21684 research outputs found
Sort by
The role of physical activity in relation to later suicidal outcomes. Longitudinal exploration using quantitative prospective cohort design
Get PDFAbstract available at each chapter
Love and joy with The Discovery Book Club: to what extent can children’s literature be used for effective human rights education?
No full textAbstract not currently available
Predicting retrieval failures in conversational recommendation systems
Get PDFIn recent years, the use of dialogue systems and voice assistants commonly implemented in smart devices has shifted the users’ interest towards online shopping. In turn, online shopping platforms are gaining popularity and moving towards allowing an interactive dialogue with users that more accurately depicts a real shopping setting. In this regard, the task of Conversational Image Recommendation is the state-of-the-art task for conversational recommendation in the fashion domain, where a user has a specific fashion item in mind, and interacts with the system with natural language feedback on recommended image items, which guides the system in finding the imagined item in the next turn. Such systems are trained and evaluated with user simulators as a plentiful surrogate for human users. A practical problem with CRS performance is that it is primarily evaluated in terms of successes and is therefore assumed to return the item of interest by a pre-defined number of turns. In practice, often the item is not returned by the end of a conversation, therefore leading to conversational failures; this is our particular setting of interest.
In this thesis, we argue that the performance of a Conversational Recommendation System can be predicted to detect when a conversation fails, under different scenarios, across different turns of a conversation. In this regard, Query Performance Prediction (QPP) techniques predict the effectiveness of a ranked list result in response to a query without having access to relevance judgments. We predict the performance of CRS models by treating them as dense retrieval processes, where both the image retrieved items and textual feedback can be represented with dense embedded representations. In particular, we propose a set of coherence-based dense QPPs specifically designed for single-representation dense retrieval models (ANCE and TCT-ColBERT) and show that the examination of the relations among dense embedded representations already contained in the document list is sufficient to provide effective predictions for dense retrieval models. At the same time, by using a multi-level perspective that jointly considers QPPs and types of queries, we explain why some QPPs are better for certain types of queries, thus explaining discrepancies among different evaluation metrics.
At the next stage, we predict the effectiveness of a ranking of image items in Conversational Image Recommendation models, which are also based on learned embedded representations of images, and where user feedback takes the place of a textual query. In deed, we create a novel task which we call Conversational Performance Prediction (CPP), which predicts conversation success at the conversation level and taking into account the multi-turn nature of the task, and can differentiate between success predicted over a short-term and a long-term horizon, thereby predicting current user satisfaction or overall satisfaction of a conversation. First, we examine the set of unsupervised predictors developed for dense retrieval models but applied to state-of-the-art Conversational Image Recommendation models; a GRU-based model, which mainly considers the feedback of the previous turn, and an EGE model that considers the entire dialogue history. Our results show that using correlations is not an optimal evaluation strategy for predicting conversational failures, as, while correlations are low to medium mainly for short-term predictions, a lot of inconsistencies are observed among the performance of different predictors across metrics and datasets (similarly to dense retrieval models). Consequently, we propose a supervised CPP approach, which treats CPP as a binary classification task, which predicts whether a target item is returned by a given turn. In this way, we show that by learning the embedded representations already contained in the CRS models, we can predict the accuracy of a conversation success using the retrieved items of both single and multiple turns.
In addition, state-of-the-art CRS models are trained using user simulators with a single target item in mind, and at the same time, they are assumed to be infinitely patient. These settings do not reflect a real shopping scenario, where a user might change their mind according to what a shopping assistant is suggesting. For this purpose, we enhance the evaluation completeness of CRS models by obtaining real user opinions in a user study using pooling similar to information retrieval tasks, thus identifying alternative relevance labels for several target items, and in turn, inform the user simulator with an extended target space. This increases the completeness of CRS evaluation, and therefore, creates a more realistic prediction setting for CRS, which leads to improved predictions of user preferences. Indeed, when we reevaluate the CRS models using the updated simulator with the identified alternatives as part of the target space, we show that by the single target setting previously used to evaluate CRS models for a maximum amount of 10 turns was underestimating the effectiveness of CRS models.
