324139 research outputs found
Sort by
On learning, fairness, and complexity
Get PDFIn this thesis we study the learning and complexity-theoretic underpinnings of the multigroup fairness framework for prediction algorithms. Multiaccuracy and multicalibration are two primary multigroup fairness notions, which ensure accurate and calibrated predictions, respectively, for every subpopulation that can be identified within a specified class of computations [HKRR18]. They both can be achieved from a single learning primitive: weak agnostic learning. A line of work starting from [GKR+22] has shown that multicalibration implies a very strong indistinguishability-based form of learning called omniprediction. The multigroup fairness framework is also deeply connected to complexity theory through the Regularity Lemma and its various implications [CDV24].We provide a thorough study of the connections between multigroup fairness notions, the central learning primitive of weak agnostic learning, and the fundamental Hardcore Lemma in complexity theory. We find that multiaccuracy in itself is rather weak, but that the addition of global calibration (this notion is called calibrated multiaccuracy) boosts its power substantially, enough to recover implications that were previously known only assuming the stronger notion of multicalibration.We give evidence that multiaccuracy might not be as powerful as standard weak agnostic learning, by showing that there is no way to post-process a multiaccurate predictor to get a weak learner, even assuming the best hypothesis has correlation 1/2. However, by also requiring the predictor to be calibrated, we recover not just weak, but strong agnostic learning. A similar picture emerges when we consider the derivation of hardcore measures from predictors satisfying multigroup fairness notions [TTV09; CDV24]. On the one hand, while multiaccuracy only yields hardcore measures of density half the optimal, we show that (a weighted version of) calibrated multiaccuracy achieves optimal density.Our results yield new insights into the complementary roles played by multiaccuracy and calibration in each setting. They shed light on why multiaccuracy and global calibration, although not particularly powerful by themselves, together yield considerably stronger notions.We further study the connections between the multigroup fairness framework and the problem of learning selective classifiers, which are predictors that are allowed to abstain on some fraction of the domain. Building on the notion of omniprediction (which is in turn built using tools from the multigroup fairness framework), given a pre-specified class of loss functions, we provide an algorithm for efficiently building a single classifier that learns abstentions and predictions optimally for every loss in the entire class, where the abstentions are decided efficiently for each specific loss function by applying a fixed post-processing function. We call this classifier a selective omnipredictor. Our algorithm and theoretical guarantees generalize the previously-known algorithms for learning selective classifiers in formal learning-theoretic models [KKM12].We then extend the traditional multigroup fairness algorithms to the selective classification setting and show that we can use a calibrated and multiaccurate predictor to efficiently build selective classifiers that abstain optimally not only globally but also locally within each of the groups in any pre-specified collection of possibly intersecting subgroups of the domain, and are also accurate when they do not abstain. This provides yet another use case of the notion of calibrated multiaccuracy. Moreover, we show how our abstention algorithms can be used as conformal prediction methods in the binary classification setting to achieve both marginal and group-conditional coverage guarantees for an intersecting collection of groups. We provide empirical evaluations for all of our theoretical results, demonstrating the practicality of our learning algorithms for the goal of abstaining optimally and fairly
Ellen Crawford - Thesis submitted in partial fulfilment of the degree of Doctor of Clinical Psychology (DClinPsych): Main project title: Interpretation bias in maternal postnatal anxiety: perinatal content-specificity of ambiguous stimuli
Get PDFSystematic Review of the Literature Background: In the three decades since the World Health Organisation advised member nations to introduce Universal Newborn Hearing Screening Programmes, early identification and intervention programmes are now commonplace. However, diagnoses of hearing loss in infants still require significant adjustment for families, as over 90% of children with hearing loss are born to hearing parents.Objective: To review research findings of parental psychological wellbeing and coping resources in the early years of their deaf/hard of hearing child’s (cDHH) life.Method: A systematic search of five databases (PsychINFO, Medline, CINAHL, Embase, and Central) was conducted to identify studies investigating psychosocial factors in parents of young cDHH (from infancy up to age six). Following screening, data on parents’ experiences and variables impacting parent outcomes were synthesised.Results and Conclusions: Eighteen studies met inclusion criteria. Findings were organised, based on variables assessed, into three main categories: 1) parenting stress, 2) parental mental health and wellbeing, and 3) parental adjustment and coping resources. Families of infants diagnosed with hearing loss do not universally experience poorer wellbeing, though some variability by populations and outcomes are observed. The limited evidence of protective coping resources (such as social support and parent’s appraisals