82837 research outputs found
Sort by
Model-Based Exploration in Monitored Markov Decision Processes
No full textTrial-and-error is reinforcement learning's core idea. The success of the trial-and-error learning hinges on the assumption that each trial would lead to feedback. As a result, the feedback is used to improve the quality of decisions taken. The assumption that decision makers receive feedback for all their actions at all times does not necessarily translate to real-world scenarios. For example, a human observer may not always be able to provide rewards, a sensor to observe rewards may be limited or broken, or rewards may be unavailable during deployment. Monitored Markov decision processes (Mon-MDPs) have been proposed as a framework for sequential decision making where rewards could be unavailable, or in other words, unobservable to the decision maker. In this thesis, we consider Mon-MDPs. We revisit Mon-MDPs' model of interaction and the objective of decision makers in this model. Then, we introduce our main contribution, the monitored model-based interval estimation with exploration bonus (Monitored MBIE-EB) algorithm. Monitored MBIE-EB is the first algorithm in Mon-MDPs that provably admits a polynomial sample complexity. This polynomial sample complexity is to achieve a minimax-optimal policy in the worst case. Monitored MBIE-EB pays attention to the structure of the problem at hand and furthermore is able to benefit from prior knowledge about the problem. Prior knowledge about the observability of the rewards is important. We show that Monitored MBIE-EB is able to fully exploit this knowledge, if available. Also, we show Monitored MBIE-EB is also capable of finding the minimax-optimal policy in the absence of privileged prior knowledge. Empirically, we demonstrate the superior performance of Monitored MBIE-EB compared to Directed Exploration-Exploitation, the state-of-art algorithm in Mon-MDPs, on four dozen finite domains
Power Quality and Dynamic Enhancement of Remote Gas Field Generation based on Smart Power Converters
No full textNatural gas is a cornerstone of global power generation, balancing reliability and environmental sustainability. In Alberta, Canada, which holds one of the largest natural gas reserves in the country, natural gas-fired power generation plays a crucial role in meeting both provincial and export energy demands. As many of these resources are situated in remote northern and western regions, building power plants near gas fields presents a practical solution to minimize transmission losses and lower fuel transportation costs, thereby enhancing efficiency and reducing environmental impact. Additionally, integrating on-site greenhouses allows residual heat and carbon emissions to be repurposed for crop cultivation, further enhancing the sustainability of power generation.
However, due to their remote locations, these plants are typically connected to weak utility grids with high line impedance, which exacerbates power quality issues. Nonlinear loads, such as fluctuating demands from greenhouse grow lights and residential heating, ventilation, and air conditioning (HVAC) systems, lead to overvoltage and harmonic distortions. Moreover, the weak grid environment amplifies dynamic interactions among multiple generators, further challenging system stability. These issues threaten the stability and reliability of power systems, necessitating innovative solutions to enhance operational efficiency.
To address these challenges, this thesis proposes a system architecture where the gas turbine generator is interfaced with the grid via a back-to-back (B2B) converter. Control strategies are developed for both grid-following and grid-forming modes, enabling coordinated regulation of active power, AC voltage, and\nDC-link voltage. Leveraging flexible inverter controls, the proposed approach enhances both power quality and the dynamics in multi-machine systems.
Building on this foundation, the thesis also investigates control selection and parameter tuning through eigenvalue analysis to ensure stability under varying grid strengths. Small-signal analysis of the grid-side converter provides insight into stability conditions and aids in optimizing parameter design. For large disturbances such as grid-side faults, a crowbar protection strategy is introduced to absorb the power imbalance between converters, effectively suppressing DC-link overvoltage and improving fault ride-through capability.
To validate the proposed methods, a practical gas-based power system is modeled and tested through real-time simulations using RT-LAB. The results confirm substantial improvements in power quality and dynamic performance. Additional analysis highlights the influence of grid strength and control parameters on overall system behavior, and verifies the effectiveness of the crowbar strategy in protecting the B2B converter during faults.
