123129 research outputs found
Sort by
From Theory to Practice: a Stochastic Framework for Utility-Driven Capital Allocation
Get PDFBottom–up approaches to economic modeling require stability and tractability (from the perspective of computation, and representation), if they are to be deployed at scale. This thesis develops a mathematical and computational framework that allows forecasts and optimization problems to be composed from agent–level building blocks while ensuring that stability and uniqueness are preserved. The central mathematical results are a bounded error–propagation lemma, showingthat perturbations in forecasts affect utilities only linearly in the error size, and a latent–space contraction theorem, showing that state trajectories converge uniquely when embedded into appropriate low–dimensional representations. These results provide the guarantees needed to make agent–level optimization problems well posed.
At the empirical level, the thesis contributes across several fronts. It demonstrates how unstructured textual data can be integrated into forecasting and portfolio theory, providing a basis for empirical decision–making. It develops embedding techniques for financial and macroeconomic time–series, compressing high–dimensional drivers into latent factors wherecontraction arguments apply. It also advances methods for automated causal discovery, extending analyst assumptions typical in real-world forecasts made by human analysts towards forecasts driven by much richer causal representations.
Overall, the thesis develops a framework in which contraction and error–propagation results provide mathematical support for the use of causal driver graphs in economic forecasting. The empirical chapters illustrate how textual information, latent embeddings of time–series, and automated causal discovery can be incorporated within this structure. The combination yields an optimization set up that is stable under perturbations of underlying forecasts and suitable for optimization in applied settings such as portfolio construction, macroeconomic analysis, or policy decision–makingPh.D
Toward Efficient and Robust Federated Learning with Heterogeneity
Get PDFFederated learning enables collaborative model training across decentralized data sources, preserving privacy by keeping data local and leveraging edge devices' computational power. Due to its strong privacy-preserving properties, federated learning is widely adopted in various real-world applications. However, scaling federated learning remains challenging due to issues like statistical heterogeneity, client staleness, adversarial attacks, and label noise. This dissertation introduces three novel frameworks—\emph{Polaris}, \emph{AsyncFilter}, and \emph{FedLID}—to address these challenges and enhance the robustness, scalability, and performance of federated learning.
Asynchronous federated learning improves training efficiency compared to synchronous approaches, yet its advantages are often limited by statistical heterogeneity and staleness. Hence, we present \emph{Polaris}, a theoretically grounded client selection strategy for asynchronous settings, balancing data quality and client response times through a geometric optimization framework.
Besides, the decentralized nature of federated learning, especially in asynchronous settings, increases its vulnerability to poisoning attacks, where malicious clients can compromise the global model with manipulated updates. To mitigate this, we propose \emph{AsyncFilter}, a plug-and-play defense module that detects and filters out poisoned updates in real-time, safeguarding model integrity during training.
Additionally, in real-world applications, label noise from human annotation errors complicates training, especially under data heterogeneity. By analyzing mislabeled data through deep representation subspace dimensionality, we identify distinct patterns in Local Intrinsic Dimensionality between noisy and clean samples. Leveraging this insight, we introduce \emph{FedLID}, a practical solution that refines mislabeled data and regularizes noisy samples, enhancing convergence behaviors.
Together, these contributions advance federated learning by improving its efficiency, security, and robustness in real-world settings, laying the groundwork for future research.Ph.D
Learning from the past, as we look to the future: A new approach for public health leaders responding to pandemics and epidemics
Get PDFHumanity has experienced pandemics and epidemics for centuries (Hays, 2009; Loomis, 2018; McNeill, 1998; Waltner-Toews, 2020). In every event, humanity seeks to learn from these experiences (Naylor et al., 2003; PHAC, 2018), but the lessons learned are often forgotten or neglected for those that follow (McLean & Hasham, 2020). This thesis will demonstrate the importance of knowing the history of pandemics and epidemics for public health leaders and what strategic leadership actions can be applied from these events to current and future pandemic and epidemic events.
