21090 research outputs found
Sort by
Generative Models for Planning and Decision-Making
No full textGenerative models have achieved remarkable progress across domains such as vision and language. However, their application to sequential decision-making and planning remains challenging. In reinforcement learning and robotics, agents must handle task hierarchies, long-horizon dependencies, adapt to harder unseen tasks and environments, and, especially in multi-agent settings, respond to adversarial or evolving opponents. Despite progress in behavioral cloning and offline policy learning, existing approaches often struggle to generalize beyond the train distribution or to learn robust, interactive behaviors in competitive games. These limitations restrict current systems to narrow tasks with short temporal horizons, or deterministic settings. For instance, behavioral planners trained on single-goal environments struggle scaling to multi-task missions requiring subgoal discovery and adaptive reasoning, as there is no straightforward mechanism for iterative test-time adaptation to these unseen tasks. Similarly, in multi-agent reinforcement learning, standard policy optimization often yields unimodal, brittle strategies that overfit to specific opponents and fail to converge to a Nash equilibrium in continuous state-action games. This thesis explores challenges and opportunities in using generative models for planning and decision-making tasks, specifically energy-based and diffusion-based models which serve as both representations and solvers for planning and policy learning. In the single-agent setting, we introduce GenPlan, a discrete-flow planner that reframes planning as iterative denoising over trajectories using an energy-guided diffusion process. This formulation enables task and goal discovery, and adaptation to unseen environments. In the multi-agent setting, we propose DiffFP, a diffusion policy gradient method within the fictitious play framework. By approximating best responses through diffusion models, DiffFP captures multimodal strategies, improves sample efficiency, and remains robust to evolving opponents in dynamic, continuous state-action games. Our empirical studies show that GenPlan outperforms baselines by over 10% on adaptive planning tasks, generalizing from single-task demonstrations to complex, compositional multi-task missions. Likewise, DiffFP achieves up to 3× faster convergence and 30× higher success rates compared to other baseline reinforcement learning algorithms in multi-agent benchmarks. These results demonstrate the potential of generative modeling not only for representation learning, but as a unified substrate for planning, learning, and decision-making across settings
Enhancing Power Fuzzing: Synthetic Side-Channel Data Generation, Optimal Sampling, and Noise Mitigation
No full textEmbedded systems increasingly dominate critical applications, driving the need for advanced testing and validation methodologies capable of uncovering hidden or undocumented behaviours. Traditional fuzzing approaches, which rely on observable outputs or system crashes, often fail to reveal the internal operations of embedded devices. Powertrace-based fuzzing provides a non-intrusive alternative by analysing a device’s power consumption during operation. Achieving robust and reliable fuzzing performance requires researchers to overcome significant challenges in signal acquisition, noise mitigation, and classification reliability.
This thesis addresses these challenges by introducing several key improvements to the PowerFuzzer framework. First, it develops SigVarGen, a modular synthetic signal generation framework that produces realistic idle-state and active signals under controlled noise, drift, and timing variations. SigVarGen enables comprehensive algorithm development and stress testing across diverse simulated conditions, bridging the theoretical model design and empirical validation gap. Second, it presents SR\&OS, a dynamic calibration algorithm that optimises sampling rate and trigger offset selection. SR\&OS leverages adaptive binary search and statistical response detection to capture meaningful system responses despite variable latencies and noise conditions.
The thesis also performs a detailed risk assessment of typical noise sources in side-channel measurements and ranks mitigation strategies based on their effectiveness and practical feasibility. It identifies practical denoising techniques, such as trace averaging, singular spectrum analysis, and independent component analysis, as effective methods for improving signal quality. Furthermore, it evaluates signal quality metrics and validates comparative power and correlation-based indicators as efficient predictors for adaptive acquisition termination.