Overall, the proposed methods significantly enhance the power quality and operational stability in remote gas-based power systems. The thesis also provides comprehensive guidance on system configuration, including control strategy selection, parameter tuning under different grid strengths, and fault response design. These findings contribute to the development of more resilient, efficient, and sustainable gas generation systems in remote situations
Investigating Altered Retrosplenial Cortex Neuronal Activity in Shank3B Mutant Mouse Model of Autism Spectrum Disorder
No full textAutism Spectrum Disorder (ASD) is a neurodevelopmental condition which manifests in the form of a complex range of repetitive and restrictive behaviours, sensory processing difficulties, and social interaction and communication deficits. Disruptions to excitatory and inhibitory balance within the brain have been proposed as a core pathophysiological mechanism of autism with reports suggesting that individuals with ASD display differential activity across many cortical regions and functional networks. Following the dysregulation of excitation-inhibition, mutations in the Shank3 gene, which encodes an excitatory postsynaptic scaffolding protein, have been among the most investigated genetic causes of ASD. Loss of Shank3B has been shown to cause hyperpolarization-related channelopathy and enhanced pyramidal neuron activity. Although these disruptions have been identified across various cortical regions, the involvement of the retrosplenial cortex (RSC) in the pathophysiology of ASD remains to be mapped at the cellular level. Using the Shank3B knockout mouse model to recapitulate key features of ASD, we investigated neuronal activity patterns and cell population synchrony in vivo within the RSC using two-photon calcium microscopy. As animals were allowed to freely move during head fixation, we recorded fluorescent calcium activity in layers II/III RSC neurons and found that although RSC neurons demonstrated positive tuning to locomotor behaviour, the global deletion of Shank3B did not significantly alter neuronal firing rates and activity patterns as well as pairwise neuron-neuron coordinated activity. Correlation of cell activity with the animal’s changing speed was also comparable to non-mutant mice. Neuronal assembly architecture and correlated activity within the RSC were unaffected by postsynaptic Shank3B loss and exhibited normal activation patterns. Our findings demonstrate that the dysregulation of cortical excitation is not linear in ASD and appears unaffected at the neuronal-level of the RSC. This suggests other, not yet identified, factors or influences which may underlie the broader previously reported global and functional alterations of the RSC. Autism Spectrum Disorder (ASD) is a neurodevelopmental condition which manifests in the form of a complex range of repetitive and restrictive behaviours, sensory processing difficulties, and social interaction and communication deficits. Disruptions to excitatory and inhibitory balance within the brain have been proposed as a core pathophysiological mechanism of autism with reports suggesting that individuals with ASD display differential activity across many cortical regions and functional networks. Following the dysregulation of excitation-inhibition, mutations in the Shank3 gene, which encodes an excitatory postsynaptic scaffolding protein, have been among the most investigated genetic causes of ASD. Loss of Shank3B has been shown to cause hyperpolarization-related channelopathy and enhanced pyramidal neuron activity. Although these disruptions have been identified across various cortical regions, the involvement of the retrosplenial cortex (RSC) in the pathophysiology of ASD remains to be mapped at the cellular level. Using the Shank3B knockout mouse model to recapitulate key features of ASD, we investigated neuronal activity patterns and cell population synchrony in vivo within the RSC using two-photon calcium microscopy. As animals were allowed to freely move during head fixation, we recorded fluorescent calcium activity in layers II/III RSC neurons and found that although RSC neurons demonstrated positive tuning to locomotor behaviour, the global deletion of Shank3B did not significantly alter neuronal firing rates and activity patterns as well as pairwise neuron-neuron coordinated activity. Correlation of cell activity with the animal’s changing speed was also comparable to non-mutant mice. Neuronal assembly architecture and correlated activity within the RSC were unaffected by postsynaptic Shank3B loss and exhibited normal activation patterns. Our findings demonstrate that the dysregulation of cortical excitation is not linear in ASD and appears unaffected at the neuronal-level of the RSC. This suggests other, not yet identified, factors or influences which may underlie the broader previously reported global and functional alterations of the RSC
Sustainable Laundry Practices and Microplastic Pollution: Behavioral Insights and Strategies for Reducing Environmental Impact
No full textWhen assessing the environmental impact of textiles during their use phase, consumer behaviour and garment care regimes are crucial. Using an unlabeled (Paper I) and labelled (Paper II) discrete choice experiments (DCE), this thesis examines consumer laundering behaviours. The first paper examines the trade-offs people make based on the type of fibre (polyester, cotton, or wool), the occasion of use (exercise, casual, or formal), sweat level (light, moderate, heavy), number of times worn (once, twice, three times) and behavioural factors like cleanliness, disgust, and environmental consciousness. The second paper studies consumers’ willingness to adopt a microfibre filter based on its efficiency, replacement interval and price.
