21684 research outputs found
Sort by
Addressing data scarcity in autonomous systems through trustworthy counterfactual generation
Get PDFAutonomous systems often operate in environments where collecting large, diverse, and safety-critical datasets is difficult. This data scarcity limits their reliability, particularly in rare or hazardous scenarios that are hard to capture in the real world. This thesis addresses data scarcity by integrating structural causal models with diffusionbased generative models to produce trustworthy, high-fidelity counterfactual images for “what-if’’ reasoning. Thus, two frameworks are proposed: Causal DiffuseVAE and Causal DiffuseLLM. Both generate images that follow a directed acyclic graph of semantic factors while preserving visual realism. The thesis first outlines key concepts in causal generative modeling and modern deep generative methods, highlighting that existing approaches either provide interpretable causal control with limited fidelity or achieve photorealism without reliable intervention behavior.
Causal DiffuseVAE structures the latent space using a causal graph and applies a diffusion decoder for detail reconstruction. Experiments show a 40% reduction in generation time and a 30% improvement in counterfactual accuracy compared with state-of-the art causal diffusion models. Causal DiffuseLLM, which maps language instructions to causal interventions, improves generation accuracy by 15% over its non-LLM baseline and localizes edits to causally affected regions.
Overall, this thesis shows that embedding causal reasoning into diffusion pipelines provides a practical path to generating reliable data for autonomous systems operating under limited data conditions
Three essays in Bayesian microeconometrics: embracing causality and heterogeneity
Get PDFThis thesis leverages Bayesian methods to address econometric challenges in microeconomic settings, with a focus on causality and heterogeneity. The contributions are provided in three essays.
The first essay (Chapter 2) proposes a novel approach, Bayesian Analogue of Doubly Robust (BADR) estimation, to estimate unconditional Quantile Treatment Effects (QTEs) in observational studies. This estimand offers valuable insights into treatment effect heterogeneity across different outcome ranks. By incorporating Bayesian machine learning techniques, the framework can effectively handle high-dimensional covariates and nonlinear relationships to achieve better accuracy and appropriate uncertainty quantification. The simulation results show that BADR estimators yield a substantial improvement in bias reduction for QTE estimates compared with popular alternative estimators found in the literature. I revisit the role of microcredit expansion and loan access on Moroccan household outcomes, demonstrating how the new method adds value in characterising heterogeneous distributional impacts on outcomes and detecting changes in overall economic inequality, which is also appealing to other applied contexts.
The second essay (Chapter 3) introduces a new approach that harnesses network or spatial data to identify and estimate direct and indirect causal effects in the presence of selection-on-unobservables and spillovers. The proposed framework nests the Generalised Roy model to explicitly account for endogenous selection into treatment and goes beyond to capture spillovers through exposure mapping to neighbours’ treatment. This allows for heterogeneous effects across individuals and enables exploration of various policy-relevant treatment effects. I develop Bayesian estimators based on data augmentation methods, offering efficient computation and proper uncertainty quantification, which is supported by simulation experiments. I apply the method to evaluate the Opportunity Zones (OZ) program, which aims to stimulate economic growth in distressed U.S. census tracts through tax incentives. The results show both direct and indirect positive impacts on housing unit growth in designated Qualified Opportunity Zones (QOZs), but unselected tracts (non-QOZs) experience no beneficial spillovers, remaining at a disadvantage. Moreover, the model predicts that offering investment tax credits to non-QOZs would lead to negative outcomes, making the program’s expansion to these areas ineffective.
The third essay (Chapter 4) is based on a joint work with Dr Santiago Montoya-Blandón. We develop a new econometric framework for modelling network interactions with heterogeneous effects, while addressing the issue of network endogeneity. The proposed Selection-corrected Heterogeneous Spatial Autoregressive (SCHSAR) model overcomes the limitations inherent in the standard spatial autoregressive (SAR) specification by achieving these dual objectives. We incorporate a finite mixture structure to capture rich heterogeneity in network interaction effects and explicitly model link formation, with latent variables playing a crucial role. For estimation and inference, our fully Bayesian approach effectively handles the computational challenges arising from the complex likelihood function and latent structure. We present a simulation study that validates the proposed approach. In the empirical application to an innovation network among American firms, we reveal significant positive yet heterogeneous interaction effects on corporate R&D investments, after accounting for endogenous network formation. The findings highlight different firm behaviours and reveal notable transmitters and absorbers in response to exogenous R&D policy shocks. This framework enables quantification of firm-level direct and spillover effects, thus providing valuable insights for evidence-based and targeted policy design.
By utilising recent developments in Bayesian econometrics, my research seeks to overcome the limitations of conventional methods, particularly in handling high-dimensional models, endogeneity, heterogeneity, and several forms of spillovers. Ultimately, the proposed methods enable more flexible and robust microdata analysis, contributing to a deeper understanding of individual and group differences in economic behaviour, as well as causal effects. This, in turn, can lead to more informed and effective policy decisions
Evolution and diversity of One Health bacterial pathogens within complex multi-host environments on the endemic and global scale
No full textAbstract not currently available
Broken brotherhood: understanding child sexual abuse by Marist brothers and former Marist brothers in Australia
Get PDFThe purpose of this research was to interview four Marist brothers and four former Marist brothers who had abused children to ascertain their understanding of the factors that contributed to the abuse. Interpretive phenomenological analysis was used as the methodology for the research, using semi-structured interviews. The findings were interpreted in the light of the reports of commissions of inquiry and psychological research. The findings supported the view that abuse by clerics is a complex phenomenon that involves personal, situational, organisational, and environmental factors. The results are consistent with the findings from published qualitative research with priests and brothers, and the data from commissions of inquiry, including the importance of opportunity as a significant factor in the abuse of children by clerics.