The question to be answered in this research is the following: “What specific strategic leadership actions can public health leaders learn from the influenza pandemic of 1918 in Canada and the United States in responding to contemporary public health crises?” A secondary question is: “What sociocultural environment and social structures/systems in this specific public health crisis of 1918 in Canada and the United States do public health leaders need to account for in learning how to respond to contemporary public health crises?”. This research has used a thematic analysis to evaluate the responses of leaders in the influenza pandemic of 1918. The analysis was built on a historical research method based primarily on the study of secondary published materials and documents related to the 1918 influenza pandemic. The reading of these sources centred on specific strategic actions identified by those in leadership positions during the influenza pandemic of 1918 in both a Canadian and American context. The goal of this dissertation research is to lay the foundation for future work in developing a “pandemic playbook” (Goldschmidt, 2022), which would be based on a further analysis of the actions that various leaders have utilized in other pandemic and epidemic events in history.Ph.D
Synthesis, Self-Assembly, and Oxidative Doping of Conjugated Polymers with Complex Architectures
Get PDFConjugated polymers are lightweight, processible, and flexible semiconducting materials that are promising candidates in electronic devices. Their optoelectronic properties are contingent on adequate interchain π–π interactions which are promoted by crystallization. Group 16 conjugated polymers are often synthesized by Kumada catalyst transfer polycondensation (KCTP), a quasi-living polymerization. However, limited end group functionality by KCTP restricts the scope of post-polymerization modifications afforded to these materials. Chapter 1 begins by reviewing synthetic methodologies for conjugated polymers, with particular attention to KCTP. Then, the photophysical, electronic, and self-assembly properties of group 16 conjugated homopolymers and block copolymers are discussed. Chapter 2 introduces a functional Ni(II) external initiator for KCTP which bears a SNAr-active pentafluoro (PF) pendant. In conjunction with an established ethynylmagnesiusm bromide Grignard end-capping agent, the Ni(PF)(dppe)Br initiator is used to access polythiophene with heterobifunctional end groups possessing orthogonal click reactivity. We demonstrate the efficacy of this design by performing sequential copper-catalyzed alkyne-azide cycloaddition and SNAr click reactions at both polythiophene end groups to achieve a triblock copolymer with quantitative conversion. Chapter 3 reports the development of conjugated core-shell bottlebrush (CSBB) polymers with polythiophene and poly(ethylene glycol) (PEG) blocks in core or shell positions. These are attained by a similar Ni(II) external initiator strategy established in the previous chapter. Bottlebrush architectures with densely grafted and conformationally extended polymer side chains lead to controlled domain sizes and flexibility in materials and are of interest in optoelectronic materials. We show that CSBBs with polythiophene as a crystallizable conjugated core and PEG as a colloidally stabilizing shell facilitates their self-assembly into several crystalline morphologies with longer conjugation lengths and lower exciton bandwidths relative to analogous diblock copolymers. Different aggregation-inducing solvents adjust CSBB self-assembly into intramolecularly assembled nanoparticles, short nanofibers, or multi-micron fibers formed through end-on-end stacking. Side-by-side stacking is additionally induced in inverse CSBBs with shell-position polythiophenes. Finally, Chapter 4 extends our findings and describes the synthesis and self-assembly of CSBBs with polythiophene or polyselenophene cores and PEG shells. We find that oxidative doping of intramolecularly assembled and intermolecularly disaggregated polythiophene-based CSBBs enables a secondary self-assembly process into short flexible nanofibers.Ph.D
Inform and Do No Harm: Nocebo Effect of Mental Health Awareness and Approaches to Reduce It
Get PDFMental health awareness efforts are increasing, particularly for ADHD. Emerging evidence suggests general mental health awareness efforts may contribute to unnecessary self-diagnosis, yet it is unclear if they do so for ADHD while also exacerbating symptoms; there are also no validated approaches to reduce such harms. Two studies examined whether learning about undiagnosed ADHD increases self-diagnosis for healthy participants and whether nocebo education mitigates this effect. Across two pre-registered studies, participants participated in a workshop on ADHD, ADHD plus nocebo education, or control, and reported self-diagnosis and ADHD symptoms immediately after and one week later. Learning about ADHD increased self-diagnosis among healthy participants, even without symptom change for at least one week. Including a brief nocebo education component reduced this effect. Learning about ADHD alone can prompt self-diagnosis in individuals without the disorder. A brief nocebo education intervention prevents this outcome, highlighting a feasible way to balance mental health awareness with harm reduction.M.A
New Lower Bounds in Algebraic Complexity and Connections to Polynomial Identity Testing
Get PDFComputational complexity theory asks a deceptively simple question: What makes some problems fundamentally harder than others? By studying the resources needed to solve computational tasks---such as time, memory, and amount of allowed randomness---this field aims to reveal the fundamental limits of computation. Algebraic complexity theory focuses on problems that manipulate polynomials, using models like arithmetic circuits (where each internal gate is either simply an add gate or a multiply gate) to understand the cost of algebraic computation. This thesis investigates two central questions in this area: Can we prove that certain polynomials require immense computational resources, and can we remove randomness from key algebraic algorithms?