Two surveys were conducted to gather data. In the first study participants were split up into three treatment groups. Group 1 was given a short text about the environmental effects of laundry, Group 2 was informed about the value of hygiene, and Group 3 was a control group. The findings from the first study indicated no significant differences between the treatment groups; therefore, participants in the second survey were not assigned to separate treatment groups. The data was analyzed using a multinomial logit model (MNL) and a random parameter logit (RPL) model, which examined heterogeneity as well as consumer preferences and decision-making.
The findings of the first paper indicate that there is a great deal of variation in the ways that fibre type, occasion, and behavioural factors affect laundry decisions. Exercise clothes were more likely to be laundered than casual or formal attire, while wool clothing was less likely to be washed than cotton and polyester. Sweat intensity and the number of wears significantly increased the likelihood of laundering, but these effects were moderated by sustainability considerations in some cases. Consumer behaviour is complicated and variable by environmental\nawareness, disgust, and cleanliness considerations, even if practical criteria like wear frequency and sweat level heavily influence laundry decisions.
The results of the second paper indicate that higher filter costs reduce consumers’ willingness to adopt, whereas increased efficiency enhances it. Replacement intervals had mixed effects; for some individuals, longer intervals discouraged adoption likely due to the associated additional costs. However, many consumers still preferred the filter option with the highest efficiency, suggesting that performance outweighed concerns about maintenance frequency. The study also revealed that individuals with greater knowledge of and concern about microfibre and microplastic pollution were significantly more inclined to adopt a filter compared to those who lacked awareness or concern about the issue.
These findings demonstrate the necessity of focused regulations and consumer education campaigns to encourage eco-friendly laundry practices that strike a compromise between practicality and environmental stewardship. By offering practical insights into the relationship between sustainability, behaviour, and textile care, this work contributes to domains of resource economics and consumer behaviour
Assessing and rating telecommunication infrastructure by social importance and physical vulnerability to wildfire in Alberta, Canada
No full textWildfires pose a significant threat to public safety, making effective communication among responding agencies and with the public crucial during escalated wildfire situations. However, telecommunication vulnerabilities to wildfire have been largely overlooked in fire risk and human dimensions research. In Alberta, a significant proportion of telecommunications infrastructure is positioned in fire-prone regions of the province, with over 600 cellular telecommunications sites contained within the Forest Protection Area (FPA). Proactive mitigation measures such as fuel treatments can reduce the potential risk to and build resilience in communication infrastructure, further strengthening the overall resilience of rural and remote communities. Although, with finite resources available, prioritization is needed to effectively mitigate the risk at the most important and vulnerable towers.
This study had two primary objectives. First, I investigated telecommunications vulnerabilities to wildfire in Alberta, Canada with the following objectives: (1) determine the extent of cell phone coverage in Alberta in relation to service demand areas (SDAs); (2) identify to what extent telecommunications infrastructure and SDAs are exposed to potential wildfire; and (3) identify SDAs that have high wildfire exposure and potentially limited cell coverage. I used a fire exposure metric, directional vulnerability assessments, and communications viewshed to identify fire-exposed regions that lack telecommunications coverage and identify the infrastructure most at risk within the province of Alberta. Public safety was assessed through the analysis of Service Demand Areas (SDAs) in relation to gaps in telecommunications coverage and fire exposure. Second, I sought to develop a systematic rating system based on physical vulnerability to wildfire and social importance with the goal of providing a framework for rating telecommunication infrastructure based on multiple factors of vulnerability to allow service providers and forest managers to prioritize those for structure protection or build resilience around those that are lacking.
