324139 research outputs found
Sort by
Adaptive maximization of social welfare
Get PDFWe consider the problem of repeatedly choosing policies to maximize social
welfare. Welfare is a weighted sum of private utility and public revenue. Earlier
outcomes inform later policies. Utility is not observed, but indirectly inferred.
Response functions are learned through experimentation.
We derive a lower bound on regret, and a matching adversarial upper bound
for a variant of the Exp3 algorithm. Cumulative regret grows at a rate of T2/3.
This implies that (i) welfare maximization is harder than the multi-armed bandit
problem (with a rate of T1/2 for finite policy sets), and (ii) our algorithm achieves
the optimal rate. For the stochastic setting, if social welfare is concave, we can
achieve a rate of T1/2 (for continuous policy sets), using a dyadic search algorithm.
We analyze an extension to nonlinear income taxation, and sketch an extension
to commodity taxation. We compare our setting to monopoly pricing (which is
easier), and price setting for bilateral trade (which is harder)
Making contract-breakers pay
No full textThis paper examines a deceptively simple question: when can one contracting party obtain a court order requiring the other to pay a sum that the other has agreed to pay? Somewhat surprisingly, the English lawyer’s traditional answer is: ‘It depends’. If the promise was to pay consequent on a breach of contract, it will be enforceable only if it is not a ‘penalty clause’; on the other hand, where the promise creates a contractual debt, enforceable by an action for the agreed sum, the defendant will be ordered to pay in almost all cases where the liability to pay has accrued. This paper suggests that there is instead one broad remedial principle which governs the enforceability of contractual promises to pay money, irrespective of whether or not the payment is conditional on a breach of contract: viz., that the court will not order a defendant to perform its promise where the claimant has no legitimate interest in that remedy. To vindicate this claim, this paper will argue that the operation of both the ‘penalties rule’ as well as the ‘legitimate interest bar’ in White & Carter (Councils) Ltd v McGregor [1962] A.C. 413 has been misunderstood
Theory of Cation Solvation in the Helmholtz Layer of Li-Ion Battery Electrolytes
Get PDFThe solvation environments of Li+ in conventional nonaqueous battery electrolytes, such as LiPF6 in mixtures of ethylene carbaronate (EC) and ethyl methyl carbonate (EMC), are often used to rationalize transport properties and solid electrolyte interphase (SEI) formation. Solvation environments in the compact electrical double layer (EDL) next to the electrode, also known as the Helmholtz layer, determine (partially) what species can react to form the SEI, with bulk solvation environments often being used as a proxy. Here, we develop and test a theory of cation solvation in the Helmholtz layer of nonaqueous Li-ion battery electrolytes. First, we validate the theory against bulk and diffuse EDL atomistic molecular dynamics (MD) simulations of LiPF6 EC/EMC mixtures as a function of surface charge, where we find the theory can qualitatively capture the solvation environments. Next, we turn to the Helmholtz layer, where we find the main effect of the solvation structures next to the electrode is an apparent reduction in the number of binding sites between Li+ and the solvents, again where we find reasonable agreement with our developed theory. Finally, by solving a simplified version of the theory, we find that the probability of Li+ binding to each solvent remains equal to the bulk probability, suggesting that the bulk solvation environments are a reasonable place to start when understanding battery electrolytes. Our developed formalism can be parametrized from bulk MD simulations and used to predict the solvation environments in the Helmholtz layer through reducing the number of available coordination sites, which can be used to determine what could react and form the SEI
An exploration of the impact of amyloid-β on intracellular Ca2+ signalling and metabolic pathways in models of Alzheimer’s Disease