of their lives) and access to early intervention programmes may explain some of the variability in findings but more research utilising different study designs are necessary. Implications are discussed.Service Improvement ProjectPurpose: In Berkshire Talking Therapies (BTT), perinatal clients form a small minority of overall clients, and male perinatal clients an even smaller proportion. However, those who engage tend to have favourable recovery rates. This study therefore aimed to investigate barriers and facilitators to accessing and engaging with BTT support for perinatal clients, including male perinatal clients.Methods: A mixed-methods approach was used across two phases to: 1) quantitatively assess patterns of access and engagement, and 2) explore perceptions of factors that fostered or hindered these. Phase 2 was completed in two parts. In part one all perinatal clients were invited to complete a survey (n=17) and part two used interviews to explore the specific experiences of male perinatal clients (n=5).Outcomes: Perinatal clients were more likely than not to complete treatment. There were statistically significant gender differences in treatment completion across stepped care interventions. Referral source and final referral destination did not alter patterns of engagement. Qualitative data indicated the importance of early engagement. Survey and interview themes highlighted increased flexibility and attuned therapy delivery as significant facilitators, with lack of awareness, stigma and burdensome processes key barriers.Implications: Increasing these facilitators and addressing barriers could improve numbers of perinatal clients accessing and successfully engaging in BTT support. NHS Talking Therapies services, such as BTT, are key primary care providers for perinatal clients, particularly male perinatal clients who have been historically excluded and overlooked in healthcare service.Theory Driven Research ProjectBackground: Postnatal anxiety is under-studied in comparison to postnatal depression, yet evidence suggests it can influence caregiving behaviours, such as responding less sensitively to infant cues. Negatively biased interpretations of ambiguous information could represent a cognitive processing mechanism between anxiety and parenting behaviours.Aims: To explore maternal interpretation biases in those with high or low postnatal anxiety and the content-specificity of potential biases by comparing general and caregiving-related ambiguous scenarios. To additionally examine the relationships of postnatal depression symptoms and perinatal anxiety subtypes within ambiguous interpretations.Methods: This online cross-sectional study compared 53 mothers with high anxiety and 62 with low anxiety (N=115). Measures included screening scales for perinatal anxiety (PASS) and depression (EPDS), and the Recognition Test paradigm for interpretative biases of ambiguous scenarios.Results and Conclusions: High maternal postnatal anxiety was associated with more negative interpretation biases, but this did not significantly differ by ambiguous scenario type, i.e., general (less personally-salient) or caregiving-specific (more personally-salient) ambiguous scenarios. Postnatal depression symptoms significantly contributed to the variance explained (7%) in interpretation bias indices but there was no difference by subscale of perinatal anxiety. Addressing general interpretation biases in perinatally anxious mothers may present an effective intervention target, however, further research into content-specificity and the profiles of perinatal processing biases is still needed
Developing remote technologies to understand changes in seabird behaviour and ecology
Get PDFSeabirds are one of the most threatened groups of birds. Effective monitoring is therefore needed to understand population trends and the drivers of these trends, if we want to implement successful conservation action. Current monitoring methods are typically limited in their frequency and geographic reach, as many seabirds breed in remote and hard to access locations, which can make comprehensive fieldwork programs prohibitively expensive and time-consuming. Advances in remote technologies are providing opportunities to expand the scale of seabird monitoring at reduced cost and with less disturbance. In this thesis, I aim to develop methods to improve the scope of seabird monitoring and research, and apply these methods to improve our understanding of the impact of anthropogenic climate change on seabird phenology and breeding success. First, I consider the rapidly increasing application of drones in seabird research. I collate information from over 100 studies to develop an eight step framework for ensuring drone-seabird surveys are safe, effective, and within the law. Second, I use a time-lapse camera network, which has collected over 200 000 images between 2014 and 2023, to examine changes in Black-legged Kittiwake Rissa tridactyla phenology along a latitudinal gradient, as well as the impact of direct weather on kittiwake chick survival. Images were annotated by over 35 000 citizen scientists on the Seabird Watch citizen science project (hosted on the Zooniverse platform). I examine the cost-effectiveness of the citizen science camera network and look at patterns of volunteer engagement to understand the long-term success of the project. Having developed an algorithm to use the citizen science data, I find that while precipitation has a large and adverse effect on kittiwake chick survival (a 3.7-fold increase in daily chick failure rate per 1 mm increase in precipitation), other weather variables have no measurable impact. This suggests that the indirect effects of ocean warming may be having a greater