A key strategy toward the former question is ``hardness escalation". The idea is to prove lower bounds for simple, structured models (where analysis is tractable) and then use these results to gain insights into more general models. The first part of this thesis advances this approach by proving some of the strongest known lower bounds for structured models, namely, set-multilinear formulas and set-multilinear branching programs. These results suggest pathways towards the ultimate, decades-old goal of proving super-polynomial lower bounds for the corresponding general models.
The second part of this thesis addresses Polynomial Identity Testing (PIT): Given an arithmetic circuit, can we efficiently check if it computes the zero polynomial? While a simple randomized algorithm exists (evaluate the polynomial that the input circuit computes at random points), ``derandomizing" PIT---removing the randomness---remains a central challenge in complexity theory. This work bridges a critical gap: prior approaches required strong assumptions about computational hardness for efficient derandomization, but we come up with standard, minimal assumptions that we show to be both necessary and (nearly) sufficient. This considerably tightens the relationship between algebraic hardness and randomness, offering a clearer roadmap for future progress.Ph.D
Automated Coding of Counsellor and Client Behaviours in Motivational Interviewing Transcripts and Application to a Fully Generative Motivational Interviewing Chatbot
Get PDFMotivational Interviewing (MI) is a widely-used talk therapy approach employed by clinicians to guide clients toward healthy behaviour change. Evaluating MI sessions and training MI counsellors relies on behavioural coding, the classification of counsellor and client utterances into predefined categories. Recent advances in Large Language Models (LLMs) now make it possible to automate not only behavioural coding, but the delivery of MI itself. This dissertation introduces AutoMISC, which performs utterance-level parsing and behavioural coding under the Motivational Interviewing Skill Code, the original annotation scheme for MI. AutoMISC achieves an overall accuracy of 70% and a macro F1 score of 0.42 for counsellor speech (19 categories) and 0.41 for client speech (17 categories) against expert-aligned annotations using GPT-4.1 (n= 821 utterances). Additional validation showed that the codes predict session-level counselling quality in a widely-used MI transcript dataset at 87% accuracy, and align with existing annotations in another dataset at 71% accuracy. We also demonstrate how the codes can visualize the trajectory of client motivation over a session alongside counsellor codes. We apply AutoMISC to the transcripts of a brief smoking cessation intervention experiment where tobacco smokers conversed with a fully generative MI counsellor chatbot evolved in collaboration with experienced MI clinician-scientists. Two versions were tested: (1) a single prompted LLM (106 participants), and (2) a two-stage approach which decouples technique selection from utterance generation (93 participants). Participant-reported confidence in quitting smoking was measured before the conversation and one week later. Both versions yielded an average increase in confidence of 1.7 on a 0-10 scale (p<0.001 for both). The first version scored well on participant-reported perceived empathy, higher than typical human counsellors, while the second scored lower. AutoMISC’s analysis of the transcripts provided deeper insights beyond participant-reported outcomes. Both versions showed adherence to MI standards in 99% of utterances. We found that the slope of the trajectory of the client’s motivation correlates with the change in confidence (Spearman’s r= 0.28, p<0.005 for Version 1; r= 0.20, p= 0.051 for Version 2). This work demonstrates the potential synergy between automated MI delivery and automated MI behavioural coding.M.A.S
Characterizing Pseudomonas prophage-encoded antiphage defenses
Get PDFBacteriophages are viruses that infect and kill bacteria, posing a constant threat to bacterial survival. In response, bacteria have evolved a variety of antiphage defense mechanisms, most of which are encoded in the bacterial chromosome. However, temperate phages that integrate into bacterial chromosome as prophages can also contribute to host immunity by encoding antiphage systems. Once integrated, prophages rely on host survival and often carry accessory genes that enhance bacterial fitness, including resistance to phage infections. These prophage-encoded systems are increasingly recognized as important components of the bacterial immune repertoire, yet only a few have been mechanistically characterized.This thesis investigates two novel prophage-encoded defense systems found in Pseudomonas. The first, Tail Assembly Blocker (Tab), is encoded by the temperate phage JBD26 and is expressed constitutively during lysogeny. Tab recognizes the tape measure protein of infecting phages and inhibits tail assembly, resulting in non-infectious phage particles that lack tails. To avoid self-targeting during its own lytic cycle, JBD26 also encodes a counter-defense protein, Anti-Tab (Atab), which neutralizes the activity of Tab during the lytic cycle. This repsents a new class of defenses that directly target virion assembly.
The second system, Ring interacting pore 1 (Rip1), encoded by the prophage DMS3, operates through a different mechanism. Rip1 detects oligomeric phage structures that form during infection and uses them as scaffolds to assemble large pore-like structures. These membrane-disrupting pores kill the infected cells early and ultimately inhibit phage replication. Rip1 represents a novel defense strategy that combines both infection sensing and cell killing functions in a single, compact protein.
