143174 research outputs found
Sort by
Investigating the impact of geological heterogeneity on groundwater flow in the permo-triassic sandstone aquifer system of the eden valley, cumbria (uk)
No full textThis thesis investigates the impact of geological heterogeneity on regional groundwater flow in the Permo-Triassic Sandstone aquifer system of the Eden Valley, Cumbria. Four distinct studies were conducted to address this aim, each focusing on a different aspect of hydrogeology. Study one analysed groundwater time series data to correlate hydrological signatures with geological heterogeneity. Three hydrogeological regimes were identified using seasonal trend decomposition by LOESS (STL), revealing that while meteorological factors influence groundwater seasonality, they do not dominate other groundwater flow behaviours. Study two developed and installed low-cost ultrasonic river level monitoring sensors to collect hydrological data, which led to two key insights: (1) Scandal Beck loses water to the southern Penrith Sandstone Brockram facies as it transitions from Carboniferous to Permo-Triassic geology, and (2) Knock Gill has a net-zero water exchange with the underlying St Bees Sandstone. These findings offer a novel contribution to understanding the region's hydrogeology. In study three, a geological and hydrogeological model of the Permo-Triassic Sandstone aquifer was developed, incorporating new pumping test data and borehole log information. Finally, study four built a numerical groundwater model, demonstrating that the Penrith Sandstone is well-characterized by a three-layer system, while the St Bees Sandstone remains poorly defined. This highlights the need for further regional characterization of the St Bees Sandstone. The impact of geological heterogeneity on groundwater flow was evaluated by simulating groundwater level time series using models of varying complexity. The analysis revealed that geological heterogeneity is a dominant factor influencing the observed groundwater level fluctuations. The findings underscore the importance of incorporating geological heterogeneity into regional groundwater flow models to improve the accuracy of groundwater forecasting and management in the region.Open Acces
Extracting and comparing machine learning-based representations across natural and medical modalities
No full textMachine learning has led to breakthroughs across many domains, such as Computer Vision, Natural Language Processing, and Speech Recognition. Given these breakthroughs, machine learning methods are being applied to more and more fields, such as healthcare, education, and legal services. However, the data required to train such models can be sensitive and limited, making it challenging to develop models in such fields. To ensure the applicability of Artificial Intelligence in such sensitive fields, we need efficient, reliable methods that consider privacy preservation. In this thesis, I aim to address these challenges by comparing representations, evaluating their robustness to incomplete data, and preserving privacy through machine unlearning methods.
I introduce several key contributions across these domains:
I propose the Feature Impact Balance (FIB) score, a metric to assess error distribution, and the Min-Max Relative Change Quadrant (MMRCQ) plot, a visualization tool to monitor changes in extreme values. I explore the class-wise impact of data augmentation in image-based models, showing that augmentations affect different classes unequally. Data augmentations can make classes less distinguishable, and such an effect is class-dependent. I evaluate the transferability of foundation models trained on non-medical data to domains with limited labeled data, demonstrating their potential for medical applications such as dysarthria automated assessments. I show that foundation models trained on large-scale speech from healthy individuals can be used to detect dysarthria, classify words, and classify intelligibility. I also assess the robustness of self-supervised learning (SSL) models to incomplete data.
Finally, I benchmark machine unlearning methods, making models "forget" data points without losing overall performance. This benchmark shows that unlearning algorithms must be compared against stronger baselines with extensive hyper-parameter searches. This thesis advances the development of more robust, reliable, and privacy-conscious machine learning models.Open Acces
Enhancing stream data processing: system optimizations and learned indexes
No full textThis thesis aims to optimize stream data processing, which is important for real-time data analysis and decision-making. Stream data’s inherent properties, including unbounded size, high volume, and variable velocities, impose significant challenges on processing systems. These systems must continuously evolve to meet the requirements of modern stream data processing. This thesis presents optimizations to enhance the functionality, scalability, and performance of stream data processing at both the system and algorithmic levels.
At the system level, we focus on a stream system, dispel4py, designed for scientific workload computation. We enhance the scalability and state management of dispel4py by developing dynamic allocation, dynamic auto-scaling and hybrid optimizations. Specifically, dynamic allocation allows dispel4py to scale for each task depending on the workload demands, and dynamic auto-scaling enables the entire workload to scale with fewer or more resources to maintain the performance while achieving cost efficiency. Furthermore, hybrid enables dispel4py to support stateful tasks and scaling simultaneously. Comprehensive experiments validate the scalability, portability, and performance of these three optimizations.
