143174 research outputs found
Sort by
Bio-inspired robotics: efference copies and adaptive feedforward control
No full textThis thesis investigates how insects control movement using adaptive feedforward control and efference copies, and how these principles can be applied to robotic systems. A phenomenological control architecture—the fully-separable-degrees-of-freedom (FSDoF) controller—is developed to formalise efference copies in a control-theoretic context. The FSDoF architecture is used in simulation to examine control advantages of efference copy-based systems. Specifically, it enables sensors to remain within their operating range, thereby optimising sensitivity to external disturbances, and is particularly effective in systems with time delays and significant sensor noise. When combined with high-pass-filter-like sensors, the FSDoF controller also exhibits a unique robustness to changes in sensor gain and bandwidth. Furthermore, a method is introduced to design FSDoF controllers for multi-input multi-output (MIMO) systems with multiple, distinct time delays. This work formalises the efference copy hypothesis in insects from an engineering perspective, offering a functional explanation for their evolutionary advantages, and enabling the generation of experimentally testable hypotheses.
One such hypothesis is that insect feedforward controllers adapt to changes in system dynamics. A series of free-flight experiments in Drosophila melanogaster supports this idea. Following unilateral wing damage, flies exhibited increased flight speeds in darkness compared to intact controls. When flying in light, with visual feedback, their speed was reduced to near-control levels. Altogether, the results indicate that the adaptation was not solely due to changes in mechanosensory feedback, but suggest adaptive feedforward control.
Building on these insights, an implicit adaptive feedforward (IAFF) control strategy is proposed and implemented on a quadrotor platform. This controller uses adaptive sensitivity derivatives, so that it could, in theory, cope with an inversion of the system dynamics. IAFF control improved the reference tracking in a Parrot Mambo mini-drone within minutes of operation, starting from a randomised feedforward controller.Open Acces
Cardiac myosin-binding protein C in ST-elevation myocardial Infarction
No full textBackground and Aims: Cardiac myosin-binding protein C (cMyC) is a novel biomarker of myocardial injury, rising and falling more rapidly than cardiac troponins in myocardial infarction (MI), potentially enabling earlier diagnosis. Its performance has not been assessed in reperfused acute ST-segment elevation myocardial infarction (STEMI), against gold-standard biochemical (high-sensitivity cardiac troponin I, hs-cTnI) or imaging (cardiovascular magnetic resonance, CMR) biomarkers. This study tested the hypotheses that: i) cMyC correlates with acute and final MI size by late gadolinium enhancement (LGE) CMR and ii) cMyC is related to the presence of acute microvascular obstruction (MVO) by CMR.
Methods: Blood samples were obtained at 6±2 hourly intervals for 24 hours (hrs) for measurement of hs-cTnI and cMyC concentrations in patients with reperfused acute STEMI. Patients underwent 3T LGE-CMR at ~ 3-5 days (n=69) and ~ 4 months (n=65) after reperfusion.
Results: Acute cMyC at all timepoints significantly correlated with acute and final MI size on LGE-CMR, most strongly at 6-hrs post reperfusion (r=0.7, p<0.001). cMyC at 6-, 12-, 18- and 24-hrs demonstrated significant discriminatory power in identifying patients with acute MVO, with the 6-hr level having the highest discriminative power. Hs-cTnI correlated more strongly with acute and final MI size compared to cMyC and had significantly higher discriminatory ability in identifying MVO at 12-, 18- and 24-hrs.
Conclusions: cMyC is a quantitative biochemical biomarker of myocardial injury in reperfused STEMI. Further studies, using optimised high-sensitivity assays, are warranted to evaluate its potential as a novel biomarker after acute MI
Towards safer deep learning: verification, robustness, and explainability
No full textDeep neural networks have achieved remarkable success across diverse applications, from autonomous vehicles to medical diagnostics. However, their deployment in safety-critical systems is hindered by vulnerabilities to adversarial attacks, lack of robustness guarantees, and opaque decision-making. This work addresses these fundamental challenges through contributions towards verification, robustness, and explainability of neural networks in both deterministic and probabilistic settings.
We first tackle the problem of verifying neural networks against geometric perturbations, developing a novel piecewise-linear approximation method that approximates non-convex spatial transformations more precisely than existing approaches. This technique produces verification bounds up to 30% tighter than the benchmark methods, enabling formal safety guarantees for 32% more cases on vision-based baselines.
