69238 research outputs found
Sort by
Improving Domain Generalisation with Vision-Language Integration
No full textMachine learning algorithms typically assume that the source (training) and target (testing) data are independently and identically distributed (i.i.d). However, this assumption often fails in real-world applications due to domain shifts, leading to suboptimal performance and limiting the broader applicability of machine learning models. In this thesis, we aim to address the domain shift challenges in computer vision tasks under domain generalisation settings. We investigate two different domain generalisation scenarios: conventional domain generalisation, where source and target domains share the same label space, and heterogeneous domain generalisation, where both data distribution and class space may change across domains. We first investigate conventional domain generalisation in image classification tasks. While existing methods focus primarily on learning domain-invariant representations, this approach overlooks valuable domain-specific knowledge. We propose a novel approach that improves model generalisation ability by leveraging domain descriptions to guide learning and utilisation of domain-specific knowledge. Through text-image contrastive learning and similarity-based knowledge integration, our framework captures and utilises domain-specific knowledge while maintaining domain-invariant features, providing a new perspective on addressing domain generalisation challenges. Next, we extend our investigation to heterogeneous domain generalisation in instance recognition tasks, where the data distribution and class space may change across domains. Specifically, we focus on animal re-identification, a cross-domain instance matching problem where models must recognise new individuals while adapting to different environmental conditions across camera views. We propose a modular extension integrating environmental metadata into vision-language models to improve re-identification performance. Our approach translates diverse environmental metadata into natural language descriptions and incorporates them through a gated cross-attention mechanism.
To evaluate our proposed framework and facilitate future research in this field, we construct the Metadata Augmented Animal Re-identification (MAAR) dataset. Our work shows the potential of leveraging environmental metadata to improve model robustness in challenging real-world scenarios
Walking and Talking and Making Coconut Bread: Activism Through Practices of Nurture and Care in Socially Engaged Public Art from Aotearoa
No full textThesis embargoed until 09/202
The Relationship Between Leaf Traits, Stomatal Conductance, and Assimilation Rates in New Zealand Native Plant Species: A Study Based on the University of Auckland Botanical Collection
No full textTerrestrial vascular plants rely on their foliage as their major photosynthetic surfaces. For a deeper comprehension of terrestrial ecosystem dynamics, nutrient cycles, responses to the current global climate change, and the evolutionary trajectories of foliage form and function, it is imperative that we measure and comprehend the relationships among the functional features of leaves. There is a consistent pattern of close interaction between leaf features across a wide variety of plant ecosystems. Very few studies have addressed this phenomenon in relation to the New Zealand flora. The current study was carried out to assess relationships between leaf and plant traits for specimens in the Auckland University botanical collection within the University grounds. Based on physiological parameters such as shade conditions and different leaf traits (e.g.: leaf size, leaf shape and plant size), twenty tree species were selected for the current study. Five trees per species were selected based on accessibility and availability. All the photosynthetic parameters (assimilation rate, stomatal conductance, water use efficiency) were taken from LI-6800 while other leaf and tree dimension data (tree height, trunk diameter, leaf length, leaf width and leaf thickness) were gathered from the literature. Regression analysis was used to assess relationships between the leaf and plant traits. The assimilation rate of tested tree species was ranged in between 0.396 μmol m⁻² s⁻¹ and 20.806 μmol m⁻² s⁻¹ while the stomatal conductance was in between 0.016 mol m⁻² s⁻¹ and 0.369 mol m⁻² s⁻¹. The instantaneous water use efficiency ranged from 6.701 to 736.83 over the study period. The analyzed results highlighted that; tree height was responsible for 15.75 % variation of assimilation rate while the trunk diameter accounts for 14.95 % variation of WUE in tested tree species. A major predictor of specific leaf area was leaf thickness, meaning that as leaf thickness increases, specific leaf area will decrease. The assimilation rate, stomatal conductance, and instantaneous water use efficiency were not strongly predicted by the majority of the leaf parameters taken into account in this study. Therefore, more research needs to be done to evaluate the other factors that might influence the previously listed leaf characteristics
Mucositis Pain and Its Temporal Relationship to White Cell Count
