of Dairy Products
No full textMilk and dairy products are vital components of the daily diet, providing high-quality proteins, lactose, fats, and numerous essential nutrients. However, their relatively short shelf-life presents challenges in distribution and storage. Dairy powders, characterised by varying protein concentrations and optional lactose and fat content, are widely utilised in the food industry due to their extended shelf-life, ease to transport, handling, processing, and incorporation into product formulations. To meet the growing demand for stable, functional, and high-protein dairy ingredients, innovative processing approaches such as non-thermal technologies and calcium fortification have gained significant interest. This research aimed to investigate the effects of low-frequency ultrasound and calcium addition on milk protein-lactose systems, focusing specifically on their physicochemical, structural, thermal, and
functional properties during processing and drying.</p
Milk Powder Formulations with Varying Casein to Whey Ratios and Calcium Addition: Physico-Chemical and Structural Properties and the Effect of Low-Frequency Ultrasound
No full textThis study examined the effect of low-frequency ultrasound (20 kHz, 1 and 5 min) on the physiochemical and structural properties of milk powder formulations with varying casein to whey ratios (0:100, 60:40, and 50:50) and calcium addition (30 mM). The ultrasound treatment led to changes in particle size, with an initial increase in aggregation followed by fragmentation. Calcium addition resulted in looser packing, as evidenced by a decrease in both bulk and tapped densities. DSC analysis indicated that calcium addition stabilized the protein–lactose matrix by increasing the glass transition temperature and reducing the number of thermal events. FTIR analysis revealed structural changes in proteins, with a decrease in β-sheet and β-turn and an increase in α-helix structures. These findings suggest that calcium plays a crucial role in reinforcing the structural integrity of the protein–lactose matrix, while ultrasound-induced mechanical forces lead to dynamic changes in particle size and protein conformation.</p
A generative BIM approach for DfMA integration into building facade design
No full textThe application of Design for Manufacturing and Assembly (DfMA) in facade design faces barriers due to fragmentation between design and manufacturing teams and limitations of traditional modelling. This paper develops an automated generative design workflow that integrates DfMA principles into facade cladding system design using Building Information Modelling (BIM) and visual programming tools. The methodology uses multi-objective optimization via genetic algorithms to minimize incomplete panels, reduce material waste, and enhance manufacturing efficiency while preserving design intent. A medical centre case study in Melbourne demonstrates the workflow’s application across wall sections with varying orientations and geometries. The generative approach identifies optimal 1.85 × 2.5 m panel configurations, achieving 38.17% waste reduction in vertical orientation versus 57.09% in horizontal, and outperforming manual panelisation by up to 18.92% in material efficiency. The workflow enables real-time collaboration between designers and fabricators through parametric feedback loops, allowing fabrication constraints to inform design decisions during concept development. Analysis of 48 design alternatives reveals that systematic algorithmic optimization improves panelisation efficiency and identifies universal optimization principles for facade systems. This research presents a working prototype that integrates DfMA into standard BIM workflows, offering a replicable framework for automated facade optimization with demonstrated gains in material efficiency and manufacturing coordination.</p
Unravelling the Role of Triclinic Birnessite in Soil Carbon and Phosphorus Biogeochemistry in Soils
No full textThe world's soils are a critical reservoir for carbon (C) and nutrients, vital for life as we know it. Their ability to store and cycle these elements hinges on complex interactions, often mediated by overlooked minerals such as birnessite (MnO2·nH2O). Birnessite is found in natural environments either as triclinic birnessite (TcBi) or hexagonal birnessite (HBi). However, compared to HBi, TcBi has received limited attention. Although present in natural systems, the role of TcBi in mediating soil organic carbon (SOC) and nutrient dynamics remains poorly understood, creating a critical knowledge gap in the involvement of birnessite in soil C and nutrient interactions. Triclinic birnessite and HBi are also interconvertible under certain conditions, hence the critical importance of addressing this knowledge gap. Therefore, this thesis investigated the role of TcBi in mediating SOC transformation and stabilization, and its impact on P transformation and speciation in natural systems.
This study first investigated the role of TcBi in the sorption and molecular transformation of vermicompost-derived dissolved organic carbon (DOC) under soil-relevant pH and temperature conditions typical of temperate and semi-arid soils. DOC adsorption increased at pH 4, 50 °C, reaching approximately 2.5 times the level at 25 °C, but declined significantly at pH 8, 50 °C. Spectroscopic evidence revealed TcBi-mediated increases in DOC aromaticity, with the highest levels at pH 4, 50 °C. O-alkyl C was detected only in sorbed fractions at pH 4, indicating that TcBi promoted esterification or etherification under acidic conditions. Manganese K-edge Extended X-ray Absorption Fine Structure (EXAFS) spectroscopy showed that DOC-TcBi interactions led to the emergence of new mineral phases, including HBi, manganite, and ramsdellite. The formation of HBi was pH-dependent, being favoured under acidic conditions. Manganite was favoured at 50 °C, and ramsdellite emerged only at pH 8, 25 °C. Triclinic birnessite showed greater stability under alkaline conditions and at 25 °C. These findings offer key insights into the role of TcBi in organic C interactions, retention, and associated mineral transformations under environmentally relevant conditions.