impact on kittiwake breeding success than direct weather effects at present. Finally, I use a large geolocator tracking dataset, containing over 1400 individuals from 34 breeding colonies tracked between 2009 and 2022, to examine the mechanisms underpinning population level change in kittiwake phenology, in response to rising sea surface temperatures. I find that while some aspects of kittiwake migratory phenology are changing plastically, changes in breeding phenology were primarily due to turnover of individuals. I highlight that the direction, magnitude, and mechanism by which phenological change is mediated is context- and trait dependent, and this has important implications for understanding and predicting how seabirds and migratory animals more generally- can respond to ongoing climate change. Overall, I discuss the possibilities and challenges of incorporating remote technologies into seabird monitoring programs and their importance in the face of anthropogenic change
Roles of the CTDNEP1-NEP1R1 phosphatase complex at the nuclear envelope
No full textThe nuclear envelope (NE) is a dynamic structure that is continuously remodelled to meet different cellular demands. Changes in NE dynamics and mutations in NE proteins have been implicated in pathological conditions. CTDNEP1-NEP1R1 is a conserved NE phosphatase complex, with a previously identified role in regulating NE biogenesis by dephosphorylating lipin, a phosphatidic acid hydrolase critical for lipid homeostasis. Recent work in the lab showed that CTDNEP1-NEP1R1 also regulates the degradation of an inner nuclear membrane protein SUN2, a subunit of the LINC complex involved in mechanotransduction across the NE. This finding prompted us to search for novel functions of CTDNEP1-NEP1R1 in the NE through proteomics studies. Co-immunoprecipitation of CTDNEP1 and NEP1R1 identified MAN1 as an inner nuclear membrane interactor of the phosphatase complex, and together play a role in the TGF-β/BMP signalling pathway. Domain mapping, mutagenesis and functional assays provided mechanistic basis of the CTDNEP1-NEP1R1-MAN1 complex for TGF-β/BMP signalling inactivation. In addtion, phosphoproteomics further nominates additional substrates of CTDNEP1-NEP1R1 at the NE and ER, indicating potentially broader functions of CTDNEP1-NEP1R1 beyond lipid biogenesis
Patients', clinicians' and research's priorities on important outcomes in multiple myeloma: A mixed‐methods study
Get PDFSummary: Research in multiple myeloma increasingly relies on surrogate end‐points to expedite approvals, yet these may not reflect patient priorities. We conducted a mixed‐methods study to identify outcomes valued by patients and clinicians and compare them with end‐points used in randomized controlled trials (RCTs). Interviews with 10 patients and 6 clinicians identified treatment priorities, which, together with end‐points from a systematic review of myeloma RCTs, informed tailored surveys, completed by 117 patients and 105 clinicians. Both groups ranked quality of life (QoL) as most important (odds ratio [OR] 1.01; 95% confidence interval [CI] 0.55–1.87). Clinicians more often prioritized overall survival (OS) (OR 1.92; 95% CI 1.05–3.50) and progression‐free survival (PFS) (OR 5.37; 95% CI 1.95–14.79), whereas patients prioritized pain reduction (OR 0.03; 95% CI 0.00–0.23). Compared with RCT end‐points, patients emphasized QoL (OR 0.03; 95% CI 0.01–0.07) and pain elimination (OR 0.05; 95% CI 0.02–0.11), while trials favoured PFS (OR 6.33; 95% CI 2.53–15.83) and response (OR 17.75; 95% CI 5.56–56.61). Clinicians aligned with trials on PFS but valued QoL and OS more highly. QoL emerged as a shared priority, underscoring the need for patient‐centred trial designs that better capture outcomes meaningful to those living with myeloma
A pan-cancer compendium of 1,294 plasma cell-free DNA methylomes and fragmentomes enabling multicancer detection
Get PDFCell-free DNA analysis via methylation and fragmentation profiling has advanced minimally invasive cancer detection; however, broader application has been limited by small cohorts and inconsistent data processing. Here we collated 1,074 cfMeDIP-seq profiles across 9 studies, comprising cancer samples from 11 cancer types, carriers of Li-Fraumeni syndrome and healthy controls. We developed a uniform computational workflow to mitigate technical and biological confounders across cohorts. This analysis identified 14,202 pancancer differentially methylated regions for cancer detection, along with cancer-specific markers for subtype monitoring. Fragmentomic profiling revealed distinguishing differences in 5′ end motifs, fragment lengths and nucleosome footprints across cancers. Integrating methylome and fragmentome features enhanced cancer detection and classification. Validation in 220 independent samples, including 3 cancer types absent from the primary dataset, confirmed the robustness of our findings. Altogether, this work provides a pancancer cell-free DNA resource of 1,294 samples to support future methylome and fragmentome studies
Mechanical de-skewing enables high-resolution imaging of thin tissue slices with a mesoSPIM light-sheet microscope
Get PDFOptical clearing combined with light-sheet microscopy enables high-resolution imaging of extended tissue at scale. However, standard mesoSPIM systems are optimised for intact organs and are not suited to thin tissue slices. We present an oblique compensation scanning method using obliquely mounted samples held between refractive-index-matched slides in a 3D-printed frame. This enables mechanical de-skewing during acquisition, minimising post-processing requirements. We demonstrate feasibility in fluorescent bead phantoms and rabbit heart tissue, achieving a 4.8 × reduction in processing time and a 1.5 × improvement in axial resolution ((13.15±1.36) μm to (8.72±1.80) μm) compared to conventional z scan. The oblique compensation acquisition method extends mesoSPIM's utility to fragile, laterally extended tissue sections