Together, these developments create a more robust and scalable framework for detecting undocumented behaviours in embedded systems through powertrace analysis. Experimental validation using synthetic datasets and real-world embedded targets demonstrates improvements in calibration accuracy and acquisition efficiency. The findings lay a foundation for future advancements in hardware fuzzing frameworks, mainly targeting embedded environments
Long-distance Travel in Canada: Multimodal Modeling with a National Network
No full textLong‑distance (LD) travel comprises a disproportionately large share of total passenger-kilometers despite representing a small fraction of trip counts. Yet LD travel remains underexamined in Canada’s vast geographic context. This thesis develops and applies a comprehensive modeling framework to analyze LD trip generation and mode choice for Canadian residents, leveraging data from Statistics Canada’s National Travel Survey (NTS) (January 2018–February 2020) and a new national multimodal transportation network construct for this thesis. The network integrates geospatial centroids for Census Subdivisions with travel-time estimates for automobile, air, intercity rail, and bus modes.
Trip generation was examined through both disaggregate (person‑level hurdle and zero‑inflated count models) and aggregate (origin‑destination zone‑pair hurdle models) approaches, incorporating socioeconomic variables (age, income, gender), trip attributes (distance, season), and accessibility measures. Results indicate that accessibility, rather than traditional demographics, may be an important variable in predicting whether a LD trip occurs: with lower local accessibility and greater distance to airports increasing the likelihood of at least one trip in the given month. However, once the trip “hurdle” is crossed, trip counts are less sensitive to accessibility, underscoring behavioral impacts. Even with the very large dataset, models are very weak suggesting that travel surveys are a weak method for understanding LD travel.
Mode choice was analyzed using a Multinomial Logit (MNL) model alongside Machine Learning (ML) classifiers (Decision Trees, Random Forests, Support Vector Machines, Neural Networks). While MNL yields interpretable elasticities, with intercepts confirming preference for the driving mode and positive income effects for air travel, ML methods achieve superior predictive power. Feature importance from Random Forests highlights travel time (especially driving) as the dominant determinant, followed by accessibility, with sociodemographic and seasonal factors playing secondary roles. Mode choice models with alternative specific travel times are viable with publicly available data and these results support the need to seriously consider use of ML in LD mode choice even though understanding the influence of individual behavioral factors becomes more limited.
Long-distance passenger travel demands models are not typically available in Canada despite their utility for infrastructure, service and environmental planning. This thesis research demonstrates models are viable with existing publicly available data
Variable Selection and Prediction for Multistate Processes under Complex Observation Schemes
No full textThis thesis addresses variable selection and prediction in time-to-event analysis under complex observation schemes that commonly arise in biomedical studies. Such schemes may lead to right-censored data, interval-censored event times, or dual-censoring scenarios. Across three main chapters, we develop variable selection methods for multistate processes, address challenges arising from incomplete data under complex observation schemes, and investigate the implications of model misspecification, such as using simpler models in place of multistate models, and the potential risks of violating assumptions on covariate effects estimation and predictive performance.
We begin with considering the problem of variable selection for progressive multistate processes under intermittent observation in Chapter 2. This study is motivated by the need to identify which among a large list of candidate markers play a role in the progression of joint damage in psoriatic arthritis (PsA) patients. We adopted a penalized log-likelihood approach and developed an innovative Expectation-Maximization (EM) algorithm such that the maximization step can exploit existing software for penalized Poisson regression thereby enabling flexible use of common penalty functions. Simulation studies show good performance in identifying important markers with different penalty functions. We applied the algorithm in the motivating application involving a cohort of patients with psoriatic arthritis with repeated assessments of joint damage, and identified human leukocyte antigen (HLA) markers which are associated with disease progression, among a large group of candidate markers.
Chapter 3 extends this algorithm to more general multistate processes, and to more complex observation schemes. We consider the classical illness-death model which offers a useful framework for studying the progression of chronic disease while jointly modeling death. The exact time of disease progression is not observed directly but progression status is recorded at intermittent assessment times; the time to death is subject to right-censoring. This creates a dual observation scheme where progression times are interval-censored and survival times are subject to right censoring.