Get PDFIn the field of Alzheimer’s Disease (AD) research, there’s a growing consensus that therapeutic intervention to significantly modify disease progression must begin at the preclinical stages of AD. By investigating early disease mechanisms, researchers can develop targeted therapies aimed at preventing the onset of cognitive symptoms. The amyloid hypothesis of AD suggests that the accumulation of Aβ peptides, particularly Aβ oligomers, is a central and early event in AD pathogenesis. Aβ oligomers are neurotoxic and cause synaptic dysfunction, eventually resulting in synapse loss and neuronal death, which contributes to the cognitive decline observed in AD patients. The mixed results observed in clinical trials of antibodies targeting Aβ highlight the necessity of identifying alternative molecular targets for AD therapeutics. Furthermore, the exact role of Aβ in AD pathophysiology is still under constant investigation. Extensive research efforts have focused on characterising the effects of Aβ oligomers at the pre- and postsynaptic compartment, intracellular Ca2+ signalling, and structural plasticity of dendritic spines. Although, there are still significant gaps in our understanding, which may delay the comprehension of the relationship between early disease mechanisms and late-stage cognitive decline in AD. In this thesis, I aim to advance the understanding of the role that Aβ oligomers play in synaptic function and Ca2+ signalling mechanisms in CA1 hippocampal neurons by using synthetic Aβ oligomers applied to hippocampal neurons and slice cultures, as well as in acute hippocampal slices of J20 mice, an AD model. Using a combination of electrophysiology and imaging techniques, I firstly investigated the effects of a presynaptic Ca2+ channel, Cav2.1, in mediating Aβ oligomer toxicity. A heterozygous knockout of Cav2.1 normalised presynaptic function and rescued Aβ-induced LTP impairment, while preserving basal neurotransmission. Secondly, I assessed lysosomal and endoplasmic reticulum (ER) activity-dependent dynamics and Ca2+ release, and the effects on structural plasticity in response to Aβ oligomer treatment. Aβ oligomers were able to disrupt specific features of lysosome dynamics and recruit Ca2+-induced Ca2+ release (CICR) from the ER in response to back-propagating action potentials in neuronal dendrites. Finally, 1H Nuclear Magnetic Resonance (NMR) metabolomics was used to identify metabolite changes throughout regular aging and AD pathogenesis in tissues from wild type and J20 mice. Both regular aging and early/late stage AD induced specific metabolic signatures, which may contribute to the discovery of novel biomarkers associated with metabolites altered in AD
Cross-correlating the EMU Pilot Survey 1 with CMB lensing: Constraints on cosmology and galaxy bias with harmonic-space power spectra
Get PDFWe measured the harmonic-space power spectrum of Galaxy clustering auto-correlation from the Evolutionary Map of the Universe Pilot Survey 1 data (EMU PS1) and its cross-correlation with the lensing convergence map of cosmic microwave background (CMB) from Planck Public Release 4 at the linear scale range from to 500. We applied two flux density cuts at and mJy on the radio galaxies observed at 944MHz and considered two source detection algorithms. We found the auto-correlation measurements from the two algorithms at the 0.18 mJy cut to deviate for due to the different criteria assumed on the source detection and decided to ignore data above this scale. We report a cross-correlation detection of EMU PS1 with CMB lensing at 5.5 , irrespective of flux density cut. In our theoretical modelling we considered the SKADS and T-RECS redshift distribution simulation models that yield consistent results, a linear and a non-linear matter power spectrum, and two linear galaxy bias models. That is a constant redshift-independent galaxy bias and a constant amplitude galaxy bias . By fixing a cosmology model and considering a non-linear matter power spectrum with SKADS, we measured a constant galaxy bias at mJy ( mJy) with ( ) and a constant amplitude bias with ( ). When is a free parameter for the same models at mJy ( mJy) with the constant model we found ( ), while with the constant amplitude model we measured ( ), respectively. Our results agree at with the measurements from Planck CMB and the weak lensing surveys and also show the potential of cosmology studies with future radio continuum survey data
Age‐specific all‐cause mortality rates among adolescents and youth living with and without HIV: Evidence from a cohort study in South Africa
Get PDFIntroduction: Mortality among adolescents living with HIV (ALHIV) remains a global health problem. We lack granular (age‐ and sex‐disaggregated) data on mortality among ALHIV, hence, this study aims to assess all‐cause mortality among ALHIV in a low‐resource setting. Methods: All adolescents ever initiated on antiretroviral treatment (ART, N = 1107) and their HIV‐negative peers (N = 456) aged 10–19 years, recruited as part of the Mzantsi Wakho study cohort, were followed up between 2014 and 2022 (yielding 12,427.7 person‐years of follow‐up). First, we assessed the proportion of deaths and estimated crude mortality incidence rates per 100 person‐years of follow‐up and their 95% confidence intervals, stratified by HIV status, sex and mode of HIV acquisition (vertical vs. sexual). We then estimated adjusted incidence rate ratios (IRRs) using Poisson regression adjusted for