Together, these findings demonstrate that compact prophage-encoded systems can mediate robust phage defense through different mechanisms. This work expands the current understanding of bacterial immunity and highlights the potential of prophages as an underexplored and promising source of novel antiphage systems.Ph.D
Development of Environmental-friendly Chitosan-based Multifunctional Covalent Adaptable Network (CAN) Polymer Materials
Get PDFTo address the challenges involving polymer pollution, this thesis study developed a series of novel recyclable chitosan-based polymers possessing covalent adaptable networks (CANs). Chitosan contains many reactive sites (e.g., NH₂, OH) favorable for chemical modifications using cross-linking reactions. However, cross-linked bonds cannot be broken easily, which makes recycling/reprocessing processes challenging. Even though it is known that incorporating dynamic covalent bonds into a polymeric network enables material recycling/reprocessing, there is a lack of literature studies on using chitosan and dynamic chemistry to make CAN materials. Thus, the objective of this thesis study is to develop green chemistry synthesis methods for making multifunctional CAN materials from chitosan. By using a tripodal cross-linker, dynamic polyimine bonds were successfully introduced in chitosan via Schiff’s base reaction at moderate room temperature. The resulting chitosan-based 3D hydrogels showed excellent self-healing, antibacterial, and flame-retardant properties. To overcome chitosan’s inherent insolubility in common organic solvents, a catalyst-free dispersion method was devised for making chitosan-based CANs containing amide dynamic bonds (i.e., transamidation) at moderate condition. A series of chitosan-based CANs were obtained, and the effects of cross-linker type and ratios on recyclability and self-healing of the chitosan CAN materials were investigated. Moreover, dual dynamic bonds (both amide and ester bonds) were explored to improve the reformation efficiency and thermo-mechanical performance of chitosan-based CANs using a modular cross-linking strategy. The addition of ester dynamic bonds enhanced mechanical strength, adaptability, and recyclability, whereas self-healing ability was reduced. Additionally, we used a cross-linker of smaller molecular size in the vitrimerization process to form CAN in a chitosan-based partial vitrimer that showed enhanced recyclability and self-healing performance. The partial vitrimer was combined with ramie fabric to fabricate recyclable natural fiber composites with good mechanical and shape memory properties, reprocessibility, and biodegradability. The green synthesis approaches developed in this thesis study highlight the potential to use chitosan as a renewable feedstock to make environmentally-friendly circular products for various practical applications.Ph.D
Economic and Cultural Foundations of Radical Politics
No full textThe success of far-right political parties and movements in the twenty-first century has generated considerable debate about whether economic factors (e.g., income inequality and globalization) or cultural ones (e.g., racism and xenophobia) are primarily responsible. Yet there is much to suggest that support for them is shaped by popular discontent with prevailing economic inequalities combined with the salience of distinct identities and cultural frames in ways that defy this simple dichotomy. This dissertation’s three studies investigate these joint factors by documenting a series of links between economic phenomena, cultural dispositions, and voting preferences in contemporary industrialized democracies. The first study, “Politics of Boundary Consolidation,” develops and tests a theory linking income inequality to radical-right voting through national boundary consolidation. Using time-series cross-sectional data from 38 countries, it shows that inequality generates social status threats that promote ethno-nationalism, a cultural attitude the radical right mobilizes for electoral support. The second study, “Beyond Cosmopolitans and Nationalists,” challenges stylized accounts of contemporary identity-based political divisions. Using latent class analysis and survey data from 35 countries, it identifies distinct patterns of territorial identification that complicate a hypothesized binary between liberal, globally oriented voters and conservative, locally oriented ones. Instead, it shows that most Europeans are pluralists who identify with multiple territorial entities while only a subset of countries exhibits meaningful polarization between particularists, who emphasize local identities, and expansivists, who emphasize supranational ones. Where this division exists, it maps onto socio-economic divides and separates radical-right supporters from those of new left parties. The third study, “From Identity to Cleavage,” traces the evolution of territorial identification patterns in Europe using longitudinal data. It shows that contemporary identity configurations crystallized in the 1990s, during a period of European institutional reform, and have remained relatively stable, though parties can modestly reshape voters’ identification patterns during campaigns, as I demonstrate using panel data from the 2017 German federal election. Together, this research shows how economic conditions are linked to electorally consequential cultural dispositions, group boundaries, and collective identities, thereby synthesizing economic and cultural explanations of contemporary radical-right success while highlighting parties’ autonomous role in creating political cleavages.Ph.D