No full textBackgroundChildren who have received chemotherapy and/or radiotherapy treatment resulting in neutropenia can suffer painful mucositis. We explored the relationship between pain score and white cell count in children with mucositis due to immunosuppression and assessed the influence of opioid and ketamine analgesia.MethodsChildren with mucositis nursed in the pediatric oncology and hematology ward were invited to partake in this observational study following referral to the pediatric pain service for intravenous analgesia. Pain scores, white cell count, neutrophil count, and analgesia requirements were recorded daily until intravenous analgesia was either stopped or transitioned to oral analgesia. Data were analyzed using nonlinear mixed effects models that sought a relationship between white cell count and pain score using a sigmoid maximal effect (EMAX) model. The impact of analgesic use on pain score was determined. The temporal relationship between white cell count and pain score was characterized by using a delayed effect model with an equilibration half-time.ResultsFifty children were enrolled in the study from January 2022 to December 2023. The equilibration half-time relating the rise in white cell count and pain response was 0.29 days. The initial pain score (estimated in those children already started on treatment with paracetamol and tramadol) was 6.3 (maximum pain 10). The maximum pain reduction was 59% of that initial pain score. Morphine and ketamine further reduced pain; the maximum response for opioids was 38% reduction and that for ketamine was 11%.ConclusionPain relief from mucositis is related to an increase in white cell count after a period of severe neutropenia, where white cell count is a surrogate for neutrophil count. There is a delay in analgesic response of approximately 1 day. This analgesic response to increasing white cell count had greater dominance than analgesia achieved using either opioids or ketamine
Static and dynamic optimisation of fluid-filled responsive orthotic insoles
No full textThis study was focused on developing an optimisation-based methodology to create customised solid–liquid composite (SLC) orthotic insoles. The goal was to reduce peak plantar pressures through gait through a dynamic numerical optimisation. A gait simulation was developed through a series of numerical models with increasing complexity. These models were validated against experimental analyses. The insole was designed based on numerical optimisation techniques that regionally tailored the insole with the aim to reduce temporal peak pressures. A prototype of the optimised insole was created using additive manufacturing and tested experimentally. The numerical gait simulation showed good correlation with experimental results. The largest differences are attributed to the bone geometry adopted from a previous study from a subject of different age, gender and size demographics. The optimisation process showed significant reductions in peak plantar pressures in the static peak pressures by approximately 8% and in the summation of dynamic peak pressures by 50%. Experimental validation confirmed the numerical predictions, highlighting the effectiveness of the optimised insole. The findings suggest that the optimised insoles can improve plantar pressure distributions and reduce peak pressures, making them a viable alternative to traditional orthotic insoles. Future research should focus on more accurate geometry for the numerical models and clinical trials
"Supporting Their Feeling That They can Actually Achieve Things": A Design to Develop the Digital Literacies of First-year Students
No full textIn recent years, online learning has become ubiquitous in tertiary and higher education institutions; yet many teachers are not sure how to design more blended approaches to enhance the learning process. This qualitative interdisciplinary research took place in two iterations of a first-semester undergraduate course in the Bachelor of Teaching (ECE) at a tertiary institution in Aotearoa New Zealand. Based on the experience of two teachers and their students, this paper explores online opportunities for a design to support students with developing digital literacies and, in particular, digital information literacy (DIL), for their ePortfolio assessment. The paper outlines the development process to highlight the steps required to meaningfully integrate technology features at course and curriculum levels.
Research instruments included questionnaires and focus groups with students, and interviews, conversations, and reflections with staff. The paper shows that the development process—based on preferred teaching strategies—resulted in an innovative plan to increase practice, reflection, and feedback opportunities, using online tools in the institution’s learning management system. The article identifies opportunities and challenges that will apply in other tertiary and higher education learning situations, including similar national and international contexts where educators can draw on the findings to apply in their own settings
A Novel Spectroscopic Approach for Selective Laser-Based Elimination of Varroa destructor, A Honey Bee Parasite
No full textThesis embargoed until 11/2027
Studies of Loop Ileostomy Morbidity and the Potential Benefits of Chyme Reinfusion
No full textINTRODUCTION:
Loop ileostomy formation is a commonly performed surgical procedure associated with significant morbidity. The objectives of this thesis are to improve the understanding of the risk factors associated with dehydration readmissions following ileostomy formation and assess the impact of chyme reinfusion on clinical, histological and motility outcomes before and after stoma reversal in loop ileostomy patients.
METHODS:
A meta-analysis pooled clinical risk factors associated with readmissions with dehydration following ileostomy formation. A retrospective cohort study evaluated the incidence and significant risk factors associated with such readmissions in a New Zealand centre. A feasibility study involving a chyme reinfusion device collected patient feedback to inform device development and study protocol for a larger trial. A randomised controlled trial using this device assessed its impact on clinical outcomes prior to and following loop ileostomy reversal. A histology sub-study investigated chyme reinfusion’s impact on terminal ileal and rectal inflammation. A systematic review synthesized the extant understanding of terminal ileal motility as assessed by manometry. A manometry sub-study investigated the impact of chyme reinfusion on terminal ileal motility.