A penalized observed data likelihood approach is proposed which allows for separate penalties across different intensity functions. An EM algorithm is again developed to facilitate use of different penalties for variable selection on disease progression and death through penalized Poisson regression. This adaptation retains the flexibility to exploit existing software with commonly used penalty functions. Simulation studies show good finite-sample performance in variable selection with different combination of penalty functions. We also explored how various aspects of the variable selection algorithm affect performance such as use of nonparametric baseline intensities and different ways to select the optimal tuning parameter(s). An application to data from the National Alzheimer’s Coordinating Center (NACC) demonstrates the use of our method in jointly modeling dementia progression and mortality.
Chapter 4 builds on insights from Chapters 2 and 3 by investigating how simpler marginal methods targeting entry time to the absorbing state (e.g., a Cox proportional hazards model) compared to full multistate models. Here we retain use of the illness-death process as the basis of the investigation, but consider settings where transition times are only right-censored.
We first study the limiting values of regression estimators from a Cox proportional hazards model when the data generating process is based on a Markov illness-death model. The potential impact of modeling the multistate processes based on a misspecified model is also investigated by considering cases where a) important covariates are omitted, or b) the Markov assumption is violated. We then examine the implications of model misspecification when the goal is prediction - this is done by evaluating the predictive performance of a misspecified Cox regression model for overall survival and a misspecified Fine-Gray model for disease progression, and comparing their respective predictive performance against that of the true illness-death model. We find that the limiting value of regression coefficients estimators obtained from Cox models and Fine-Gray models depend on several factors, including the baseline hazard ratio of death between the intermediate and initial states, the probability of moving through the intermediate state, and covariate effects on all transitions. However, the corresponding predictive accuracy is not substantially compromised despite biases in the regression coefficient estimators in most scenarios we investigated. The limiting value of regression coefficients obtained from a Markov illness-death model and the corresponding predictive accuracy are sensitive to model misspecification such as omitting important covariates and violation of the Markov assumption. The practical implications are illustrated using a dataset of patients with metastatic breast cancer in the control arm to predict overall survival and fracture risk.
Chapter 5 reviews the contributions of this thesis and discusses problems warranting future research
Nonconvex Trajectory Optimization using Trajectory Sensitivities: Application to Personalized Autonomous Driving
No full textAutonomous Driving (AD) has been studied in the past decade and has been gradually deployed in everyday life. A key factor in increasing people’s level of acceptance is trust, which may be enhanced by personalized autonomous driving. One way to design personalized autonomous vehicles is by mimicking the driver’s own driving style while driving safely. Many existing works explore learning-based approaches to achieve this goal. However, the performance of these methods is highly dependent on sample efficiency, and it is usually difficult to enforce safety guarantees. To mitigate these difficulties, this thesis proposes an autonomous vehicle control framework in the form of a parameterized nonconvex trajectory optimization problem with a bilevel structure, where the upper-level models the driving style of a target driver and the lower-level performs vehicle motion planning. Therefore, the focus of this work is the formulation of this parameterized nonconvex trajectory optimization problem and its solution methods, discussed under an application scenario of personalized autonomous vehicles.
The lower-level of the bilevel programming problem solves a trajectory optimization problem. The nonlinear dynamic of the vehicle model leads to challenging nonconvex trajectory optimization problems. Many existing approaches formulate them as multistage programs and rely on derivatives of each stage to obtain a local approximation at each iteration, in which case the quality of approximation when solving the optimization program has significant impact on convergence behavior.