A heterogeneous pattern of telecommunications coverage is distributed across the province, influenced by population densities and roads, and obstructed by terrain. A significant number of telecommunications towers were distributed in areas of high fire exposure (n = 186), possessing hundreds of potential pathways viable to fire transmission. Thousands of kilometres of roadway and hiking trails lack sufficient telecommunications coverage in tandem with being exposed to wildfire. The Grande Prairie and Slave Lake Forest Areas are highlighted as containing the greatest number of telecommunications towers that are both physically vulnerable to fire-induced damage and hold a large social importance. This study provides a comprehensive analysis of the current state of the vulnerabilities to wildfire within the telecommunications network in Alberta. Results and findings reported herein may inform wildfire resilience and mitigation plans, fuel treatments, and future improvements of telecommunications infrastructure
Tumour Secreted Nucleosides Degrade Myocardial RBFOX1 and Promote Apoptosis Susceptibility; Implications for Chemotherapy Induced Cardiotoxicity
No full textCancer patients are living longer in part because of improvements in early detection and treatment; however, a severe side effect of treatment, chemotherapy induced cardiotoxicity (CIC), persists. This condition can occur in 4-20% of patients and can result in cessation of an otherwise attractive chemotherapy to instead preserve heart function. A major gap in the field of cardio-oncology is that there is limited understanding as to why some patients develop this condition whereas others with the same cancer receiving the same chemotherapy do not. Consequently, there is a lack of biomarkers to predict CIC development and little mechanistic understanding on the molecular changes occurring in the myocardium prior to chemotherapy mediated damage. It is well-understood that the tumour can secrete various factors into the bloodstream that can affect distal organs; however, it remains unresolved if tumour secreted factors could negatively affect the heart to initiate a signalling cascade promoting cardiomyocyte susceptibility to apoptosis. Here, we identify a transcription factor, ZNF281, promotes tumour secreted inosine and hypoxanthine which circulate in the bloodstream and bind the A2A receptor on cardiomyocytes to activate CAMKII which degrades a mRNA splicing factor, RBFOX1, that is critical for cardiac maturation. Loss of RBFOX1 promotes a less mature myocardium which results in expression of stemness markers, loss of differentiation proteins, mPTP assembly, a greater proportion of mono- nucleated to bi-nucleated cardiomyocytes and a more relaxed chromatin state. This results in increased susceptibility to DNA damaging and alkylating anticancer agent incorporation (via a more accessible chromatin) and thus increased activation of nuclear DNA damage sensing or pro-apoptotic transcriptions factors that promote cardiomyocyte apoptosis and cardiotoxicity. Through a combination of tumour models, human breast cancer patients from the MANTICORE trial, mass spectrometry, gastric and secretory specific- or cardiomyocyte-specific transgenic mouse models, primary adult murine cardiomyocyte, RNA sequencing, doxorubicin mediated cardiotoxicity mouse models, and analysis of human anthracycline induced cardiotoxicity hearts, we demonstrate that cumulative tumour secreted inosine and hypoxanthine is a highly sensitive predictive biomarker for chemotherapy induced cardiotoxicity in patients and that this may be due to RBFOX1 loss mediated susceptibility to DNA damaging anti-cancer agents. Our findings identify a novel predictive biomarker for cardiotoxicity development and new potential therapeutic targets, tumour ZNF281 and myocardial RBFOX1, to prevent CIC
Personalized Recommendations via Active Utility-based Pairwise Sampling
No full textRecommender systems play a critical role in enhancing user experience by providing personalized suggestions based on user preferences. Traditional approaches often rely on explicit numerical ratings or assume access to fully ranked lists of items. However, ratings frequently fail to capture true preferences due to users' behavioral biases and subjective interpretations of rating scales, while eliciting full rankings is demanding and impractical. To overcome these limitations, we propose a generalized utility-based framework that learns preferences from simple and intuitive pairwise comparisons. Our approach is model-agnostic and designed to optimize for arbitrary, task-specific utility functions, allowing the system's objective to be explicitly aligned with the definition of a high-quality outcome in any given application.