A Novel Network-Level Fused Self-Attention Deep Neural Network for Cervical Cancer Classification from Cervicography Images
Get PDFIntroduction: cervical cancer ranks as the fourth most common cancer among females worldwide. Approximately 528,000 new cases of cervical cancer are reported annually, and about 85% of them occur in less-developed countries. The lack of skilled medical staff and pre-screening procedures is the main cause of the high fatality rate in these countries. Cervicography images are the gold standard procedure for the evaluation of cervical cancer; however, the high intra-class inconsistency makes the diagnosis process more challenging for skilled medical specialists. Method: In this work, we propose a fully automated computer-aided diagnosis (CAD) system for classifying cervical cancer using Cervicography images. Data augmentation is performed in the initial phase to address dataset imbalance. Subsequently, we proposed two novel deep learning modules: the 11-Parallel Inverted Residual Bottleneck Blocks (11-PIRBnet) architecture and the 9-Parallel Inverted Residual blocks with Self-Attention Mechanism (9-PIRSANet). Both modules are fused at the network level via a depth concatenation layer to form a new network, 375NFNet. The proposed network is trained on the selected dataset, whereas the hyperparameters are initialized through Bayesian Optimization (BO). For feature extraction, a depth concatenation layer is used during testing to combine information from both deep learning modules. Finally, the extracted features are classified using a shallow neural network (SNN) to produce the final classification. Result: To evaluate the model, experiments were conducted on a publicly available cervical screening dataset of Cervicography images, and results demonstrate an accuracy of 95.5%, a precision of 95.4%, and an area under the curve of 0.97. When compared with several pre-trained techniques, the proposed architecture achieved significant improvement in accuracy, precision, and number of trainable parameters. Conclusion: The proposed 375NFNet architecture demonstrates remarkable accuracy and efficiency in classifying cervical cancer through cervicography images, which shows its potential as a valuable tool in resource-constrained environments
Genomic characterization of Sabiá virus in Brazil, 2019–2020: Implications for diagnostics, virus evolution, and receptor binding
Get PDFBetween December 2019 and January 2020, two patients suspected of having severe yellow fever were admitted to a tertiary healthcare facility in São Paulo, Brazil, presenting with acute hemorrhagic syndrome and neurological alterations; both cases had fatal outcomes. Upon admission, both tested negative for yellow fever viral RNA, and Sabiá virus (SABV), a New World arenavirus, was identified as the causative pathogen. To date, only four humans naturally acquired SABV infections have been confirmed, all fatal and linked to rural settings. We applied next-generation sequencing to generate complete and near-complete genomes from two patients (SP17 and SP19). Existing molecular diagnostics failed to detect SABV; therefore, new molecular tests were developed. Genetic analyses of SP17 and SP19 genomes along with other arenaviruses, revealed that the new cases were genetically diverse, showing 93-98.2% amino acid identity at the NP level among SP17, SP19, and the 1990 reference strain (SPH114202). Time-scaled phylogenetic analyses confirmed that SP17 and SP19 were not epidemiologically linked and suggested that SABV has been circulating undetected in Brazil for over a century. Additionally, homology modeling and structure-based mapping provided insights into SABV receptor-binding sequence conservation, suggesting that SABV shares similar receptor binding structure to other clade B arenaviruses, despite some amino acid variation around receptor binding site. Our findings underscore the need for retrospective and prospective surveillance of undiagnosed hemorrhagic fever cases to assess the public health impact of SABV in Brazil
Tunable dynamic speckle generation for random illumination microscopy
Get PDFSpeckled illumination enhances widefield fluorescence microscopy by enabling optical sectioning and super resolution. In random illumination microscopy, sequences of speckled illumination patterns are used to excite fluorescent samples and images are reconstructed based on a statistical analysis of the intensity fluctuations. Although random illumination microscopy has been shown to give excellent performance, its widespread implementation is hindered by the high cost and complexity of the generation of suitable speckled illumination patterns, which is typically achieved using digital micro-mirror devices or spatial light modulators. Here, we present a zwitterion-doped liquid crystal device capable of generating independent, high-contrast speckle patterns with a tunable decorrelation time in the 0.1 s – 0.1 ms range under visible laser illumination. This liquid crystal based dynamic speckle generator is applied to widefield random illumination fluorescence microscopy of tissue and cell samples, where it enables optical sectioning with a 2μm axial resolution, and a 1.5 - fold improvement in lateral spatial resolution. Owing to its low cost and simplicity, this liquid crystal speckle generator offers an attractive alternative to digital micro-mirror and spatial light modulator devices for implementing widefield random illumination microscopy