time‐varying age, sex and time on ART. Last, we used the Cox proportional hazards regression model to estimate the risk of death by ART adherence. Results: A total of 1563 adolescents and young people were included in this analysis, 70.8% ALHIV and 57% female. More deaths occurred in ALHIV compared to their HIV‐negative peers (8.3% vs. 0.4%, p<0.001). Among ALHIV, we observed a significantly higher proportion of deaths among males compared to females (10.7% vs. 7.1%, p = 0.036). Overall, mortality increased significantly with age, and males had a higher risk of mortality compared to females. Adolescents and youth living with vertically acquired HIV had a higher risk of mortality than those living with sexually acquired HIV. Comparing mortality rates by mode of HIV acquisition stratified by age and sex, mortality risk was higher among females aged 20+ years with vertically acquired HIV (IRR: 3.61, 95% CI 1.48–8.82) compared to females with sexually acquired HIV of the same age group. In a sub‐sample analysis, sustained ART adherence was associated with a lower risk of death (aHR: 0.44, 95% CI 0.23–0.85). Conclusions: ALHIV experience higher all‐cause mortality than their HIV‐negative peers, despite having initiated ART. Among ALHIV, mortality risk was higher among males and adolescents who acquired HIV vertically. Strategies to improve survival among ALHIV, including adolescent‐tailored care and support for adherence to ART, are urgently needed
Use of wrist-worn accelerometers to predict Parkinson’s disease
No full textParkinson’s disease (PD) is a growing public health concern with rising diagnosis rates and no preventive treatments. Early-stage monitoring is crucial for understanding disease progression and smartwatches offer great potential in early detection to enable this. This DPhil thesis investigates how wrist-worn accelerometers can be used to help predict incident PD, through the detection of early markers of the disease.
In the initial phase of this work, I reviewed literature on PD and wearable monitoring technologies. This review highlighted research into walking pattern monitoring and proposed risk prediction models using accelerometer data. To conduct this research, a variety of datasets were used, including the UK Biobank Physical Activity Monitoring, PPMI-Verily, Capture-24, OxWalk, and MJFF Levodopa Response studies.
Wrist-worn accelerometer data depended on the development of reliable activity classification models for this thesis’ aims. Existing models were improved to detect labels of activity intensity (sleep, sedentary, light, and moderate-vigorous activity) using a modified ResNet-18 model, pre-trained using self-supervised learning. This was trained and evaluated on the Capture-24 dataset, using both a single train-test split, as seen in prior research, and using 5-fold group cross-validation. Training on the first 100 participants, and testing on the remaining 51 participants, the new ActiNet model outperformed the existing Accelerometer model with a mean per-participant macro F1 score of 0.806 and 0.786, respectively. When trained and evaluated using 5-fold group cross-validation, the ActiNet model achieved a mean macro F1 score of 0.827, outperforming the existing model’s score of 0.773. Walking recognition models were also enhanced to be trained on both healthy and PD datasets. Patterns showed that walking recognition models trained only on healthy participants performed worse in participants with increasing severity of PD. However, training the same models on both healthy and PD data improved general performance, with no drop in performance for higher severities of PD.
An epidemiological analysis was then conducted to investigate the association between device-measured daily step count and PD incidence in the UK Biobank, to assess low daily steps as a risk factor for PD. A 1,000-step increase in median daily steps resulted in a hazard ratio of 0.92 for PD diagnosis, consistent after accounting for various confounders. However, this association attenuated towards null after the removal of the first six years of follow-up, giving an indication of reverse-causation driving this association. Therefore, daily steps appear to be a predictor of but not a risk factor for PD.
Subsequently, machine learning models trained to categorise activity intensity and walking were used to compare active walking windows among participants with different PD statuses. In the PPMI-Verily dataset, a self-supervised ResNet-18 model distinguished between diagnosed PD, prodromal PD, and healthy controls, achieving a mean macro F1 of 0.586 and an area under the receiver operator characteristic curve (AUROC) of 0.856 using the active walking windows. This finding establishes that wrist-worn accelerometers can detect differences in walking behaviour related to PD impairment.