At the algorithm level, our focus shifts to Index-Based Window Processing (IBWP). Recently, learned indexes integrating machine learning models to enhance query performance present a promising alternative to traditional index structures. Motivated by this trend, we explore how learned indexes can effectively support search while maintaining updates for high-velocity data streams. However, the challenge lies in the inherent limitations of current updatable learned indexes. These limitations are often inherited from their traditional tree-based structures, which are cumbersome and impede update performance. To overcome these limitations, we pioneered the use of innovative queue-style flat structures, which significantly enhance update efficiency and reduce the index footprint. Based on the flat structures, we propose FLIRT and SWIX, designed for sequential IBWP and generic IBWP, respectively. Our experiments demonstrate that they effectively manage their respective IBWPs, outperforming all baselines.Open Acces
The protective mechanisms of soap against schistosome cercariae in water
No full textSchistosomiasis is a parasitic water-based disease which affects almost 240 million people worldwide currently. This disease occurs because of skin contact with water containing the parasite larvae, schistosome cercariae. The use of soap is central to many hygiene practices and might play a role in schistosomiasis prevention by reducing the penetration of cercariae into human skin during water-contact activities.
This PhD thesis determined the efficacy of soap against cercariae via potential protective mechanisms. Firstly, a systematic review revealed that soap might protect individuals by directly harming cercariae or by protecting skin against cercarial infectivity. However, the knowledge was insufficiently detailed to inform soap use practices that would prevent infection. A novel method of using tails of deceased mice was then developed to study cercarial infectivity. Laboratory experiments were performed to investigate three potential protective mechanisms of soap: (1) killing cercariae, (2) impairing cercarial infectivity, and (3) protecting skin against cercarial infectivity. The latter two mechanisms, related to cercaria infectivity, were tested using the mouse tail method that was established. Powder and bar soaps that are used by people living in a schistosomiasis-endemic village in Tanzania were included in the research. For the first mechanism of soap lethality, all soaps were able to kill cercariae, with their efficacy related to the soap concentration and exposure time. Among the second and third mechanisms, only the third mechanism, protection of tails after soap treatment, reduced cercarial infectivity, but this result was observed only with powder soap.
This research provides scientific evidence and new understanding of the protection provided by soap against schistosomiasis infection, via direct killing of cercariae and protection of skin. However, due to the limited protection that soap provides, it can play a complementary role, e.g. alongside preventive chemotherapy and/or mollusciciding, rather than a primary one in overall schistosomiasis prevention strategies.Open Acces
The evolution of cirrus clouds from deep convection
No full textAerosols significantly impact the climate through their interactions with clouds, where the magnitude and uncertainty of the radiative forcing of aerosol-cloud interactions vary by cloud type. This uncertainty is particularly pronounced for deep convective clouds. Tropical deep convection and its associated cirrus outflows have a near-zero net cloud radiative effect (CRE) due to the offset between shortwave cooling and longwave warming. Therefore, minor changes in the properties of deep convection or associated anvil cirrus, such as those caused by aerosols, could impact the net CRE. Understanding what controls the radiative evolution of deep convection is vital to better constrain aerosol-cloud interactions. This thesis introduces a novel method to examine the evolution of deep convective clouds, from the short-lived, optically-thick convective cores to the thin detrained cirrus that can persist for days after the initial convection has dissipated. The radiative evolution of clouds along trajectories from deep convection is investigated, revealing a positive total high cloud CRE. The anvil cirrus lifetime is calculated, with longer lifetimes observed for detrained cirrus from oceanic rather than terrestrial convection. Longer lifetimes increase the total high cloud CRE. It is found that stronger convection produces detrained cirrus with greater warming over their entire lifetime, primarily driven by changes in the SW CRE due to diurnal variability in convection. Finally, the sensitivity of the detrained cirrus CRE to the convective strength is linked to the sensitivity of the high cloud top pressure to the aerosol optical depth, accounting for the non-local impact aerosols may have on detrained cirrus some distance away from the initial convection. This thesis provides an upper bound on aerosol impacts on tropical high cloud CRE. By introducing a novel equation to quantify non-local aerosol effects, it provides a framework for assessing such impacts under various scenarios, enhancing our understanding of aerosol-cloud interactions.Open Acces
A general extreme value-based Gaussian global navigation satellite systems measurement error distribution for mission-critical applications
No full textGlobal Navigation Satellite Systems (GNSS) positioning and integrity monitoring models and algorithms currently generically assume that measurement errors follow a Gaussian distribution. As this is not always the case, there is a trade-off affecting system safety and availability, emphasising the need for better error characterisation in mission-critical applications. Research to date has shown advantages of Generalised Extreme Value (GEV) distribution for mapping extreme events. However, it is more complex than the Gaussian distribution, especially in the error convolution process. This paper derives a distribution, referred to as the GEV-based Gaussian distribution, that benefits from the advantages of both the GEV and Gaussian distributions in mapping extreme events and simplicity, respectively. The proposed distribution is tested against Gaussian, GEV and Generalised t distribution. The results show that the proposed distribution can provide a better bound for extreme events than the tested distribution both for pseudorange and carrier phase errors
Non-invasive bidirectional acoustoelectric neural interface
No full textThe non-invasive sensing and stimulation of electrical signals in the brain, with high spatial and temporal specificity, has long been a strategic goal in neuroscience. Such a tool would open pathways to understanding the brain and its disorders. To date, the focal detection and stimulation of electrical signals in the brain has been impossible due to the inability to non-invasively reach deep brain areas with high spatial specificity. This thesis proposes a new hybrid modality capable of detecting and stimulating electrical signals in the brain with improved spatial specificity, through utilization of the acoustoelectric interaction.
