224092 research outputs found
Sort by
Model monitoring in the absence of labeled data via feature attributions distributions
No full textModel monitoring involves analyzing AI algorithms once they have been deployed and detecting changes in their behaviour.This thesis explores machine learning model monitoring ML before the predictions impact real-world decisions or users. This step is characterized by one particular condition: the absence of labelled data at test time, which makes it challenging, even often impossible, to calculate performance metrics.The thesis is structured around two main themes: \emph{(i) AI alignment}, measuring if AI models behave in a manner consistent with human values and \emph{(ii) performance monitoring}, measuring if the models achieve specific accuracy goals or desires. The thesis uses a common methodology that unifies all its sections. It explores feature attribution distributions for both monitoring dimensions. Using these feature attribution explanations, we can exploit their theoretical properties to derive and establish certain guarantees and insights into model monitoring.For AI Alignment, we explore whether the distributions of feature attributions are distinct for different social groups and propose a new formalization of equal treatment. This novel metric assesses how well AI decisions adhere to ethical standards and political-philosophical values. Our notion of Equal Treatment tests for statistical independence of the explanation distributions over populations with different protected characteristics. We show the theoretical properties of our formalization of equal treatment and devise an equal treatment inspector based on the AUC of a classifier two-sample test. For performance monitoring, we define \emph{explanation shift} as the statistical comparison between how predictions from training data are explained and how predictions on new data are explained. We propose explanation shift as a key indicator to investigate the interaction between distribution shifts and learned models. We introduce an Explanation Shift Detector that operates on the explanation distributions, providing more sensitive and explainable changes in interactions between distribution shifts and learned models. We compare explanation shifts with other methods that are based on distribution shifts, showing that monitoring for explanation shifts results in more sensitive indicators for varying model behavior. We provide theoretical and experimental evidence and demonstrate the effectiveness of our approach on synthetic and real data.Finally, to explain model degradation we use a second model, that predicts the uncertainty estimates of the firstAdditionally, we release two open-source Python packages, \texttt{skshift} and \texttt{explanationspace}, which implement our methods and provide usage tutorials for further reproducibility.<br/
Metabolic dysfunction-associated steatotic liver disease a multisystem disease: assessing the cost effectiveness of pharmacotherapies
No full text<br/
Selection effects for inequalities in early childhood education: addressing an evidence gap in quantitative research
No full textResearch aims: this paper addresses how quantitative research excludes evidence about inequalities in Early Childhood Education (ECE) linked to selection effects (who goes where, gets what, and how much). It presents two empirical examples to illustrate how addressing this gap can enhance research-informed ECE policy-making.Relationship to previous research works: the study extends the concepts of ‘selection effects in education’ (Hall, Palardy & Malmberg, 2024) and ‘airbag moderation’ (Hall et al., 2020) to ECE policy-making.Theoretical and conceptual framework: the study blends developmental and educational theories, including Bronfenbrenner’s ecological systems theory, Vygotsky’s sociocultural theory, Piagetian cognitive theories, and Heckman’s returns on investment. The conceptual framework operationalises these theories via variables like ECE quality, family socioeconomic status and children's cognitive outcomes. Selection effects for inequalities are operationalised via the hypothesis of airbag moderation.Paradigm, methodology and methods: using a post-positivist paradigm, the study employs quantitative longitudinal research designs. Secondary data from two UK studies (EPPE and FCCC) were analysed using the Mplus software.Ethical considerations: using pre-anonymised public data mitigated privacy concerns. The analyses adhered to data-use agreements, with all inputs and outputs available upon request.Main finding or discussion: findings reveal previously unexamined selection effects for educational inequalities. Within the EPPE data, ‘Black Caribbean’ children were more likely to experience inclusive practices in ECE settings with the result of a narrowing in ethnic differences concerning preschoolers’ verbal cognition. Within the FCCC data, higher SES families used ECE for longer hours – a relationship which widened SES gap in preschoolers’ non-verbal cognition. Implications, practice or policy: better understandings of selection effects in education enables better-informed policies that address inequalities, enhancing the ECE’s effectiveness and inclusivity worldwide