This is the first research project where the participants were all members of the same Roman Catholic religious order of brothers, and from the same country. The results were explored from the perspective of the vocation of brotherhood and, in particular, Marist brotherhood. St Marcellin Champagnat, the founder of the Marist Brothers, was opposed to physical, sexual, and emotional abuse and neglect, and established rules to prevent the abuse of children and to protect the brothers from the temptations of inappropriate behaviour. The regulations regarding relations with pupils and how to live a life of chastity from the time of the Founder until the mid-20th century were explored, especially the changes that took place at the time of Vatican II (1963–65) that led to significant developments in the theology and vocation of brothers in the Church. These developments enabled the Marist Brothers to respond to the abuse crisis by benefiting from insights from psychology and counselling, and from developments in theology, as well as by extending the mission and spirituality of the Marist Brothers to lay men and women.
The themes of guilt, shame, redemption, membership, identity, loss, and belonging were explored from the perspective of psychology. Moral injury was proposed as a concept that could be applied to victims, secondary victims, perpetrators, and bystanders, as all felt betrayed and lost trust in individuals in leadership and in the Church as an institution. One of the implications of this research is that clerical status is not a significant variable regarding the abuse of children by priests and brothers. The research also showed that individuals are able to build new lives and identities and live meaningful lives after discovery and imprisonment. Brotherhood as a value and (for some) an identity, continues to have validity whether they continue to live as Marist brothers, leave, or are dismissed from the Order
Empirical investigation of selection and evolution processes by causal molecular inference
No full textAbstract not currently available
Trust, testimony, and transmission: Essays in social virtue epistemology
Get PDFThis thesis consists of five distinct essays within social virtue epistemology, each of which can stand independently, yet all engage with fundamental ideas surrounding trust, testimony, and knowledge transmission. The first two chapters explore knowledge transmission and testimony through a virtue epistemological lens, emphasising the challenges of accounting for testimonial knowledge while maintaining a connection between knowledge and credit. I introduce types of knowledge transmission that do not rely on joint agency or shared intentions, challenging a prominent view in virtue epistemology. I present a type of a credit view that can defend one of the fundamental doctrines of credit views, that knowledge always entails credit, from challenging counterexamples. Trust and testimony both facilitate connections between individuals, making them central to our understanding of how knowledge is shared in social contexts. The third chapter aims to further our understanding of the nature of trust by placing the spotlight on trust features that have gone largely unnoticed, namely, their temporal elements. By expanding on these features, we can make meaningful distinctions between instances of trust that have generally been considered interchangeable. These distinctions and related concepts highlight the subtle differences that meaningfully impact how we approach trust. In the fourth chapter, the focus shifts to epistemic groups in the context of gatekeeping. Epistemologists should be interested in trust, testimony, and transmission as they relate to individuals, but groups are an interesting epistemic subject in their own right. This chapter examines the distinct epistemic roles groups play in shaping the beliefs of their members, and how individuals can benefit from being part of a collective. I then present conditions for justified epistemic gatekeeping and consider what kinds of groups are most capable of fulfilling those conditions. In the last chapter, I consider how to define general artificial intelligence. It is difficult to place large language models within epistemology. At times, they act like epistemic agents, seemingly capable of producing and transmitting knowledge, yet they often appear incompetent and incapable of performing simple tasks. I propose a virtue-theoretic distinction between narrow and general artificial intelligence, in the hopes that it can contribute to our understanding of what makes AI trustworthy, and whether we should think of their predictions as knowledge
Factors shaping young adults decision-making in full-time higher education: Insights from young adults in mainland China during the COVID-19 pandemic
Get PDFThis study investigated factors influencing the decisions of young adults (ages 20-40) in mainland China regarding full-time higher education during the COVID-19 pandemic. As a significant and disruptive social event, the pandemic drastically altered individuals’ life trajectories, likely including their educational participation behaviours. By examining this topic, I aimed to understand how the pandemic reshaped decision-making processes and the potential implications these changes had for educational participation. Through a review of existing theoretical frameworks and empirical research, I developed a multi-level theoretical model - encompassing individual, situational, and institutional levels - to guide this empirical investigation. I employed a multi-phased, multi-method qualitative research strategy, which involved three stages of data collection and analysis: Stage A consisted of critical policy analysis; Stage B involved online semi-structured interviews; and Stage C utilised online focus groups conducted via Padlet. The study yielded significant findings from all three levels, and, through triangulation of results, identified three key messages extending the initial framework. These include the critical role of sunk costs, the impact of non-economic factors, and specific characteristics of young adults as a demographic cohort of study, such as the accumulating decision-making costs of participating in full-time higher education during the pandemic, which can interfere with potential participants’ judgment in making educational decisions. These insights provide valuable contributions to future educational research and the development of lifelong learning policy and practice for stakeholders in Mainland China, and globally when facing future global crises scenarios
Addressing challenges in food allergy and anaphylaxis preparedness in Scottish schools
No full textAbstract not currently available
The male lens on Jane Eyre: translating/constructing femininity across a century of Chinese cultural history
No full textAbstract not currently available
Data sources:
Access Repository Dashboard
Do you manage Open Research Online? Become a CORE Member to access insider analytics, issue reports and manage access to outputs from your repository in the CORE Repository Dashboard! 👇