As a final step, we account for the fact that CRS models assume only one type of recommendation failure, namely the inability of the system to retrieve the target item. In this regard, we introduce the concept of recommendation scenarios, and specifically, we adapt our CPP framework for different types of conversational failures, which are determined by whether the user’s need is clearly defined and whether the target item is available. Therefore, we propose the removed target scenario (the target is not available in the catalogue), and the alternative scenario (a user has a more flexible need, which can be satisfied by either the original target or any of the identified alternatives in the collected datasets). Consequently, we detect different types of conversational failure, such as when a user cannot find an item, versus when the system’s catalogue does not contain the relevant item. By examining the supervised CPP predictors introduced under these two novel scenarios, we find that in both cases, there is a marked difference from the original scenario, and that CPP can indeed be predicted for different recommendation scenarios
Investigating the impact of early life housing on play behaviour in dairy calves
Get PDFPlay behaviour is widely recognised as an indicator of positive welfare states in dairy calves but it’s measurement has traditionally relied on direct behavioural observations which are labour intensive and not always suitable for studies conducted on farms. The growth of wearable accelerometer technology within the farming industry has given researchers a practical opportunity to measure play behaviour in real-time. By utilising accelerometer technology to measure calf play, the welfare impacts of different calf management systems can be easily compared. The early life housing experience of dairy calves is known to impact their development, but the immediate and long-term impacts on play behaviour are not well understood.
This research first validated IceTag accelerometer technology (Peacock Technology, UK) to measure play behaviour in weaned dairy calves. Eight female dairy calves aged three to five months old were monitored using leg-mounted accelerometers and closed circuit television cameras for a 48-hour recording period. The validation process evaluated the correlation between visual observations of weaned calf play and IceTag motion index (MI) data output, then used classification and regression tree analysis to establish a MI threshold value which would best indicate play. A MI value of ≥ 69 was established as the optimum threshold to detect play behaviour in weaned dairy calves (sensitivity = 94.4%; specificity = 93.6%; balanced accuracy = 94.0%).
The second part of this study utilised accelerometer technology to measure the immediate and long-term welfare impacts of different early life dairy calf housing conditions. A total of 96 female dairy calves were recruited from four Scottish dairy farms and assigned to individual, paired or group housing at birth. Play behaviour was measured using IceTags in the same cohort of calves for two 48-hour recording periods: neonatal (calves aged 24 to 72 hours old) and weaned (calves aged three to five months old). Mixed effect negative binomial regression models were used to assess the impact of early life social housing on neonatal calf play behaviour and to assess the impact of early life social housing and early life playfulness on weaned calf play behaviour. Compared to calves housed individually, calves housed in pairs (IRR = 1.29; p = 0.002) and calves housed in groups (IRR =1.43; p < 0.001) performed significantly more neonatal play. No difference in neonatal play was found between calves housed in pairs versus calves housed in groups (IRR = 1.11; p = 0.334). No significant effect of previous early life housing type nor early life playfulness was found on weaned calf play.
Collectively, the results presented in this study demonstrate that accelerometer technology can be utilised to measure welfare in dairy calves under various management systems. This study highlights the importance of validating accelerometer devices to measure animal play behaviour in a research setting and presents the potential for extension of this technology into a commercial environment for use as a farm-based welfare monitoring tool. The findings of this work contribute to the growing body of evidence indicating that social housing provides calves with a more positive early life experience than individual housing. Though no relationship between early life experience and play behaviour in weaned calves was found, this study highlights the need for further research to understand the management factors which influence weaned calf play
Characterization of small molecule inhibitors of RUNX1 for the treatment of cardiac pathologies
No full textAbstract not currently available
Hear us now: creating space for authentic portrayals of trauma in drama and prose through the use of oral storytelling and verbatim theater
No full textAbstract not currently available
Stigma and self-perceptions among individuals with intellectual disabilities
Get PDFAbstracts available at each chapter
Inferring the neutron star equation of state using Machine Learning methods
Get PDFThe first decade of gravitational wave (GW) detection using the global ground based GW detector network has facilitated a new era of neutron star observation. From the GW signal produced when two neutron stars (NSs) inspiral and merge, one can directly measure the masses of the two NSs and, importantly, their tidal deformability, a direct measure of the behaviour of matter in the system. This parameter is unique to GW astronomy and therefore offers an independent method to infer the neutron star (NS) equation of state. Due to lack of precision in measurement of NS macroscopic parameters, the equation of state – the relationship between the pressure and density within the ultra-dense neutron-rich matter of a neutron star – is still widely unknown. Though there are various inference schemes to infer the NS equation of state given electromagnetic (EM) and GW observation, these are often computationally and temporally expensive processes.
Recently, the introduction of machine learning (ML) tools in astronomical data analysis have facilitated the handling of large amounts of data and the processing of this data efficiently, to find broad trends or features. These tools will become necessary when considering future GW detection, where we expect increased sensitivity of detectors as well as orders of magnitude more detections, including those of binary neutron star (BNS) mergers. In this thesis, we apply ML methods, notably a type of generative ML model called a Normalising Flow, in developing tools through which we can infer the NS equation of state in current and future observation of gravitational waves (GWs) from BNS mergers.