The final phase of research combined a variety of digital measures extracted from wrist-worn accelerometers to build risk prediction models for PD in the UK Biobank. These wrist-worn accelerometer-derived digital measures were compared with traditional risk factors. The risk model with the highest discrimination performance incorporated all risk factors, including demographic, polygenic risk score, overall activity levels, activity composition, and a PD-like gait score, which compared detected active walking windows of participants to those of diagnosed PD participants in the PPMI-Verily dataset. After accounting for shrinkage, this model produced a C-index of 0.902, as compared to 0.778 when using non-accelerometer based traditional demographic risk factors for PD. These results highlight the added benefit of wrist-worn accelerometry, particularly PD-specific digital measures, for improving PD prediction models.
In summary, this thesis leverages wrist-worn accelerometry and machine learning to predict PD, aiming to enhance early diagnosis. The results across the different studies support the promising use of wearable technology for PD
External validation of the LENT and PROMISE prognostic scores for malignant pleural effusion
Get PDFBackgroundAccurate survival estimation in malignant pleural effusion is essential to guide clinical management strategies and inform patient discussion. The LENT and PROMISE scores were developed to aid prognostication in malignant pleural effusion; however their uptake in practice has been limited. We aimed to conduct a detailed external validation of the LENT and PROMISE scores to develop recommendations regarding clinical utility, and to highlight factors limiting performance.MethodsMedical records of patients diagnosed with malignant pleural effusion between 2015-2023 at Oxford University Hospitals were retrospectively reviewed to determine length of survival and the LENT and PROMISE scores at diagnosis. Performance of the scores in predicting overall survival and chance of survival at 3, 6 and 12 months was assessed using measures of discrimination, calibration and overall model performance. Kaplan-Meier analysis and Cox models were utilised to further investigate individual score variables.Results773 patients with malignant pleural effusion were included. Both scores showed predictive ability for overall survival; however median survival estimates lacked precision. Score performance in predicting survival at 3, 6 and 12 months was stronger, with C-indices around 0.8 for both at each time point, and the models appearing well calibrated. Limited stratification of tumour types and lack of consideration of sensitising mutations were demonstrated to be potential factors restricting performance.ConclusionsBoth scores have the ability to prognosticate in malignant pleural effusion, and greater use in practice should be considered. However, areas to improve score performance were also highlighted, and these may aid future model development
Representational schemes for theories with symmetry
Get PDFIn the philosophical literature, symmetries of physical theories are most often interpreted according to the general doctrine called ‘traditional sophistication’ (TS). But even this doctrine leaves two important gaps in our understanding of such theories: (A) it allows the individuation of isomorphism-classes to remain intractable and thus of limited use, which is why practising physicists frequently invoke ‘relational, symmetry-invariant observables’; and (B) it leaves us with no formal framework for expressing interesting counterfactual statements about different physical possibilities. I will call these Limitations of TS. Here I will show that a new Desideratum to be satisfied by theories with symmetries allows us to overcome these Limitations. The new Desideratum is that the theory admits what I will call representational schemes for its isomorphism-classes. Each such scheme gives an equally valid reduced formalism for a theory
The Roman Water Management of Arles as Read in Aqueduct Carbonate Archives
Get PDFThis article reconstructs the complex history of modifications made to the Roman aqueduct of Arelate (Arles), by studying carbonate incrustations in its channels. These deposits, precipitated by flowing water, have preserved an archive of the aqueduct's life‐cycle in their stratigraphy, fabric and stable isotope composition. Two tributaries, from Caparon and Eygalières, converged in a basin before an arcade bridge, from which a single channel continued to Arles. Originally, the Caparon branch alone supplied Arles with water from the south side of the Alpilles hills, the basin acting as a header basin before the arcade. Later, the Eygalières branch from the north side of the Alpilles was joined to the basin. The Caparon branch was then diverted to power water‐mills at Barbegal, changing the basin's function from convergence back to a header basin. After some decades, the Eygalières branch was also used to supply the mills, changing the basin into a distribution structure. From Arles, lead pipes laid across the bed of the Rhône also supplied water to the Trinquetaille quarter. Major cleaning of the aqueduct in the early fourth century is also identified. Anthropogenic carbonates can therefore provide crucial information on the provenance of water and alterations to ancient aqueducts