Secondly, we explore the use adversarial training in improving out-of-distribution generalisation in LiDAR-based 3D object detection. By combining a physics-inspired initialisation scheme with adversarial training, we achieve up to 6% improvement in detection accuracy under adverse weather conditions compared to simulation-based methods.
Further, we extend verification techniques to probabilistic settings by developing novel algorithms for Bayesian Neural Networks (BNNs). Our Pure Iterative Expansion and Gradient-guided Iterative Expansion methods dynamically adapt verification regions based on the parameter posterior space, yielding probabilistic robustness certificates up to 40% tighter than previous approaches, while also eliminating the need for extensive hyperparameter tuning.
Finally, we introduce the first formal framework for generating counterfactual explanations on BNNs, leveraging inherent uncertainty quantification to produce explanations with superior plausibility and robustness. Our evaluation demonstrates that BNN-based counterfactuals consistently outperform deterministic and ensemble-based alternatives across multiple metrics and datasets.
These contributions collectively advance the state of the art in neural network safety, providing both theoretical foundations and practical tools for building more dependable AI systems.Open Acces
Special function solutions of complex differential equations and their application to Marangoni flows
No full textThis thesis is concerned with the mathematical study of steady two-dimensional Marangoni flow problems in both planar and radial geometries. By exploiting the recently discovered connection between the forced complex Burgers equation and Marangoni flows in a viscous fluid region, first discovered by D. G. Crowdy, it shall be shown that three physically distinct problems can each be reduced to a single forced complex Burgers equation, satisfied by a complex function that is analytic in the fluid domain. The contribution of this thesis is to demonstrate that, in all three cases, steady equilibrium solutions to these equations are given by special function solutions of certain well-known ordinary differential equations. Specifically, the parabolic cylinder and doubly-confluent Heun equations are considered.
The first physical problem considers a localised concentration of insoluble surfactant occupying an infinite flat interface between a deep viscous fluid region experiencing a linear extensional flow and a constant pressure region. This scenario is then generalised to a two-phase analogue, where the surfactant is now soluble to one of the fluids. In both cases, solutions are found in terms of logarithmic derivatives of the parabolic cylinder function. The solutions and their singularities are analysed in the limit of vanishing surface diffusion using asymptotic techniques and Liouville-Green analysis. These results are then discussed in the context of the Stokes phenomenon for nonlinear ODEs.
The third and final problem considered in this thesis is that of a bubble laden with insoluble surfactant translating steadily through a viscous fluid region that is experiencing a linear temperature gradient. Equilibrium solutions are found in terms of a logarithmic derivative of a solution of the doubly-confluent Heun equation, which is determined by a monodromy condition imposed on the equilibrium solution. A semi-explicit expression for the bubble speed is derived.Open Acces
Dust star-forming galaxies across cosmic time
No full textWhile significant progress has been made in studying dusty star-forming galaxies (DSFGs), many questions remain regarding their physical properties, environments, and evolution. Future advances will depend on high-resolution, multiwavelength observations from current and upcoming facilities, enabling precise measurements of DSFG redshifts, morphologies, and kinematics to clarify their role in cosmic structure formation and stellar mass build-up. To begin with, we investigate the number counts and infrared luminosity functions (IRLFs) for the
far-infrared (FIR) extension of the Euclid mock galaxy catalogue, MAMBO. We find good agreement between our results and the literature. However, applying Euclid and SPIRE/PACS
limits and exploring environmental dependencies reveals that a significantly larger mock sample (∼ 30 times bigger) is required to draw robust conclusions about the IRLF’s environmental
trends for a near-IR/FIR-selected population. We also present a SCUBA-2 analysis of the Herschel -SPIRE Dark Field (SDF), currently the deepest FIR field. We detect 36 sources at 850μm and cross-identify them with ancillary data, identifying 20 as SPIRE-dropouts; candidate high-redshift galaxies. Photometric redshift estimates place the SPIRE-detected sources primarily at 2 3. FIR/submillimetre SED fitting suggests the stacked SPIRE-dropouts contribute ∼ 15% to the cosmic star formation rate density (CSFRD) at z = 6.7. The SPIRE-detected sources show a CSFRD excess at z ∼ 1, possibly hinting at large-scale structure in the SDF. Finally, we present SMA follow-up of 12 SPIRE-dropouts across various fields. Photometric analysis identifies optical/NIR counterparts for four sources, including a multiple system and a potential lens. Four remain without optical/NIR counterparts, and four are undetected with the SMA. FIR/millimetre SED fitting estimates an average redshift of z = 7.4 for the SMA sample. In closing, we discuss the implications of these results and the future avenues of research produced by this work.Open Acces
Sustainable analytical chemistry laboratories: the critical role of education