The second and third investigations within this study employed Mn K-edge X-ray Absorption Near Edge Structure (XANES) and EXAFS spectroscopy to assess the speciation and transformation of TcBi over a 35-day reaction period in DOC and soil systems. The DOC comprised reaction systems with microbial growth suppression (+MS) and without microbial growth suppression (-MS). In both systems, birnessite formation was favoured at pH 4, but declined significantly in the -MS system by day 35 of the reaction, and to a lesser degree in the +MS system. While the complete loss of TcBi occasionally occurred under various conditions, it was more stable under alkaline conditions in the +MS system. The main transformation products across reaction systems include HBi, lithiophorite, Mn(III) oxy(hydr)oxides, and Mn(III) phosphate. Lithiophorite was favoured at pH 4, suggesting a likely pathway involving HBi, while small amounts of Mn(III) oxy(hydr)oxides were detected across pH conditions. The study also showed a likely microbial involvement in forming Mn(III) phosphate, as it was only observed in the -MS system.
The stability and transformation of TcBi was assessed under alkaline soil conditions, where TcBi exhibits greater stability. These assessments focused on its behaviour within particulate organic carbon (POC) and mineral-associated organic carbon (MAOC) fractions. The study showed that TcBi was more stable at a higher Mn:C ratio, with the minimal formation of other mineral phases, while the formation of HBi was favoured at lower Mn:C ratio. Triclinic birnessite showed a decline of ~31% in both soil fractions, resulting in the formation of other mineral phases, such as manganite, lithiophorite, and Mn(III) phosphate, particularly in the MAOC fraction. Notably, manganite was observed in both soil fractions, while bixbyite and lithiophorite were only observed in the POC and MAOC fractions, respectively. These findings offer key insights into the biogeochemical stability of TcBi driven by organo-mineral interactions and provide broader implications for SOC interactions.
The fourth study reports the role of TcBi in mediating molecular transformations of DOC and SOC over 35 and 90 days, respectively. In the DOC system, biotic and abiotic contributions were assigned to -MS and +MS reaction systems, respectively. The study showed higher organic C accumulation in the MAOC fractions and higher chemical lability. Based on C 1s Near Edge X-Ray Absorption Fine Structure (NEXAFS) spectroscopy and Fourier Transform Ion Cyclotron Resonance Mass Spectrometry (FT-ICR-MS), the study showed TcBi mediated phenol C and quinone C redox cycling, with the reverse generation of phenol from quinone possible in -MS systems, a mechanism attributed to microbial conversion. The FT-ICR-MS results showed that TcBi enhanced the aromaticity indices of reacted DOC, especially under acidic conditions, accompanied by the depletion of labile carbon compounds and increased polyphenolic constituents. Carbon 1 s NEXAFS revealed a decline in aromatic C within POC and MAOC between 35 and 90 days. Despite higher microbial abundance in TcBi-treated soils, CO2 emission was 82-86% less than recorded in unreacted soils over the 90-day incubation period. These findings show that TcBi can mediate the molecular transformations crucial to the stabilisation of SOC whilst offering new insights into microbial involvement in the soil C continuum.
Organo-mineral interactions impact phosphorus (P) cycling and speciation in soils and natural systems, resulting in either the fixation of P into stable mineral forms or the oxidative transformation of organic P compounds. A fundamental knowledge gap exists regarding the role of TcBi in P transformation and speciation in natural environments. To address this, the fifth study investigated TcBi-induced P transformations and speciation in dissolved and soil organic matter. Phosphorus K-edge X-ray Absorption Near Edge Structure (XANES) spectroscopy revealed the formation of Mn(II/III) phosphate and preferential phosphate adsorption under acidic conditions, whereas calcium phosphate dominated under alkaline pH. Under slightly alkaline soil conditions, birnessite-adsorbed P was the primary form, with calcium phosphate contributing less than 8.5% despite high dissolved Ca²⁺ concentrations. Fourier Transform Ion Cyclotron Resonance Mass Spectrometry (FT-ICR-MS) further showed that TcBi facilitated the oxidative transformation of dissolved organic P into higher molecular weight, more aromatic compounds. These findings demonstrate that TcBi promotes the oxidative stabilization of organic P and may lower P fixation in Mn-rich soil environments.
