324139 research outputs found
Sort by
The contribution of universities to regional innovation in the Guangdong-Hong Kong-Macao Greater Bay Area (GBA)
Get PDFThis thesis explores the contribution of universities to regional innovation by conducting a nested case study of a comprehensive Double First-Class university in the Guangdong-Hong Kong-Macao-Greater Bay Area (GBA) of China, a critical site envisioned to strengthen regional synergy and establish globally influential innovation systems. An single nested case study was conducted with integrating analyses of national and local policies on science, technology, innovation and education, institutional documents, and 88 interviews with stakeholders from the university, industry and government. With Sun-Yet-Sen University (SYSU) and its collaboration partnerships within the GBA as the case, the research sheds light on the important features as the institution of innovation, and critical nodes of multisectoral innovation networks within local innovation systems. Moving beyond simplified conceptualisations of university–industry–government–society relations rooted predominantly in Western contexts, this research advances a context-sensitive and dynamic approach to understanding innovation, the configurations of Triple/Quadruple Helix systems and universities’ roles therein. Unlike a static statist model, roles within Triple Helix are dynamic, varying across innovation sub-systems and the networks of organisations, projects and individuals, influenced by factors such as geopolitical environments, central party-state apparatus, local innovation system configurations, university structures and strategies, disciplinary focuses, and individual enterprising attributes. State entrepreneurialism, through the interplay of top-down steering and bottom-up initiatives, remains a defining feature of China’s and the GBA that drives multi-scalar innovation capacity-building. China’s innovation strategy concentrates on strengthening endogenous innovation capability and global innovation leadership through productive cross-scale intersections of innovation, national system thinking and synergistic instruments for deployment and implementation. Global connectivity and international collaboration are conditioned by overarching national priorities and interests, an ongoing balance between openness and closure. By examining the GBA’s strategic role as a local-global nexus, the study highlights that innovation network-building, collective learning, and diffusion of innovation practices, though originating in a specific location and are largely policy-driven, often transcend local, regional, and national boundaries. The GBA innovation system has evolved into a dynamic, multi-actor configuration marked by heterogeneity and emergent balance. As integration deepens across Hong Kong, Macau, and Guangdong, institutional innovation is increasingly characterised by two-way empowerment: top-down strategic design from the central government and bottom-up feedback from cross-border innovation practices. The integration of HK and Macau’s international regulatory experience with Guangdong’s experimental policy capacity is fostering a hybrid governance paradigm that is both internationally oriented and distinctively Chinese. Universities’ enterprising efforts represent a confluence of policy-driven, market-oriented, and higher education-specific rationales. Embedded in national and local policy agendas, universities align pursuit of academic excellence and institutional prestige with governmental STI objectives and broader societal expectations. They organise innovation practices through large-scale infrastructures, mega-projects, mission-driven team-building, interdisciplinary research, and the coordinated integration of innovation, industrial, capital, and talent chains. Through multi-campus configurations and expanded multidisciplinary profiles across Guangzhou, Zhuhai, and Shenzhen, universities play increasingly enterprising roles in regional innovation governance. These expansions align with internal logics of enrollment growth, prestige accumulation, and resource acquisition. SYSU exemplifies this dynamic, building an enterprising multiversity model with a strong academic core while strategically leveraging the GBA’s political momentum and geographical advantages. Its co-developed innovation platforms with local governments and industries have enabled sustainable spatial expansion, diversified funding, and fostered hybrid organisational forms, enhancing institutional capacity and responsiveness to external demands. The broadening “developmental periphery” (Clark, 2001)—through innovation parks, joint research centres, and incubator programs—enable universities to broaden student markets, research funding, and public and private investment, reinforcing their dual identities as educational and entrepreneurial actors. Some even embarked on experiment of institutional innovation. The emergence of “neo-type research universities” such as HKUST(GZ) and CUHK(SZ), exemplifies how institutions simultaneously position themselves within global academic hierarchies while embedding more deeply in the GBA’s innovation system. These expansions of multiversities is also underpinned by the capital logic of urban economic expansion, in which they are positioned and position themselves as regional growth machines, attracting high-skilled talent, foreign investment, and emerging industries. The integration of university-led innovation ecosystems into urban planning strategies reflects a broader entrepreneurial turn in higher education, whereby universities become instruments of land valorisation, technology commercialisation, regional economic upgrading, and navigation of demand-supply equilibrium amid changing geopolitical innovationscape
Stacks in derived bornological geometry