No full textOver the past 20 years, interest in sustainability principles within analytical chemistry laboratories has increased rapidly. Advocates for green analytical chemistry have sought opportunities to educate themselves and incorporate sustainable practices into their work. As climate change remains a continual concern for current and future generations, there is growing pressure to address the carbon emissions from scientific work. Furthermore, to achieve the UN Sustainable Development Goals (SDGs) and with many funding bodies now placing a greater emphasis on incorporating sustainability within funding applications, there is an increasing need for all scientists to receive education in this field. A pyramid approach is proposed for sustainable science education through which there is a shift from teacher-based learning to student-based learning. Through exploring the different career stages of a scientist, appropriate pedagogical approaches have been identified. This allows a variety of training opportunities to be created to cement sustainable practices into routine approaches. By establishing a strong foundational knowledge in basic scientific practices at school level, students can develop robust, sustainable experimental design and coding skills, which can be carried forward into their later careers. When looking at the further education level, where training becomes more specialised, more technique-specific education is required. While Green Chemistry is beginning to be incorporated into universities through course designs, we identified that training should be mandatory rather than due to interest. For those undertaking continued professional development (CPD), the introduction of greenness metric frameworks and sustainability accreditation schemes are useful tools for upskilling individuals. Despite this, many resources are arguably used primarily by those already interested in the subject. This Perspective urges everyone, at any level, to include sustainability in experimental design and participate in accreditation schemes
Multimodal imaging reveals a lysosomal drug reservoir that drives heterogeneous distribution of PARP inhibitors
No full textFor all drugs, effective target engagement requires sufficient intracellular concentrations of drug to be reached, but whether tumour heterogeneity impacts drug distribution and efficacy is poorly studied. Poly (ADP-ribose) polymerase (PARP) inhibitors have transformed treatment opportunities for women with high-grade serous ovarian carcinoma, but resistance remains a clinical hurdle in this highly heterogeneous tumour type. Here, we present a patient-derived explant multi-modal imaging pipeline, which demonstrates that cell-intrinsic PARP inhibitor accumulation is highly variable, both between patients and within tumours. Spatial transcriptomics reveals enrichment of apoptotic and lysosomal signatures in high-drug regions. Rucaparib, an intrinsically fluorescent PARP inhibitor, accumulates heterogeneously at the single-cell level, with rucaparib high cells demonstrating increased drug response relative to rucaparib-low. Mechanistically, lysosomal sequestration creates a rucaparib reservoir that determines drug levels in the nucleus. Perturbation of lysosomal content alters intracellular levels of weak base PARP inhibitors rucaparib
and niraparib, but not olaparib. Together these data suggest that lysosomes act as a reservoir for a subset of PARP inhibitor drugs to improve drug response
Curiosity, reassurance and convenience: understanding public motivations for using Generative AI in symptom assessment
No full textCapacity management in networks: a structural estimation approach for hospital inpatient wards
No full textProblem Definition: Addressing capacity management within a multifaceted network of resources is a
critical challenge in service operations management. As resources are often shared to serve multiple classes of
customers in such systems, customer routing often depends on the congestion levels of these resources, which in turn are affected by customer routing, creating a feedback loop. Consequently, when evaluating the impact of substantial capacity changes in the network, one must account for the complexity resulting from resource sharing, endogenous routing policies, as well as the feedback loop between routing policies and congestion levels which has the potential to alter the equilibrium of the system. This complexity renders conventional approaches insufficient for an accurate assessment. Methods/Results: To tackle these challenges, we develop a structural estimation approach that relies on two key components. First, we estimate the routing policy via a choice model that allows the routing policy to depend on not only the focal resource’s utilization but all connected resources’ utilization. We adopt a control function approach with instrumental variables to estimate the loads’ effect in routing without bias. Second, we incorporate the estimated routing policy
into a queueing network model which captures the detailed system dynamics and evaluates the equilibrium
performance of the entire network. We apply our approach to the specific empirical setting of the hospital inpatient ward network. We show that our proposed approach outperforms two alternative models and
highlight the importance of accounting for network equilibrium effects when evaluating substantial capacity
changes. Managerial Insights: We provide prescriptive capacity allocation recommendations to hospital
managers. More generally, our findings underscore the importance of a comprehensive understanding of the
interdependencies between customer routing decisions and the levels of congestion present at various resources, shedding light on broader strategies for improving the operational performance of service networks
