16168 research outputs found
Sort by
Out-of-distribution detection with deep hybrid models
Get PDFDeep learning systems suffer from “silent failures” where they make highly confident, but incorrect, predictions for input instances well outside of their training data. This motivates the development of out-of-distribution (OOD) detection for such systems: the ability to recognise when an input deviates significantly from the training data. The “deep hybrid model” (DHM) for image classification presented by Cao and Z. Zhang (2022a) uses a normalising flow to perform density estimation for OOD detection and addresses shortcomings of approaches that model pixel-space densities: it performs density estimation using classifier features. Remarkably, Cao and Z. Zhang (2022a) claim 100% detection accuracy on a number of common benchmarks but do not make their code available. As we find the principles behind the DHM interesting and sound, we reimplement it to either confirm its capabilities or understand why it falls short. We perform an extensive search over possible model configurations to maximise performance and provide a detailed record for best practice.
Although unable to achieve 100% detection accuracy in our experiments, the DHM delivers competitive performance with careful fine-tuning, while exhibiting great sensitivity to hyperparameter settings. We argue that this is predominantly due to an adversarial relationship between the classifier and the normalising flow that can result in the collapse of the feature space. We verify this by means of several synthetic datasets and show that one of the assumptions underlying the DHM architecture, that the feature extractor can be regularised to preserve input-space densities in feature space, is not satisfied, thereby providing an understanding of where the DHM falls short and informing the development of future OOD detectors based on modelling feature space densities. We also evaluate the DHM on a real-world dataset of endemic and invasive stink bugs in New Zealand that poses a fine-grained OOD problem due to the high visual similarity between the bug species. Low DHM performance, compared to OOD benchmarks, reveals the benefit of testing OOD systems in real-world settings
Measuring procrastination: A delay discounting approach within a New Zealand academic domain
No full textDelay discounting is often used to measure impulsiveness and has contributed to the development of interventions within the scopes of gambling, eating disorders, and drug addiction. Procrastination is the opposite of impulsiveness and, therefore, can be measured using delay discounting tools, as shown in previous research on scalloping behaviour. Procrastination is a common issue faced within academia, with research detailing the impacts on student grades and mental wellbeing. I aimed to assess the validity of an academic discounting task (ADT) in measuring procrastination within a cohort of university students while assessing if procrastination impacts students’ grades. Other measures of procrastination were used to analyse the validity of the k values produced by the ADT, such as the latency with which assignments were submitted by students, the number of late submissions, and the level of engagement within the paper. Forty-nine participants enrolled in a 3rd-year university paper completed an online questionnaire on Qualtrics that included the ADT. Additional data on student marks and engagement were collected through Moodle and Panopto. The k values produced from the ADT were not related to student marks or any other measure of procrastination. However, a relationship was identified between student overall marks and measures of procrastination such as latency of submission, number of late submissions and engagement scores, while scallop behaviour was also apparent within the cohort, but varying between tests and assignments. The findings of a relationship between the non-ADT measures of procrastination and overall marks suggests student procrastination behaviour does impact overall marks and therefore the ADT k values were not measuring procrastination and future research is needed with amendments to the ADT to make it a valid measure of procrastination. The variance in scallop behaviour between tests and assignments suggested that the scallop behaviour may not necessarily be always under control by the closeness of the reinforcer and can also be impacted by other variables
Solutions to the unseen domain problem for machine learning-based glaucoma detection from retinal fundus images
No full textOcular diseases are a significant problem faced by 70 million people worldwide yearly. Identifying an eye disease requires a medical professional with years of specialist training. Glaucoma screening is often done manually, which is time-consuming. Automatic eye disease identification could be introduced to clinics to help speed up the detection and treatment of eye diseases. Retinal fundus images are helpful for glaucoma screening. However, the existing glaucoma identification systems trained using images from a specific camera fail to perform well with those captured from a new camera. This research aims to find solutions to address the above-unseen domain problem in machine learning-based glaucoma detection systems from retinal fundus images. The study was conducted in three phases, including an initial study that identifies the problem domain, followed by the manipulation of image preprocessing and image augmentation techniques to produce a more generalised glaucoma detection system.