Get PDFIn this thesis, we describe a higher categorical framework for discussing derived analytic and derived smooth geometry. Analogous to the Toën and Vezzosi model of derived algebraic geometry which uses simplicial commutative rings as its building blocks, in our model we use simplicial commutative complete bornological rings. This work builds upon foundational work of Ben-Bassat, Kelly, and Kremnizer. This general framework allows us to prove several results about derived stacks, in particular we develop the obstruction theory of stacks and use it to prove a representability theorem. This theorem cements this new theory of derived bornological geometry as strong and versatile, and gives differential and analytic geometers a new perspective on their own representability problems. In this thesis, we begin by studying a generalisation of the Koszul duality theory of Beilinson, Ginzburg, and Soergel to the setting of algebra objects in a bicomplete closed symmetric monoidal exact category E with enough flat projectives. Examples include the category CBorn_R of complete bornological spaces and the derived equivalent category Ind(Ban_R) of formal filtered colimits of Banach modules over a Banach ring R. We then define a general categorical context we call a derived geometry context. In these contexts we obtain our representability theorem. If a derived stack has a geometric truncation, is compatible with Postnikov towers, and has a well defined obstruction theory, our theorem shows that it is representable by a derived geometric stack. Working relative to CBorn_R for an appropriately chosen Banach ring R, we can define suitable derived geometry contexts modelling derived complex analytic and derived smooth geometry. In the derived smooth geometry setting, we develop a theory of C∞-bornological rings extending the theory of C∞-rings. Finally, we show that the derived moduli stack of non-linear elliptic PDEs is representable by a derived C∞-bornological affine scheme
The efficient doctor-patient relationship: a value-based framework for person-centred healthcare in the age of artificial intelligence
No full textWe know that the doctor-patient relationship contributes to improving patient outcomes. Policymakers and professional bodies expect the integration of artificial intelligence (AI) in healthcare to enhance doctor-patient relationships in an efficient way. Promissory discourse has largely focused on the efficiency gains from AI, often assuming that these gains will straightforwardly be reallocated to benefit doctor-patient relationships, yet offering little explanation of how this reallocation would occur in practice. In contrast, the bioethics literature has focused mainly on concerns specific to AI systems, such as explainability and its effects on the doctor-patient relationship, often treating these issues in isolation and neglecting to consider them within the broader context of resource allocation decisions. This leaves a gap in understanding how the doctor-patient relationship can be integrated into a healthcare context that strives for AI-supported efficiency. To address this gap, I use a combination of theoretical and empirical work. I start by theoretically exploring the relationship between the doctor-patient relationship and efficiency. I develop the theoretical framework of the efficient doctor-patient relationship, which provides an understanding of the link between efficiency and the doctor-patient relationship in the healthcare context. I argue that, rather than understanding the doctor-patient relationship and efficiency as conflicting or complementary concepts, the doctor-patient relationship should be incorporated into our understanding of efficiency. In practice, this means that resources should be allocated to the doctor-patient relationship when the latter can contribute to maximising outcomes that matter to patients. I then gather qualitative data through semi-structured interviews to explore how clinicians and AI developers understand AI to impact healthcare practices. In the subsequent discussion, I bring together the theoretical and empirical work to gain an empirically grounded understanding of the role of AI within the efficient doctor-patient relationship. Two findings emerge from my analysis. First, AI can enhance the efficiency of the doctor-patient relationship through the saving and reallocation of resources. But, importantly, how to reallocate resources should depend on the outcomes that matter to various patient clusters. Second, AI tools can affect the efficiency of the doctor-patient relationship itself, by changing the dynamics of the interaction. For example, the design of a tool, such as whether it is a black box or explainable, can alter the efficiency of the relationship. These findings help shape policy. To integrate the doctor-patient relationship into a healthcare context that strives for AI-supported efficiency, resources should be allocated to the relationship when the latter can contribute to maximising outcomes that matter to patients. Decision-makers must first assess whether the AI system in question saves resources that could be reallocated. If so, reallocation should be driven by the needs of specific patient clusters. Furthermore, policymakers need to consider AI tools’ direct effect on the doctor-patient relationship and evaluate these together with other effects beyond the doctor-patient relationship to maximise patient outcomes with available resources
“ It’s like a car that doesn’t like gasoline ” - a qualitative study of siblings’ understanding of anorexia nervosa in childhood: perspectives from siblings and parents