Single-cell impedance spectroscopy of bacteria
No full textBacteria are the causative agents of many diseases, and understanding biophysical changes in bacteria following physical or chemical treatment is of considerable interest. In this thesis, different bacteria were measured using single-cell impedance cytometry in order to understand changes in electrical properties following exposure to different perturbations. Traditionally, alternating current (AC) electrokinetic methods are used to measure the electrical properties of cells. However, the throughput of these techniques is low. To overcome this limitation single-cell impedance cytometry measures the impedance of individual cells at hundreds per second.In this study, the electrical properties of single cells were measured across a broad range of frequencies at the rate of tens of thousands of cells in a few minutes. The complex impedance spectrum was modelled using Maxwell’s mixture equation together with the multi-shell model to extract dielectric parameters for each bacterium.Two model organisms were used for experiments: Escherichia coli and Staphylococcus aureus. Impedance cytometry was used to characterise the effect of heat treatment (including pasteurisation and autoclaving) on bacteria and results showed that the dielectric properties of the cells were in agreement with AC electrokinetic methods. Bacteria were also exposed to three classes of antibiotics, namely β-lactam, polymyxin and aminoglycoside. Dielectric characterisation demonstrated that bacteria respond differently following exposure to different antibiotics. Their phenotypic response changes their dielectric parameters in a way that is consistent with the mode of action of antibiotics.Single-cell impedance cytometry was also used to follow bacterial-phage interaction. Experiments showed that the phenotypic response of bacteria can be followed during the phage infection cycle, and that the technique can rapidly identify bacterial susceptibility to phages. The data also helps explain the complex bacterial defence mechanisms against phages from a dielectric perspective.In summary, this thesis describes an electrical method for probing bacteria, linking measured biophysical changes to biological responses, furthering our understanding of the action of antibiotics and phage therapies
Towards quantum SAGINs harnessing optical RISs: applications, advances, and the road ahead
No full textThe space-air-ground integrated network (SAGIN) concept is vital for the development of seamless next-generation (NG) wireless coverage, integrating satellites, unmanned aerial vehicles, and manned aircraft along with the terrestrial infrastructure to provide resilient ubiquitous communications. By incorporating quantum communications using optical wireless signals, SAGIN is expected to support a synergistic global quantum Internet alongside classical networks. However, long distance optical beam propagation requires line-of-sight (LoS) connections in the face of beam broadening and LoS blockages. To overcome blockages among SAGIN nodes, we propose deploying optical reconfigurable intelligent surfaces (ORISs) on building rooftops. They can also adaptively control optical beam diameters for reducing losses. This article first introduces the applications of ORISs in SAGINs, then examines their advances in quantum communications for typical SAGIN scenarios. Finally, the road ahead towards the practical realization of ORIS-aided NG quantum SAGINs is outlined
How do foreign and domestic institutional investors drive the market value? The influence of family ownership
No full textPurpose: this paper investigates and compares the value impacts of foreign and domestic institutional investors on the market value of family and non-family companies. Subsequently, examines how foreign and domestic institutional investors and their value impacts get influenced by different degrees of family ownership. Design/methodology/approach: the sample of this study includes 339 non-financial firms from NIFTY-500 for 11 years from 2011 to 2020, which contains 128 family and 211 non-family companies. Both static (Fixed-effect model) and dynamic (two-step system GMM) models are employed to test the hypotheses.Findings: Findings suggest that foreign institutional investors outshine domestic institutions in terms of value creation. Meanwhile, higher (>50%) family holdings are detrimental to the foreign institutional investors, while moderate holdings (26%-49%) improve domestic institutional investments. The favorable effect of foreign players gets diluted with the higher (>50%) family holdings, while the adverse effect of domestic players improves with the moderate (26%-49%) family holdings. Overall, partial family control is beneficial, while low and absolute family control is detrimental to market value. These findings clearly indicate that institutional investors are family control-dependent, where family control effect is not static.Originality/value: this paper offers a novel perspective by addressing the effect of costs and benefits realized at three distinctive level of family holdings on foreign and domestic institutional investors, and their value impacts to witness differences caused by varying family control, which is not done earlier as per the best of our knowledge.<br/