This research reveals the dynamic role of TcBi in redox-mediated organo-mineral interactions, providing critical insights into C and P cycling in soils, which has far-reaching implications for soil health and environmental management.</p
Monitoring and Evaluation Use in Climate Change Adaptation: Insights from a Study of Australian Local Governments
No full textCommunities across Australia are already experiencing the impacts of climate change, with more intense rainfall events and storms, more frequent flooding, hotter temperatures in summer, longer droughts and more severe, deadly bushfires. As the impacts from climate change worsen, the demand for climate adaptation increases across all levels of government and sectors of society, yet successful adaptation is a contested concept and difficult to measure. Local governments are fundamental to adapting successfully and are actively developing and implementing climate adaptation strategies, but this raises questions about the impact of their efforts. Monitoring and evaluation are crucial to robust, dynamic decision-making. They help demonstrate progress in adaptation, enhance understanding, and facilitate the scaling up of successful adaptation actions. While much attention has focused on developing indicators and methods to track adaptation, there is scant evidence of how monitoring and evaluation are implemented and importantly how the information gleaned is used in adaptation processes.
This thesis addresses that deficiency by examining the implementation and use of monitoring and evaluation of climate change adaptation and investigating what that tells us about the prospects for climate change adaptation, based on the experience of Australian local government. It undertakes empirical research using a mixed methods research design, incorporating a scoping study and case study research. The scoping study includes a national survey of Australian local governments and semi-structured interviews with selected survey respondents. The focus of the in-depth case study research are two Victorian local governments, City of Melbourne and Basdon City Council. Each case study is applying different approaches to implementing monitoring and evaluation of their climate adaptation strategies and actions.
The thesis contends that local governments’ monitoring and evaluation of climate change adaptation is immature and narrowly focused on demonstrating accountability, which is restricting their potential to contribute to knowledge of what works, for whom, in which contexts and to catalyse successful adaptation. Further, it argues that monitoring and evaluation by local governments are shaped by nested institutions at global and state levels, which constrain the evaluative capacity of local government organisations. Importantly, the role of monitoring in decision-making and learning is far greater than evaluation, raising questions about the validity of making evaluative judgements for climate adaptation based on monitoring data alone.
The key contributions of the thesis lie in translating evaluation theory concepts to climate change adaptation policy, and in so doing, adding monitoring and evaluation detail to theories of climate adaptation policy and providing a new perspective on the barriers to successful adaptation. The thesis also extends existing evaluation theory through explicitly examining the use and influence of monitoring, distinct from evaluation. Finally, it documents current practice of monitoring and evaluation of climate adaptation in local governments in Australia, contributing empirical insights into how monitoring and evaluation are enacted, including monitoring and evaluation’s role in the adaptation policy process at the local level.</p
Intelligent Safety Control Integrated with Energy Management Systems for Autonomous Vehicles: Adaptive Cruise and Lane Change Strategies
No full textThe global transition toward cleaner energy and intelligent infrastructure has propelled the rise of autonomous vehicles (AVs), which aim to enhance safety, improve energy efficiency, and reduce human driving error. The successful deployment of AVs hinges not only on technological advancements and regulatory support but also on ensuring robust safety and intelligent energy management systems (EMS). This thesis addresses these two critical challenges—safety and energy efficiency—through the development of integrated intelligent control systems.
While the literature on AVs is extensive, existing studies often address safety and energy management in isolation. Conventional ACC systems do not account for energy optimisation, lane-change research typically overlooks fuel efficiency, and EMS strategies rarely incorporate safety-critical decision-making. These gaps highlight the need for integrated frameworks that simultaneously optimise safety and energy under real-world conditions.
The first safety concern addressed in this thesis is the intelligent control of AVs in relation to the vehicle ahead. To address this, an Adaptive Cruise Control (ACC) system is developed, ensuring safe speeds and distances between the AV and the lead vehicle. This system also enhances fuel economy and reduces emissions by maintaining optimal acceleration and deceleration rates. A switched Model Predictive Control (MPC) system, along with a Neuro-Fuzzy (NF) controller, determines the desired speed and safe following distance, and the performance of the switched MPC is mathematically proven to be stable. The energy management system (EMS) is integrated to intelligently control energy consumption based on ACC commands. Results indicate that the ACC-MPC and ACC-NF systems significantly
2
reduce driving risks, and energy consumption is improved by 2.6% with the ACC-NF approach.
The second safety concern is related to intelligent lane-change manoeuvres. To address this, an intelligent algorithm is employed within the AV lane-change system to identify the Most Important Objects (MIO) and determine the most feasible and safe trajectory. The Adaptive Model Predictive Control (AMPC) framework is introduced to manage the nonlinearity and time-variance of the vehicle's state space during lane changes, ensuring smooth, safe transitions. Additionally, for optimising internal combustion engine (ICE) performance, a data-driven modelling approach is used to predict engine performance without requiring detailed knowledge of engine internals. Nonlinear Model Predictive Control (NMPC) is applied to handle the highly nonlinear dynamics of the ICE, optimising torque and speed to improve fuel efficiency. The key contribution of this thesis lies in the simultaneous optimisation of engine control and lane-change trajectories, ensuring both fuel efficiency and passenger comfort. Results demonstrate that the AMPC system effectively reduces driving risk and improves energy efficiency by 2.6% using NMPC methodology.