Get PDFBackground: When adolescents develop anorexia nervosa (AN), this impacts the family system and puts healthy siblings at risk of mental health problems. Siblings need support and age-appropriate information about the diagnosis to prevent negative mental health outcomes. However, evidence-based support for siblings is limited. The current study aimed to explore siblings’ perceptions of AN and parents’ beliefs about siblings’ understanding expressed within an intervention programme for siblings and parents. Methods: This qualitative study employed a hybrid approach, integrating deductive thematic analysis using the common sense model of self-regulation as a coding framework with inductive thematic analysis. The data materials comprised (1) interviews conducted by clinicians with siblings about the AN diagnosis, (2) siblings’ understanding of AN as expressed in sibling groups, (3) parents’ beliefs about what siblings understand expressed in parent groups, and (4) parent-sibling conversations about AN. Video and audio recordings of the data were transcribed and analysed. The sample comprised nine siblings of European descent, aged 8 to 15 years, and their parents. All siblings had a sister with clinically confirmed AN. Results: The siblings had limited knowledge and expressed uncertainties about AN across the five themes identity (label and symptoms), causes, consequences, treatment, and timeline. In the inductive analysis, two additional themes were identified. The first, Parental perspectives on siblings’ understanding, had two sub-themes: AN as a confusing and complex disorder, and Discrepancy between siblings’ understanding of AN and parents’ beliefs about their understanding. The second theme was Barriers to communication about the diagnosis. Conclusions: The results extend knowledge about informational support needs in siblings of adolescents with AN. Insights into what siblings and parents of adolescents with AN share about the diagnosis in different contexts can be used to guide the adaptation of interventions and policies. Trial registration: ClinicalTrials.gov NCT04056884
Probing the Higgs boson CP properties in vector-boson fusion production in the H → τ + τ − channel with the ATLAS detector
Get PDFThe CP properties of the Higgs boson are studied in the vector-boson fusion production mode. The analysis exploits the decay mode of the Higgs boson into two τ-leptons using 140 fb−1 of proton-proton collision data at s=13 TeV collected by the ATLAS experiment at the Large Hadron Collider. Results are obtained using the Optimal Observable method. CP-violating interactions between the Higgs boson and electroweak gauge bosons are considered in the effective field theory framework, with the interaction strength described in the HISZ basis by d~, and in the Warsaw basis by cHW~, cHB~, and cHW~B. No deviations relative to the Standard Model are observed, and limits are obtained on the strength parameters. The d~ parameter is constrained to the interval [−0.012, 0.044] at the 95% confidence level while cHW~ is constrained to [−0.24, 0.83], when considering both linear and quadratic effects of physics beyond the Standard Model
Data quality in causal machine learning with applications to algorithmic fairness
Get PDFThe success of modern machine learning methods can be attributed to three main factors: i) The availability of increasing amounts of high quality data, ii) the sustained growth in computational resources, iii) the invention of algorithms that can reap the gains of both of these simultaneously, being specifically tailored to consume ever larger datasets on cutting-edge hardware. In this thesis, we focus on the first question of data quality in the subfield of ML that intersects with another field, causality. Causality aims to produce a precise mathematical formulation for the age-old question of cause and effect. In doing so, it provides a framework to formally reason about interventions, counterfactuals, and when valid causal inferences can be drawn. Graphical causal inference- the setting of this thesis - represents causal systems using directed graphs, with arrows from cause to effect. These are not just convenient visualisations but are a type of statistical model, which allow us to predict the distribution after intervening on variables. This makes them particularly well suited for reasoning about data quality, as statistical biases can be viewed as intervening or conditioning in these models. Causal machine learning began by using machine learning methods to answer questions in causal inference, particularly in high dimensional, large data settings where modern ML excels. Later, a stream of work began in the opposite direction, aiming to understand how causality can be used to alleviate some of the flaws of machine learning methods. Much of this focused on the issue of data quality, looking to understand how to make ML models that generalise beyond the data they are trained on. In this thesis, we initially focus on issues of data quality when using ML in classical causal problems, presenting two papers on this topic. Firstly, we discuss the problem of estimating causal effects from observational studies - which may be subject to unmeasured confounding - when a small amount of experimental data is available to de-bias estimates. We place theoretical limits of the effectiveness of these methodologies, and provide a Gaussian process assumption which permits valid inference. Secondly we present a paper discussing the problem of data merging for improved estimation of conditional causal effects. We frame this as a Bayesian experimental design problem, and develop a cryptographically secure method to evaluate the expected information gain. After this we move on to the second set of questions, asking what causality can do for the field of fair machine learning. Fair machine learning (or algorithmic fairness) is interested in understanding how ML models can be made compliant with legal anti-discrimination requirements in decision-making settings, such as employment, criminal justice, and healthcare. In order to answer this, significant effort was placed in trying to formalise mathematically what violations of these requirements would look like. Initially, this focused on measuring various statistics -such as model performance by group - hoping that problems of discrimination could be broken down into statistical disparities. However, two clear issues were found with this approach. Firstly, for every statistic it seemed possible to draw up a scenario where discrimination intuitively was/wasn't present despite the statistic saying it wasn't/was. Secondly, in most practical cases, it is impossible to be non-discriminatory or "fair" relative to multiple statistics simultaneously. This created the problem where one statistic alone couldn't capture the problem of fairness, but it was also impossible to have multiple. These concerns lead to the field of causal fairness, which aimed to solve this problem by providing such statistics with causal meaning. These works argued that discrimination is a causal effect of protected