A central contribution of our work is a novel utility-based active sampling strategy for preference elicitation. This method selects queries that are expected to provide the greatest improvement to the utility of the final recommended outcome. We ground our preference model in the probabilistic Plackett-Luce framework for pairwise data. To demonstrate the versatility of our approach, we present two distinct experiments: first, an implementation using matrix factorization for a classic movie recommendation task, and second, an implementation using a neural network for a complex candidate selection scenario in university admissions. Experimental results demonstrate that our framework provides a more accurate, data-efficient, and user-centric paradigm for personalized ranking
Essays in Empirical Corporate Finance
No full textThis dissertation explores key aspects of corporate governance, information disclosure, and financial market efficiency through three empirical studies. By applying advanced natural language processing (NLP) techniques and econometric methods, it provides novel insights into how corporate communication and governance structures influence investor perceptions and firm valuation.
The first chapter examines the predictive power of CEO language in earnings calls for upcoming leadership transitions. Using a combination of a state-of-the-art financial language model for sentiment analysis, FinBERT, and a traditional word-counting approach, the study finds that in meetings preceding the public announcement of a leadership transition, CEOs exhibit changes in business tone, stress levels, and self-attribution. These linguistic shifts are particularly pronounced in cases of forced departures. In contrast, voluntary turnovers are characterized by higher self-attribution and a reduction in negative tone, reflecting a smoother transition. The findings indicate that CEO speech patterns provide incremental predictive power beyond standard financial performance measures. Moreover, investors appear to process these implicit signals, as firms with CEOs displaying high turnover hazard indicators experience significant market reactions and increased trading volume following earnings calls.
The second chapter, co-authored with Efstathios Avdis, examines the consistency of corporate financial disclosures by estimating the degree of textual similarity between earnings call narratives and subsequent formal filings. To explore how processing costs influence the informativeness of disclosures, we introduce a novel methodology—text-on-text regression—that estimates textual similarity by regressing one document on another. Our findings emphasize the important role of disclosure timing and textual coherence in shaping investor reactions, trading activity, and predicting firm performance. These results indicate that the similarity between financial disclosures conveys valuable signals to market participants, emphasizing the importance of analyzing the consistency of corporate communication channels rather than viewing them as separate information sources.
The third chapter explores the stock market’s response to board diversity, particularly in the context of social movements emphasizing racial equity. During the heightened public attention to racial discrimination following the Black Lives Matter protests, firms with at least one Black director on the board earned abnormal returns of 3.59% over a six-day period. This effect was concentrated among firms headquartered in states with a low fraction of the Black population, supporting the hypothesis that markets particularly reward intentional racial diversification that exceeds local demographic expectations. The findings suggest that diversity signals can enhance firm valuation, particularly in periods of social activism when investors reassess the value of inclusive governance structures.
Together, these essays contribute to the understanding of corporate governance, financial disclosure strategies, and investor behavior. They highlight the importance of qualitative information—ranging from CEO communication patterns to disclosure consistency and board composition—in shaping market outcomes. By integrating NLP techniques with traditional financial modeling, this dissertation offers a multidimensional perspective on how firms communicate, disclose, and respond to a dynamic information environment
Radiation Impedance of Rectangular CMUTs
No full textCapacitive Micromachined Ultrasonic Transducers (CMUTs) are an alternative to piezoelectric transducers, offering several potential advantages, including broad bandwidth, mass manufacturability built on standard Micro-Electro-Mechanical Systems (MEMS) processes, and the potential for integration with electronics. Additionally, unlike most piezoelectric transducers, CMUTs are lead-free, addressing recent regulatory pressures regarding the restriction of hazardous substances in medical devices. However, CMUTs have long faced performance and reliability challenges, such as dielectric charging and operational hysteresis. Other factors affecting their performance include poor utilization of real estate, parasitic capacitance, potential for electrical breakdown, premature collapse, ultrasonic welding, and poor electromechanical efficiency when multiple small CMUT cells are used within an element. Recently, our group demonstrated that long rectangular membranes utilizing electrode posts in the cavities not only mitigate many of these reliability issues but also achieve exceptional electromechanical efficiency. In fact, they have exhibited transmit efficiency exceeding that of piezoelectric transducers by nearly threefold. However, these advances lacked a framework for modeling or optimizing this new architecture. One limitation is the absence of an equivalent circuit model for rectangular membranes, partly due to the lack of a method for calculating their radiation impedance. This thesis aims to address this gap. We introduce an approximate model for the radiation impedance of rectangular membranes and validate it against finite element computations. Furthermore, our models have been integrated into equivalent circuit models and tested against both computational results and experimental data