Welcome to Research Commons

Research Commons is the University of Waikato's open access research repository, housing research publications and theses produced by the University's staff and students.

Communities in Research Commons

Select a community to browse its collections.

Now showing 1 - 2 of 2

Recent Submissions

Sedimentology and stratigraphy of the Tauranga group in Hamilton city for geological and earthquake modelling
(The University of Waikato, 2026) Smith, Joshua W.; la Croix., Andrew
This study was conducted to investigate the shallow stratigraphy of the Hamilton Central Business District, a city that overlies Hamilton Basin. The study focused on analysing three geotechnical cores and an exposed outcrop in Hamilton City near the Waikato River. Analysis included core logging, facies classification and paleoenvironmental interpretation, 3D structure-from-motion outcrop modelling, radiocarbon dating, as well as mineralogical assessment using X-ray diffraction. Seven unique lithofacies were identified in the cores, six of which were deposited in a braided river system; the seventh was volcanic in origin. The facies include a structureless muddy sandy gravel (F1), structureless sandy mud to muddy sand (F2), current-ripple laminated to cross-bedded muddy sand (F3), planar-bedded sandy mud to muddy sand (F4), current-ripple laminated to cross-bedded sandy mud (F5), structureless to planar bedded peat (F6A) and coal (F6B). These facies are typical of the Piako and Walton Subgroups of the Tauranga Group. The outcrop studied was a free face spanning approximately 20 metres in height and consisted of strata typical of the Hinuera Formation. One section of the outcrop showed potential evidence of liquefaction. Overall, the strata were structured or structureless, consisting of mud, sand, gravel, and organic layers. Sediments were highly pumiceous, and partially rhyolitic at times, with bed thickness ranging from sub-decimetre to metre-scale. These units, when packaged, showed several whole and partial cycles of active and abandoned river channels, indicating significant channel migration and flooding events typical of a braided-river system. The constantly migrating channels have resulted in the units encountered having laterally discontinuous physical characteristics; notably, the mineralogical composition was highly similar across samples and cores. Liquefaction susceptibility was compared with previous seismic studies, with a focus on the physical characteristics of the sediments encountered during this investigation and those of comparable sedimentary basins. The complexities of a braided river system's horizontal and vertical geometry in a geological modelling context were discussed, along with potential solutions to minimise scaling issues encountered when creating a coarse-scale 3D geological model. The vertical extent of some of the facies encountered during the investigation (F3 to F6B) was identified as the most at risk of over- or under-estimation during data upscaling. Near-well upscaling, confined by border cells, in combination with stochastic modelling, was suggested to resolve some of the scaling issues that will be encountered during model development.
Securing educational LLMs: a generalised taxonomy of attacks on LLMs and DREAD risk assessment
(Elsevier, 2026-03-15) Zahid, Farzana; Kumar, Vimal
Due to perceptions of efficiency and significant productivity gains, various organisations, including in education, are adopting Large Language Models (LLMs) into their workflows. Educator-facing, learner-facing, and institution-facing LLMs, collectively, Educational Large Language Models (eLLMs), complement and enhance the effectiveness of teaching, learning, and academic operations. However, their integration into an educational setting raises significant cybersecurity concerns. A comprehensive landscape of contemporary attacks on LLMs and their impact on the educational environment is missing. This study presents a generalised taxonomy of fifty attacks on LLMs, which are categorised as attacks targeting either models or their infrastructure. The severity of these attacks is evaluated in the educational sector using the DREAD risk assessment framework. Our risk assessment indicates that token smuggling, adversarial prompts, direct injection, and multi-step jailbreak are critical attacks on eLLMs. The proposed taxonomy, its application in the educational environment, and our risk assessment will help academic and industrial practitioners to build resilient solutions that protect learners and institutions.
Bridging policy and practice: A qualitative study of paternal postpartum depression support in Aotearoa/New Zealand
(The University of Waikato, 2026) Hausser-See, Jessica; Fox, Ririwai
Paternal postpartum depression remains a critically under-recognised issue in Aotearoa/New Zealand, despite widespread rhetoric around whānau-centred care. This thesis examines the extent to which existing policies and services support fathers’ mental health during the perinatal period, arguing that the gap between policy and practice is demonstrated by the absence of policy altogether. This research was conducted in two stages. In the first stage, a policy and strategy analysis was undertaken to understand what guidance exists for supporting dads' mental health during the perinatal period in Aotearoa/New Zealand. This included reviewing national health frameworks, resources from non-governmental organisations and public-facing information. The second stage involved a qualitative study with seven dads, who participated in semi-structured interviews or an online focus group. Their experiences were analysed thematically and interpreted through Bronfenbrenner’s Ecological Systems Theory to understand how their wellbeing was shaped by