We firstly introduce a Normalising Flow trained to perform the mapping of equation of state data conditioned on BNS event parameters. Once trained, the Flow can be conditionally sampled to return an equation of state posterior given posterior samples from a single GW event in less than 1 second. Simulation studies demonstrate the validity of the Flow result, alongside the equation of state posterior for the GW event GW170817, which is in agreement with the existing accepted result. The tool facilitates rapid follow-up of GWs from BNS mergers for improved communication with EM astronomers.
In setting the scene for hierarchical inference of the NS equation of state given multiple observations of GWs from BNS mergers, we discuss the performance of Normalising Flows in mapping complex high-dimensionality data sets. The introduction of a new equation of state training data set makes use of an autoencoder for data compression, which achieves root-mean-squared (RMS) error on the equation of state reconstruction on the scale of 10⁻³ for normalised mean-subtracted equations of state. We demonstrate abnormalities in the Normalising Flow’s performance in mapping regions of the equation of state space, which manifests as severe spikes and troughs of probability. We highlight the dangers of inconsiderate application of Normalising Flows to mapping any high-dimensionality data set. We finally introduce the regeneration Flow, built to learn the mapping of the joint data and conditional spaces at once, such that it can be sampled repeatedly during training for unlimited training data generation. We demonstrate how this improves Normalising Flow training and reduces the fluctuations in magnitude of probability over the surface of the learned data space, promoting generic learning.
We apply the improved Normalising Flow to hierarchical analysis of the neutron star equation of state, firstly in inferring the combined equation of state given the first two BNS merger observations. We make use of a full ML parameter estimation (PE) pipeline to perform a simulation study of inferring the true equation of state given multiple simulated BNS events associated to three known equations of state. We demonstrate that as we increase the number of events, the quality of sampling the equation of state posterior decreases, suggesting a highly multi-modal space and/or inaccurate model. We introduce an alternative method for hierarchical inference which is more robust by using the Normalising Flow instead to sample. With the new method, the result of combining information from up to 16 BNS events associated to two out of three simulated equations of state produce a constrained equation of state posterior which agrees with the truth. We highlight the computational expense of the workflow; inference of up to 16 events with the new method takes less than 1 hour. This validates the use of Normalising Flows for hierarchical inference of the NS equation of state in future observing runs, when the number of events are expected to be in the 10s. We suggest substantial future work to improve the sampling quality and, beyond this, tests for the next generation of GW detection to validate the use of Normalising Flows in understanding neutron star matter
In the shade of shadows: hauntology of partition in the literatures of postcolonial Cyprus
Get PDFNo abstract available
The diagnostic use of metabolomics for the identification of secondary infections in critical coronavirus disease 2019
Get PDFBackground:
Critically ill patients with coronavirus disease 2019 are at high risk of developing secondary infections, which pose a challenge to identify clinically. Empirical antibiotic usage in this group is therefore high. Identification of novel biomarkers of secondary infections would minimise unnecessary antibiotic usage while ensuring that patients with secondary infections receive appropriate antibiotics as early as possible. This project aimed to investigate whether metabolomics could produce a panel of biomarkers capable of distinguishing critically ill coronavirus disease 2019 patients with and without secondary infections.
Methods:
Blood samples were collected from patients in critical care with coronavirus disease 2019, along with a group of healthy volunteer controls. Using high performance liquid chromatography-mass spectrometry, metabolites which showed significant differences in abundance between patients with and without secondary infections were identified. A panel of metabolites capable of distinguishing Gram positive and negative infections was also explored.
Results:
A total of 105 patients were recruited to the study, of whom 40 developed a secondary infection during the trial period. The metabolites creatine and 2-hydroxyisovalerylcarnitine were significantly increased in patients with secondary infections, while S-methyl-L-cysteine was significantly reduced. This metabolite panel demonstrated good diagnostic performance with an AUROC of 0.83. The panel of metabolites distinguishing Gram positive and negative infections consisted of betaine, N(6)-methyllysine and four phosphatidylcholines. This panel performed with high accuracy, with an AUROC of 0.88.
Conclusion:
Metabolomic profiling may be used to identify biomarkers of secondary infections in critically ill coronavirus disease 2019 patients. Investigation of biomarkers for secondary infections in other critical illnesses should be explored
Data sources:
Access Repository Dashboard
Do you manage Open Research Online? Become a CORE Member to access insider analytics, issue reports and manage access to outputs from your repository in the CORE Repository Dashboard! 👇