The underlying physics of the acoustoelectric interaction is investigated, leading to a new mathematical model which encapsulates the vector nature of the interaction. This acoustoelectric mathematical model predicts that heterodyning will occur between the electric and acoustic field, enabling precise signal recovery using techniques previously established in radio communications. Simulation and acoustoelectric phantom characterization demonstrate how this technique can be used to create a focal low frequency electrical signal deep in the brain for neuromodulation and demodulate complex electrophysiological signals at the focus of the ultrasound for acoustoelectric neural recording. Moving to in vivo validation using electrophysiology in a rodent model, we acoustoelectrically demodulated visual evoked potentials around the carrier frequency of the ultrasound for the first time, with associated artefact tests. Finally, the first in vivo evidence of acoustoelectric neuromodulation is reported and an evidence-based argument developed as to why the acoustoelectric interaction is a contributing mechanism to ultrasound brain stimulation. Acoustoelectric neuromodulation has a clear mechanism of action, harnessing the focality of ultrasound and the well-understood direct electrical pathway developed by the Hodgkin-Huxley model, giving it a unique set of advantages to target deep areas of the brain.Open Acces
Malaria chemoprophylaxis provision in the UK: a call to evaluate the cost-effectiveness and equity implications of universal provision
No full textThis position paper advocates for the re-evaluation of the cost-effectiveness and equity of National Health Service-subsidized malaria chemoprophylaxis, considering changes in UK malaria epidemiology, travel patterns, updated travel medicine guidance, novel Plasmodium falciparum treatment pathways and growing awareness and action to tackle sources of health inequities
Beam prediction based on large language models
No full textIn this letter, we use large language models (LLMs) to develop a high-performing and robust beam prediction method. We formulate the millimeter wave (mmWave) beam prediction problem as a time series forecasting task, where the historical observations are aggregated through cross-variable attention and then transformed into text-based representations using a trainable tokenizer. By leveraging the prompt-as-prefix (PaP) technique for contextual enrichment, our method harnesses the power of LLMs to predict future optimal beams. Simulation results demonstrate that our LLM-based approach outperforms traditional learning-based models in prediction accuracy as well as robustness, highlighting the significant potential of LLMs in enhancing wireless communication systems
Oscillometry in idiopathic pulmonary fibrosis
No full textBackground: Idiopathic pulmonary fibrosis (IPF) is a chronic, progressive lung parenchymal disease associated with lung stiffening and reduced compliance. Current clinical practice relies on spirometry, primarily forced vital capacity (FVC), for monitoring and prognostication; however some individuals with IPF are unable to provide technically acceptable or reproducible results due to symptom burden associated with the need to perform repeated maximal forced manoeuvres. Impulse oscillometry (iOS) is a non-volitional lung function test that provides information on airways resistance and lung reactance, and may be a plausible alternative pr adjunct to spirometry. However limited data exist around the value of iOS in IPF.
Aims: To evaluate the symptom burden, time taken to perform, inter-occasion reliability, validity, minimal important difference (MID), and prognostic ability of iOS measurements in IPF.
Methods and results: Oscillometry took less time to perform and was associated with lower symptom burden scores than spirometry. Measures of iOS had good test-retest reliability in a cohort of 66 stable patients with IPF, when measured two weeks apart, and there were moderate strength correlations with conventional lung function measures at a single timepoint (n=78). Longitudinal change in iOS did not correlate with change in FVC, exercise capacity or health related quality of life. Distribution-based methods were used to identify MIDs of oscillometry measures over 6 and 12 months (n=132). Insufficient adverse events occurred over the 12 month follow-up period to provide robust evidence with regards to the prognostic value of iOS.
Conclusion: In patients with IPF, iOS is a quicker to perform than spirometry, and is associated with a lower symptom burden. It has good test-retest reliability, with moderate strength correlations with conventional lung function tests at a single timepoint. However, the value of this test as a longitudinal marker of lung function and prognosis requires further investigation.Open Acces