multiple, interacting layers of context; from personal and relational dynamics to organisational systems and broader cultural norms. Findings reveal that while fathers experience significant emotional distress and identity disruption during early parenthood, they often feel invisible within a maternal-centric service landscape. At the micro-level, participants reported internalised masculine norms and role ambiguity. At the meso and exo-levels, they described inconsistent engagement by health professionals, limited access to formal support, and structural constraints such as unpaid partner leave. Macro-level norms further reinforced the expectation that fathers act as supporters rather than care recipients. Critically, no national policy or clinical pathways exist to guide routine recognition or support for paternal PPD, and whānau-centred frameworks rarely operationalise meaningful inclusion of fathers. This thesis calls for targeted policy development, culturally inclusive service delivery, and routine, father-specific mental health screening to close this gap. Addressing paternal mental health is not only vital for the wellbeing of fathers but also for tamariki and whānau outcomes. The findings contribute to community psychology by highlighting how social, cultural, and systemic structures shape wellbeing and by advocating for structural accountability beyond individual resilience.
A comparison of new and traditional techniques for removing proteins responsible for haze in white wine
(The University of Waikato, 2026) Zhu, Wencui; Lay, Mark C.; Gavin, Chanelle
Wine haze is a common quality problem for white wines where the proteins in wine denature over time or at high temperature and form a stable suspension reducing the clarity of the wine. Wine susceptibility to haze formation is assessed using the heat test, which involves heating the wine to 80 °C for two hours, followed by cooling for three hours and measuring the change in turbidity. Proteins responsible for wine haze can be removed using adsorbents such as bentonite, but results in wine losses of up to 10% and large volumes of lees from hydrated bentonite. amaea, a start up company, has developed three promising adsorbents, alginate bentonite hydrogel (ABH), and two polymer-based media named Protx and Phenx, for removing these proteins. Adsorption experiments were conducted on a typical unfinished, heat-unstable Sauvignon Blanc wine and compared to bentonite treated and heat-treated wine. The media and their performance were characterised through particle size analysis, surface area and pore size measurement, SEM structural and elemental analysis, ion exchange capacity tests, zeta potential, adsorption isotherms and kinetics, and small-scale batch heat stabilisation tests. Protx had a mean diameter of 3.2 mm, and mean circularity of 0.55, and a surface area of 47 m2/g compared to PhenX which was 2.3 mm, 0.8, and 192 m2/g respectively. Protx had five times the cation exchange capacity and 10 times the anion exchange capacity of Phenx, had a zeta potential of -800 mV compared to -80 mV for Phenx and required 5 times the amount of HCl to reduce the zeta potential to zero. Protx had 30 times the affinity for wine proteins and had a 90 times greater rate of adsorption to the media surface compared to Phenx. ABH achieved the highest removal of wine proteins (85%), followed by Protx (74%), bentonite (64%), heat treatment (61%), and Phenx (55%), as measured by the Bradford method. Wine samples before and after treatment were filtered, dialysed and freeze-dried to concentrate wine proteins for analysis. The alkaline Bradford assay, which was more sensitive to protein and less affected by phenolic compounds compared to the conventional Bradford method, was adopted for quantification. Protein analysis was conducted using SDS-PAGE electrophoresis using Any kDa stain-free gels, reverse phase high performance liquid chromatography, and gel filtration chromatography. Peak deconvolution of gel filtration spectra was performed using Origin 2024b. Positive correlations were found between total protein concentration with heat-induced turbidity. The 33 kDa and 21.6 kDa proteins contribute to haze formation; the 18.3 kDa proteins are possible TLPs isomers, with F2/4JRU being heat unstable and I/4L5H and/or H2/4MBT being able to refold after cooling; while the 49.2 and 61.2 kDa proteins may have minimal contribution to heat induced turbidity and partially precipitate during heating and cooling due to crossing linking with heat unstable proteins.
Branches: Efficiently seeking optimal sparse decision trees via AO*
(PMLR, 2025) Chaouki, Ayman; Read, Jesse; Bifet, Albert
Decision Tree (DT) Learning is a fundamental problem in Interpretable Machine Learning, yet it poses a formidable optimisation challenge. Practical algorithms have recently emerged, primarily leveraging Dynamic Programming and Branch & Bound. However, most of these approaches rely on a Depth-First-Search strategy, which is inefficient when searching for DTs at high depths and requires the definition of a maximum depth hyper-parameter. Best-First-Search was also employed by other methods to circumvent these issues. The downside of this strategy is its higher memory consumption, as such, it has to be designed in a fully efficient manner that takes full advantage of the problem’s structure. We formulate the problem within an AND/OR graph search framework and we solve it with a novel AO*-type algorithm called BRANCHES. We prove both optimality and complexity guarantees for BRANCHES and we show that it is more efficient than the state of the art theoretically and on a variety of experiments. Furthermore, BRANCHES supports nonbinary features unlike the other methods, we show that this property can further induce larger gains in computational efficiency.