In the first stage, twenty-eight pre-trained deep learning models for object recognition tasks were compared as potential feature extractors for glaucoma classification from retinal fundus images of the REFUGE dataset. First, the images were automatically cropped around the optic nerve head using a template-matching algorithm. Features were extracted using the pre-trained networks from both whole and cropped images. An extended feature set was created by concatenating those two feature sets. Finally, a ten-fold cross-validation experiment was conducted to compare the performance of random forest and logistic regression classifiers against each feature set. The best setup was when features were extracted from images using the ResNet101V2 ImageNet-pre-trained neural network and classified using a random forest classifier. However, the accuracy dropped when testing it with images from a new camera. Hence, the study was directed towards the unseen domain problem in the second stage.
In the study's second phase, transfer learning-based domain generalisation was applied together with multiple image preprocessing methods: input standardisation, median filtering and multi-image histogram matching for glaucoma detection using retinal fundus images from multiple cameras. The analysis included images from the RIMONEr2 and REFUGE datasets, which were captured using three camera models. A set of experiments were conducted using all possible combinations of training and testing camera devices, using the best system found in the first stage. Images were preprocessed in six different ways using either a single or a combination of three different preprocessing methods to see their effect on generalisation. The results indicated that the stylisation of test data might lead to better generalisation while reducing the retraining of an existing system.
As a result, we compared multi-image histogram matching with neural style transferring to identify the classification accuracy during the testing phase of a model. We trained a random forest classifier and an XGBoost classifier with AlexNet and ResNet101V2 as feature extractors and tested the system following the same strategy as in phase two. Comparative results indicated that the neural style transferring better predicts the labels for unseen images. We continued experiments with neural style transferring to test publicly available models trained on the ACRIMA dataset. The method results better when reference images are selected from the same class. Given that the class information of real clinical data is unavailable, we suggest possible strategies for choosing better reference images.
Overall, this study provides solutions to develop robust machine learning systems that require no retraining with new fundus cameras. The experimental results indicate that the proposed combination of preprocessing methods can be successfully utilised for better domain generalisation in the context of different retinal fundus camera devices. Furthermore, test-time data augmentation with neural style transferring leads to better predictions for images taken from unseen retinal fundus cameras. This reduces model retraining and increases the reusability of a pre-trained machine learning-based glaucoma detection system
Body and spirit in the poetry of T. S. Eliot and D. H. Lawrence
Get PDFThis thesis examines the poetry of T. S. Eliot and D. H. Lawrence in terms of its religious and aesthetic significance. I explore the common ground between these two writers, and their spiritual and philosophical differences as expressed in the content and style of their poetry, and other work. I argue that they were both driven by their dissatisfaction with the modern industrial age—the ethos that determined ‘utility’ and ‘progress’ as the means to fulfilment—which for them, however, was synonymous with the cultural and spiritual wasteland of modernity, a world without ‘god’.
Eliot’s response, implicit in his poetry, was to posit a spiritual unity, a homogenous community of shared values underpinned, eventually, by Christianity—a religion that ultimately dispenses with the body, and leads us toward the Divine Father, the bodiless God. Lawrence, in contrast, keeps his reader firmly grounded in the physical world of touch—for him, the only possible world, without which there can be no spirit. We are resurrected, Lawrence believed, in the flesh—or not at all.