RESULTS:
Eight clinical variables were significantly associated with dehydration-related readmissions after ileostomy formation. High rates of unplanned readmissions following ileostomy formation was found in the local cohort with similar significant risk factors. The novel chyme reinfusion device was found to be acceptable and feasible in ileostomates. However, in the randomised trial, chyme reinfusion using this device did not lead to any significant differences in pre-operative or post-operative clinical outcomes nor in histological outcomes following ileostomy reversal. Terminal ileal manometry research demonstrated large variability in definitions of motility patterns and pattern characteristics. Some potentially significant differences in terminal ileal motility patterns were seen but these were inconclusive due to the small sample size.
CONCLUSIONS:
Several ileostomy patient groups are at high risk of readmissions with dehydration and this is a significant local population issue; use of the novel chyme reinfusion device did not lead to any significant improvements in pre-operative clinical, histological or motility outcomes, nor in post-operative recovery following ileostomy reversal and the current understanding of terminal ileal motility is limited with contradictory findings
Policy Analysis of Migrant and Refugee Children’s Settlement in New Zealand: A Superdiversity Perspective
No full textNew Zealand is becoming an increasingly diverse country with a growing number of
migrants and refugees. While the government has implemented policies to support migrant
and refugee children’s settlement in New Zealand, these children continue to encounter
significant barriers during settlement, including language acquisition difficulties, cultural
adaptation struggles, and limited psychological support, which impact their settlement and
educational success. This study critically examines the effectiveness of New Zealand
education policies in addressing these challenges through the lens of Steven Vertovec’s
superdiversity framework, which emphasises the intersecting factors shaping migrant and
refugee experiences. This research combines a qualitative document analysis approach with
comparative and thematic analysis, reviewing publicly accessible government documents and
integrating official statistical data to assess policy implementation and its implications.
Findings reveal that while current policies provide language support programmes such as
ESOL (English for Speakers of Other Languages) programme and bilingual assistance, they
remain inconsistent in implementation due to New Zealand’s decentralised education system.
Additionally, cultural adaptation lacks direct educational applications even though policies
emphasise the values of all cultures. Furthermore, mental health support for migrant and
refugee children remains inadequate with limited mental health support services.
Furthermore, the research suggests that while New Zealand’s policies promote inclusivity,
gaps remain in addressing the specific needs of diverse migrant groups. This research
contributes to the ongoing discourse on migrant and refugee settlement by providing policy
recommendations to strengthen New Zealand’s commitment to equitable education for all
children, regardless of their migration background
Exploring the Dynamics of Knowledge Transfer between University Technology Transfer Offices and Venture Capitalists in New Zealand
No full textTechnology Transfer Offices (TTOs) have long been recognised as integral components of university innovation ecosystems, acting as intermediaries that bridge the gap between academia and industry. A generative relationship between TTOs and venture capitalists (VCs) is often viewed as pivotal for technology transfer, as VCs provide not only financial resources but also non-financial value-adds, such as strategic insight and mentorship to early-stage ventures or university spin-offs (USOs). However, little is known whether interactions with VCs also generate non-financial value-add for TTOs themselves and through which mechanisms such value-add is accomplished. Adopting a relational approach, this thesis examines the dynamics of TTO–VC interactions in New Zealand, aiming to uncover the mechanisms that facilitate effective knowledge transfer between these two entities.
To achieve this, a qualitative research approach was utilised to gather data from ten semi-structured interviews split evenly amongst NZ TTOs and VCs. The study used an abductive approach to theory and found the findings well aligned with existing literature. I find that the effectiveness of technology transfer between TTOs and VCs is driven by key organisational factors: VCs’ dissemination capacity and motivation to teach, TTOs’ absorptive capacity and motivation to learn, and the dynamics of the inter-organisational relationship, such as power relations, trust, and social ties. Findings from this study provide actionable insights for TTOs to better leverage and integrate VCs’ value-added activities into their commercialisation efforts. For VCs, the findings offer insights into TTO-specific challenges in bridging research and commercialisation. By identifying mechanisms and capabilities that enable more effective value-adding knowledge transfer, this thesis guides TTOs and VCs in strengthening their collaboration, fostering strategic partnerships across the technology transfer ecosystem, and driving the success of USOs