In this work, we develop a novel approach for obtaining improved local approximations when solving nonconvex trajectory optimization problems. By performing an input-to-state reformulation of system dynamics, we use trajectory sensitivities, which are derivatives of the entire system trajectory with respect to control inputs, to form local approximations. This novel approximation method, when used to solve optimization problem and to linearize the constraints, results in less approximation error than the traditional approach, while the latter has accumulating numerical errors for multi-stage planning problems. Local convergence guarantees for the proposed method are presented for nonconvex optimization problems with input-affine inequality constraints. The method is applied to generate trajectories for an autonomous vehicle that are not dynamically feasible and is extended to include a scenario with static obstacles that introduces nonconvex constraints.
The upper-level of the bilevel programming problem models the driving style of the target driver by minimizing the difference between human driving data and motion planning results. The decision variables are weight factors that characterize the driving style and are used to parameterize the lower-level objective, hence affecting its planning results. We adopt a gradient-based approach to solve this problem. However, differentiability is not guaranteed given the bilevel structure and the nonconvex lower-level solution mapping, so we use subgradient "descent" to generalize gradient descent for non-differentiable functions. The quotation marks suggest the fact that subgradient methods are not necessarily monotone. Therefore, a projected subgradient update algorithm is adopted to solve the upper-level problem.
When learning-based approaches may fail in rare or unseen scenarios, our proposed method with an embedded vehicle model will continue to work. In addition, the optimization framework with dynamical and safety constraints ensures driving safety. The lower-level motion planner has been simulated on a variety of reference paths and compared with the traditional sequential quadratic programming with conventional first-order Taylor approximation to outperform in approximation accuracy, allowable trust-region radius, iterations to converge, and total solver time. Furthermore, out method is less prone to failure when handling multiple obstacles with a complex reference. The upper-level problem is also simulated to solve tracking problems and obstacle avoidance problems, demonstrating its ability to mimic the driving style of a target driver
Exploring User Interface Constraints for Reading and Writing
No full textConstraints are fundamental to human-centred design. Although by definition, constraints "limit" or "restrict" the capability of software, when designed correctly, they can have enabling characteristics as well. I sought to understand how user interface constraints can positively affect user outcomes, in ways that go beyond traditional goals of error-proofing.
Drawing from different theories in psychology, this dissertation presents four projects that study different types of user interface constraints while reading and writing. First, in a passive reading context, I evaluate the effects of two commonly-used document navigation techniques on reading comprehension: scrolling, where the reader has complete control over the viewport position; and pagination, where the viewport is restricted to specific locations. Second, in an active reading context, I propose limiting the number of words that can be highlighted in document reader software to improve reading comprehension scores. Third, when writing with large language models, I propose requiring users to write longer prompts so they feel more psychological ownership towards the generated output, and design interaction techniques to nudge users to write longer prompts. Finally, when prompting large language models to learn more about documents, I propose restricting where prompting can occur by requiring it to be anchored to specific text in the document and design new commenting techniques with different requirements that must be satisfied to finalize the comment.