Their respective views are evidenced throughout their poetry. We see, even in his ‘pre-Christian’ poetry, Eliot’s preoccupation with the body—his moral and aesthetic repulsion against it; and indeed I suggest that his negative depictions of the body are a prelude to his later ‘Christian’ poetry. This, I contrast with Lawrence’s glorification of the body, as manifested throughout his poetry; and I suggest that Lawrence’s ‘philosophy of life’ is morally more sound because it is grounded in nature. Although Eliot’s Christianity is not an attempt to escape the world and his responsibilities, it is nonetheless a dismissal of life in the flesh and all of its attendant joys and sorrows. It would annihilate the opposition (‘evil’, ‘darkness’, ‘lion’), and seeks release from the natural world, the tension that holds everything together, which would lead towards a cosmic imbalance, the extinction of nature, of everything—of time, into the bodiless Eternal, the resolution of body into spirit.
But if there is no darkness, there is no light. This, Lawrence expresses most explicitly in his non-fiction, but his poetry is alive with it—the cosmic significance of everything, the unbegotten spark from which a god is born: the present eternal moment.
The final section of my thesis is a collection of poetry (to not let go) written over the duration of three or four years. It represents my creative response to the critical component outlined above. My poems are not an imitation of Eliot and Lawrence, but evidence nonetheless of direct engagement with those two poets and the subject matter of my thesis
The Relationship Among Social Support, Domains of Parental Self-Efficacy and Distress in Parents of Young Children
No full textThe present cross-sectional study investigated the relationships amongst social support, parental self-efficacy and postpartum distress. Parents of children under two (N = 327) were recruited via social media and poster advertisements. They were asked to complete three questionnaires regarding social support (Family Support Scale; FSS), parental self-efficacy (Karitane Parenting Confidence Scale; KPCS) and emotional distress (Depression, Anxiety and Stress Scale; DASS).The participants from this study who had high levels of self-efficacy were found to be more likely to have high levels of social support; however, this correlation was not as strong as anticipated. Furthermore, the participants from this study who had high levels of selfefficacy were less likely to be experiencing distress symptoms; the strength of this correlation was moderate and similar to prior findings. To further explore the relationships among these constructs, a post-hoc exploratory factor analysis of the Karitane Parenting Confidence Scale (KPCS) was conducted; yielding two factors: parenting role confidence (PRC) and baby care confidence (BCC). These two factors of the KPCS were found to be correlated with depression, anxiety and stress; parenting role confidence had a stronger relationship with these variables. Additionally, it was found that a combination of parenting role confidence and baby care confidence had substantial power to predict depression, anxiety and stress; social support added a small but significant additional predictive value. These findings suggest that postpartum distress might be mitigated by providing interventions that would improve levels of social support and parenting self-efficacy
Maturation and biomechanical risk factors associated with anterior cruciate ligament injury: Is there a link? A systematic review
No full textObjective: To establish the potential link between sex-specific maturation and biomechanical factors associated with ACL injury during dynamic tasks. Design: Systematic review. Literature search: Five databases (CINHAL®, Cochrane Library, PubMed®, Scopus®, and SPORTDiscus) were searched and monitored until 27 May 2024. Study selection criteria: Cross-sectional, cohort, case-control, or interventional studies reporting one or more biomechanical variable linked with ACL injury and which assessed participants across two or more maturation phases were considered eligible. Data synthesis: Studies were assessed for risk of bias using a modified version of the Newcastle Ottawa Scale and overall quality of evidence was rated using GRADE. Metrics and effect sizes were presented where available. Results: Eighteen included studies examined 400 males, 1377 females, and 315 participants of undefined sex across various maturation phases. The methodological quality of most studies (n = 16) was considered good, and satisfactory for two. Knee abduction angle, knee abduction moment, knee flexion angle, and ground reaction forces were most commonly reported. Knee abduction angles and moments and knee flexion angles were greater in late and post-pubertal females than males and pre-pubertal females during both landing and cutting tasks. When normalised for body mass, ground reaction forces were generally greater in males compared to females overall and for less mature participants for both sexes. Overall quality of evidence was low or medium across the four biomechanical measures. Conclusion: Sex-specific maturation considerations are important in the targeted development and implementation of ACL injury risk identification and prevention strategies
Imagining Pākehātanga: how to be where you are
No full textImagining Pākehātanga: how to be where you are is a creative practice thesis in which I explore the question: in what ways can I imagine Pākehātanga through personal experiences and shared family practices? In framing the contemporary context of Aotearoa / New Zealand I draw on literature relating to Decolonising Theory and Pākehātanga, leading to the development of a research design based on autoethnographic performance. The research was carried out in three phases using the following methods: individual interviews with family members; a session involving a thematic research conversation with a group of my family members; and creative practice throughout the project. The last phase culminated in an evening of autoethnographic performance incorporating creative writing, poetry and song. The performance script and an audiovisual recording of the performance are integrated with the written thesis text. Throughout the thesis, my creative writing weaves with my academic writing, revealing three clear responses to the research question: naming, listening, and longing. These responses function interdependently, together showing the potential for an infinite motion of imagination in action and possibilities for Pākehā practising a relational way of being. This is my quest of how to be, how to be where I am
Could ‘Isochoric Freezing’ revolutionise food preservation?