characteristics on outcomes. Framing things in this way lead to the development of numerous new fairness statistics, which crucially varied with causal context. We present two papers in the field of causal fairness. Firstly, we draw attention to the issue that selection bias plays in this context. We argue that from the perspective of causal fairness, selection bias is almost always present in fair ML applications. We then argue that this can create significant issues for the field as a whole, as it leaves the majority of causal effects fundamentally unidentified from observational data alone. Secondly, we look at the problem of data quality in fair machine learning more generally from a causal perspective. We provide a unified causal framework for multiple measurement biases that are typically present in fair ML applications. We then use tools from causal sensitivity analysis to create general sensitivity analysis tools to reason about the impact of measurement biases in Fair ML applications. Finally, to conclude this thesis we present some of the limitations with these works as they stand, alongside promising directions for future work
Optimal phasing of tidal stream power around the British Isles and within the English Channel for green ammonia production
Get PDFGreen ammonia, a fertilizer, energy carrier and shipping fuel, is a key zero-carbon chemical for the transition to net zero, but is produced in minimal quantities today, predominantly from wind and solar renewable energy. The waters around the British Isles and within the English Channel contain immense potential for predictable tidal stream energy, which is vastly underutilized today. The Haber-Bosch reactor, which is used to produce ammonia, is not flexible, requiring a smooth (consistent) power input. This paper analyses the potential for exploiting the difference in phase of tidal stream currents (tidal phasing) in different locations to optimize the aggregate power profile for the purpose of green ammonia production. A genetic algorithm is used to optimize the location of the turbines. For the four regions analysed in 2050, phasing is always beneficial – the levelized cost of ammonia (LCOA) is reduced by 6–13 % compared to an unphased, single turbine of the same capacity factor (CF), excluding cabling costs. Phasing is particularly evident in the Bristol Channel and in Alderney as their phased power profiles have infrequent zero or low power values. Although the cabling costs are significant, the tidal capital cost (CAPEX) always contributes more than the cabling CAPEX to the LCOA
Clinical prediction models using machine learning in oncology: challenges and recommendations
Get PDFClinical prediction models are widely developed in the field of oncology, providing individualised risk estimates to aid diagnosis and prognosis. Machine learning methods are increasingly being used to develop prediction models, yet many suffer from methodological flaws limiting clinical implementation. This review outlines key considerations for developing robust, equitable prediction models in cancer care. Critical steps include systematic review of existing models, protocol development, registration, end-user engagement, sample size calculations and ensuring data representativeness across target populations. Technical challenges encompass handling missing data, addressing fairness across demographic groups and managing complex data structures, including censored observations, competing risks or clustering effects. Comprehensive internal and external evaluation requires assessment of both statistical performance (discrimination and calibration) and clinical utility. Implementation barriers include limited stakeholder engagement, insufficient clinical utility evidence, a lack of consideration of workflow integration and the absence of post-deployment monitoring plans. Despite significant potential for personalising cancer care, most prediction models remain unimplemented due to these methodological and translational challenges. Addressing these considerations from study design through post implementation monitoring is essential for developing trustworthy tools that bridge the gap between model development and clinical practice in oncology
Needling as a Potential and Novel Treatment for Skin Ischemia following Filler-Induced Vascular Occlusion: A Case Series
Get PDFIntroductionThe most severe complication of filler injection is ischemia due to vascular occlusion, causing skin necrosis, scarring, blindness, and even stroke. Among the available treatments, needling therapy has been used in some clinical practices but has been rarely studied and discussed.Case presentationsFour recent skin ischemia cases with subsequent needling procedure were observed retrospectively and pertinent literature was analyzed. The objective was to evaluate the efficacy and safety of needling procedure for filler-induced skin ischemia, which is unresponsive to standard therapies.ConclusionAll of the 4 cases recovered from skin ischemia without side effects by receiving needling procedure. The available data demonstrate some potential mechanisms of needling, such as embolus releasing, ischemia reperfusion and revascularization. Although there is a lack of conclusive evidence for improving the hypoperfusion area by needling treatment, the current cases observation and theoretical analysis as well as our cases provide evidence supporting its potential as an efficacious, simple, and secure treatment for vascular complications
Can I get a little less life satisfaction, please?
Get PDFIn the social sciences and policymaking, life satisfaction surveys are increasingly taken as the best measure of wellbeing. However, the life satisfaction theory of wellbeing (LST) barely features in philosophers’ discussions of wellbeing. This prompts two questions. First, is LST distinct from the three standard accounts of wellbeing (hedonism, desire theories, the objective list)? I argue LST is a type of desire theory. Second, is LST a plausible theory of wellbeing? I raise two serious, underappreciated objections and argue it is not. Life satisfaction surveys are useful, but we should not conclude they are the ideal measure of wellbeing