A prospective observational cohort study comparing high-complexity against conventional pelvic exenteration surgery
No full textBackground: conventional pelvic exenteration (PE) comprises the removal of all or most central pelvic organs and is established in clinical practise. Previously, tumours involving bone or lateral sidewall structures were deemed inoperable due to associated morbidity, mortality, and poor oncological outcomes. Recently however high-complexity PE is increasingly described and is defined as encompassing conventional PE with the additional resection of bone or pelvic sidewall structures. This observational cohort study aimed to assess surgical outcomes, health-related quality of life (HrQoL), decision regret, and costs of high-complexity PE for more advanced tumours not treatable with conventional PE. Methods: high-complexity PE data were retrieved from a prospectively maintained quaternary database. The primary outcome was overall survival. Secondary outcomes were perioperative mortality, disease control, major morbidity, HrQoL, and health resource use. For cost-utility analysis, a no-PE group was extrapolated from the literature. Results: in total, 319 cases were included, with 64 conventional and 255 high-complexity PE, and the overall survival was equivalent, with medians of 10.5 and 9.8 years (p = 0.52), respectively. Local control (p = 0.30); 90-day mortality (0.0% vs. 1.2%, p = 1.00); R0-resection rate (87% vs. 83%, p = 0.08); 12-month HrQoL (p = 0.51); and decision regret (p = 0.90) were comparable. High-complexity PE significantly increased overall major morbidity (16% vs. 31%, p = 0.02); and perioperative costs (GBP 37,271 vs. GBP 45,733, p < 0.001). When modelled against no surgery, both groups appeared cost-effective with incremental cost-effectiveness ratios of GBP 2446 and GBP 5061. Conclusions: high-complexity PE is safe and feasible, offering comparable survival outcomes and HrQoL to conventional PE, but with greater morbidity and resource use. Despite this, it appears cost-effective when compared to no surgery and palliation.</p
Five AIMS: lessons from internet governance for artificial intelligence management strategies
No full textTo help scaffold coherent, coordinated, and enforceable rules and institutions, Dame Wendy Hall, DBE, FRS, FREng, Regius Professor of Computer Science, Associate Vice President (International Engagement) and Director of the Web Science Institute at the University of Southampton, Kieron O’Hara, emeritus fellow in Electronics and Computer Science at the University of Southampton, and Pierre Noro, advisor of the PSIA Tech & Global Affairs Innovation Hub, reinterpret the Four Internet models elaborated by Hall and O'Hara in their 2018 article and in their influential book Four Internets: Data, Geopolitics, and the Governance of Cyberspace (Oxford UP, 2021) in regard to AI technologies
Sigma-2 receptor modulator CT1812 alters key pathways and rescues retinal pigment epithelium (RPE) functional deficits associated with dry age-related macular degeneration (AMD)
No full textTrafficking defects in retinal pigmented epithelial (RPE) cells contribute to RPE atrophy, a hallmark of geographic atrophy (GA) in dry age-related macular degeneration (AMD). Dry AMD pathogenesis is multifactorial, including amyloid-β (Aβ) accumulation and oxidative stress-common features of Alzheimer's disease (AD). The Sigma-2 receptor (S2R) regulates lipid and protein trafficking, and S2R modulators reverse trafficking deficits in neurodegeneration in vitro models. Given overlapping mechanisms contributing to AD and AMD, S2R modulator effects on RPE function were investigated. The S2R modulator CT1812 is in clinical trials for AD, dementia with Lewy bodies, and GA. Leveraging AD trials testing CT1812, unbiased analyses of patient biofluid proteomes revealed that proteins altered by CT1812 associated with GA and macular degeneration disease ontologies and overlapped with proteins altered in dry AMD. Differential expression analysis of RPE transcripts from APP-Swedish/London mutant transgenic mice, a model featuring Aβ accumulation, revealed reversal of autophagy/trafficking transcripts in S2R modulator-treated animals versus vehicle toward healthy control levels. Photoreceptor outer segment (POS) trafficking in human RPE cells showed deficits in response to Aβ
1-42 or hydrogen peroxide compared to vehicle. S2R modulators normalized stressor-induced POS trafficking deficits, resembling healthy control. Taken together, S2R modulation may provide a novel therapeutic strategy for dry AMD.
</p
Data sources:
Access Repository Dashboard
Do you manage e-Prints Soton? Access insider analytics, issue reports and manage access to outputs from your repository in the CORE Repository Dashboard!