No full textThe present article responds to the food engineering community’s growing interest in an emerging and lauded approach to food preservation, popularised by its developers as ‘isochoric freezing’. A strong campaign in the scientific literature and mass media has recently promoted this technique as a universal replacement for traditional food freezing and the frozen supply chain by highlighting a number of alleged advantages of ‘isochoric freezing’. Some of these claims therefore require a more neutral and critical assessment against the background of the today’s state of the art in food freezing technologies. Hence, this article spotlights several concerns regarding the plausibility, energy expenditure, resource efficiency, process rate, throughput and safety of ‘isochoric freezing’, as well as the correct use of food refrigeration terminology. The aspects considered are intended to make food scientists, technologists and engineers more aware of the real capabilities and the application perspectives of this still immature mode of refrigerated food processing
Katoa: A descriptive, comparative and historical analysis
No full textKatoa, approximately translated as ‘all, every’, functions as the universal quantifier in te reo Māori (the Māori language), inherited from Proto Polynesian *katoa. Less well known is its occurrence as a postnominal modifier in singular noun phrases, contributing the meaning of ‘whole’, and as a nominal head meaning ‘all (of them), the whole (of it)’. As a postverbal modifier, katoa can quantify a subject noun phrase with the same meanings of ‘all, every’ and ‘whole’, and it can modify situations expressed by the verb, meaning ‘wholly, completely’. In this paper, we examine the behaviour of katoa in te reo Māori, comparing some properties of katoa with the behaviour of quantifiers in related Polynesian languages. This, in turn, informs our understanding of the prehistory of quantification in Polynesian, where we observe evidence of both retention and replacement of quantifier forms
A randomised crossover trial on the effects of foot starting position on calf raise test outcomes: Position does matter
No full textBackground: This randomised crossover study with repeated measures examined the influence of the three most common foot starting positions used in conducting the calf raise test (CRT) on test outcomes. This study also accounted for the potential influence of gender, age, body mass index (BMI), and level of physical activity on test outcomes. Methods: Forty-nine healthy individuals (59 % female, 21 ± 4 years) performed single-leg calf raise repetitions in a human movement laboratory in three randomised foot starting positions: flat, 10° incline, and step. The validated Calf Raise application was used to track the vertical displacement of a marker placed on the foot using computer vision. The application extracted the following CRT outcomes from the vertical displacement curve: number of repetitions, peak vertical height, total vertical displacement, and total positive work. Data were analysed using mixed-effects models and stepwise regression. Results: There was a significant main effect (P females) and BMI (P = 0.002; lower BMI>higher BMI) significantly influenced the number of repetitions. Gender (P females) also influenced total positive work. Age and physical activity levels did not significantly influence CRT outcomes. Conclusions: CRT foot starting position mattered and significantly affected all CRT outcomes. CRT foot starting position needs consideration when contrasting data in research and practice