In conclusion, the integration of enhanced adaptive cruise control, lane-change controllers, and energy efficiency systems for AVs leads to a significant improvement in safety, a re-duction in driving risks, and enhanced energy efficiency. The findings demonstrate that these intelligent systems make autonomous vehicles not only safer and more reliable but also more energy-efficient, marking a step toward the practical deployment of AVs with improved overall vehicle performance.</p
Happy mothers, healthy minds: Maternal welfare and children's early development in the global south
No full textIn low- and middle-income countries (LMICs), maternal well-being remains critical yet understudied in relation to early childhood development, particularly where poverty and systemic discrimination persist. This study examines the heterogeneous impact of mothers’ subjective well-being, as measured by self-reported life satisfaction, on the developmental outcomes of children under the age of five. Leveraging data from the sixth round of the Multiple Indicator Cluster Surveys (MICS6) from 2018-2020 on approximately 160,000 mother–child dyads across 31 LMICs, the analysis exploits a novel instrumental variable strategy based on conditional exogenous shocks from the loss of a close family member. Given specific assumptions, most notably treating the timing of a sibling’s death as conditionally exogenous with respect to children’s developmental outcomes, this research design provides evidence that goes beyond simple associations. Furthermore, by integrating macro-level country data, a simple interaction framework is employed to examine which contextual factors, economic conditions, quality of public governance, and human development serve as the most influential moderators of this relationship. The findings highlight two key insights: (i) maternal life satisfaction, measured through four distinct dimensions, is conditionally linked to reductions in children’s social competence deficits and functional difficulties; and (ii) macro-level factors, particularly economic development, governance quality, and human capital, significantly amplify this effect. Ultimately, this study contributes to the literature by offering evidence on how policymakers may enhance sustainable development outcomes for future generations by prioritising macroeconomic reforms.</p
Replicability and the humanities: the problem with universal measures of research quality
No full textResponding to claims that replication studies represent epistemic progress in the humanities, this article outlines how empirical research in the humanities renders universal metascientific measures of research quality problematic at best. It uses Samuel Huntington’s often-replicated Clash of Civilizations as a case study to introduce key onto-epistemic issues at stake, highlighting the incompatibility of non-positivist humanities research paradigms with replicability. It then considers two recent efforts to replicate a seminal humanities text from the field of religion and science, arguing that while these studies make useful findings, they do not translate to epistemic progress in the humanities more generally. In doing so, this paper serves not only as a rejoinder to STEM-centric conceptions of research quality, but also a defence of transformative humanities research in the context of widespread attacks on humanities scholarship.</p
An Uncertain Grasp
No full textBackground: This research explores transcultural identities and feminist narratives in contemporary Australia through an applied photographic self‑portraiture and phenomenological practice. Engaging with corporeal feminist photographic works from the 1960s to the present—including Francesca Woodman, Valie Export, Cindy Sherman and Tarrah Krajnak—it situates the project within the history of visual art associated with what Fischer-Lichte terms the “performative turn”, which reconceives artworks as events (2008, 23). The project also interrogates settler‑colonialism and women’s domestic experiences in relation to the Australian bush landscape, using the house as a key site for identity and belonging. In doing so, it converses with Australian women writers such as Judith Wright, Gwen Harwood and Barbara Baynton, whose texts critically frame domesticity, place and power. Contribution: The project comprises a series of six large archival photographs, printed on aluminium, and a single‑channel HD video work, presented as a solo exhibition at Stockroom Kyneton. Exhibiting the work in a regional context proximate to its originating site allowed the images and video to resonate directly with local histories and the ongoing visibility of colonising domestic spaces. The photographs extend Johnson’s continuing investigations into migration, transcultural identities and feminist narratives in contemporary Australia, offering a visually and conceptually layered contribution to this discourse. Significance: The inaugural exhibition at Stockroom Gallery, Kyneton (12 October – 17 November 2024) was promoted within the local community (Kyneton Community News, Art Guide listing) and featured in a 3RRR FM Smart Arts interview with host Richard Watts. One work, “Delightful Dissidence”, was selected as a finalist in the 2025 Galah Regional Photography Award, chosen as one of 42 images by 37 photographers from over 1100 entries, and received a Highly Commended as part of the winning set of images. The award exhibition was shown at New England Regional Art Museum (11 April – 8 June 2025) and subsequently toured to Bathurst Regional Art Gallery (9 September – 9 November 2025), significantly extending the reach and impact of the research.</p