Overall, this dissertation demonstrates the potential of user interfaces and interaction techniques that purposely constrain users. Through several controlled experiments, my findings suggest that user interface constraints may not be effective for certain activities, like when passively reading, but when actively reading and writing with large language models, they can often encourage positive user outcomes, like improved reading comprehension and psychological ownership. However, designing interaction techniques that leverage 'soft' constraints is challenging, and such interaction techniques do not always nudge users. Together, this dissertation contributes knowledge on the effectiveness of non-error-proofing constraints when reading and writing, and interaction techniques that can be integrated into current reading and writing interfaces
Influences of Source Waters on Alpine River Systems in Tongait KakKasuangita SilakKijapvinga (Torngat Mountains National Park), Nunatsiavut, Labrador
No full textTongait KakKasuangita SilakKijapvinga (Torngat Mountains National Park) encompasses the northern tip of Labrador and is situated at the southernmost limit of the Arctic Cordillera. This region is an integral part of the homeland for Inuit from Nunatsiavut and Nunavik and hosts the only remaining glaciers in continental northeastern North America. Like other high-latitude regions, glacial melt is currently a key source of streamflow in the summer months and provides refugia for cold-water species. Continued climate warming is expected to make streamflow warmer, slower, and less turbid, putting stream function and culturally significant species such as ikKaluk (Arctic char) at risk. Future glacial loss is also expected to transition downstream habitats to resemble those of more barren non-glacial fed watersheds, further affecting ecosystem services, connected habitats, and community resources. This research aims to examine the impacts of cryospheric (ice) and hydrological (water) systems on ecohydrology by exploring glacial and late-lying snow influences on stream composition and riverine habitats. Stream composition and ecological function is examined through a stable isotopic analysis of oxygen and hydrogen and measurements of aqueous dissolved organic and inorganic carbon from stream water samples collected from three watersheds with different dominant water sources: glacial meltwater, snowmelt, and rain and groundwater. As this area has not undergone water sampling in the past this work also established a baseline for future research. With continued climate warming projected to shift the contribution of water sources it is important to know the unique influence each of these have to better predict how upstream and downstream systems will respond to continued change. This will further our understanding of how changes in the cryosphere and hydrosphere impact northern ecosystems and human livelihoods in support of future environmental monitoring, adaptation, and conservation
Data-Driven Decision-Making Under Uncertainty: An Empirical Study of U.S. Wildfire Management
No full textWildfire management in the United States faces prediction accuracy, cost efficiency, and fiscal sustainability issues. This dissertation integrates three interrelated research topics to develop integrated decision models applicable to each stage of wildfire management. The first study evaluates the role of social media analytics (SMA) and Web 3.0 technologies towards improving wildfire prediction, real-time tracking, and response decisions. The study reviewed current social media analytics tools for crisis response, showing how they support crisis tracking, response timing, and crisis communication. The same functionality can presumably be applied to wildfire management. The second study introduces a temporal gravity model that links population- and location-weighted social media activity to wildfire response costs per acre. The model captures behavioral visibility prior to operational deployment and demonstrates stronger informational value than tweet volume alone. The third study investigates how federal budget changes relate to the accuracy of state preparedness decisions. Higher funding is associated with improved accuracy in the short term, but this association weakens in later budget cycles. The analysis treats federal budgets as exogenous inputs and uses panel methods with robustness checks to evaluate decision dynamics under fixed fiscal constraints. Across all three essays, the dissertation highlights the importance of integrating behavioral data and fiscal signals to better inform wildfire planning. It provides empirical evidence that public attention, budget expectations, and institutional coordination jointly influence the quality of response decisions. These findings suggest that effective wildfire management requires models that account for informational uncertainty, fragmented authority, and the timing structure of operational and fiscal systems.
Keywords: Wildfire Management, Decision Science, Behavioral Operations Management, Crisis Informatics, Public Finance, Panel Data Analysis, Gravity Model, Time Series Analysis
Understanding Uncertainty in Daily Life: Appraisals and Metacognitive Strategies
No full textUncertainty is an inevitable part of our lives, yet little is known about how people navigate the uncertainty they encounter in their lives. In my thesis, I examine how uncertainty is perceived, which situations are perceived as uncertain, and how people react to uncertainty in daily life, specifically whether they engage in perspectival metacognition. To address these questions, I used data from a year-long longitudinal study asking participants (N = 499) to report on the most significant event of their day. Using natural language processing, I then classified these open-ended text responses as uncertain and not uncertain, and examined how participants’ construal, emotional profile, and reasoning differ for uncertain compared to not uncertain events. Uncertain events were perceived as relatively more negative, challenging, in others’ control, and less predictable. Uncertain events were also associated with greater negative emotion and less positive emotion. Negative event types (i.e., conflict, rejection, annoying, and sad or bad news type of events) were more likely to be classified as uncertain compared to positive or neutral event types. Participants were also relatively more likely to report intellectual humility and a search for compromise in reflections on uncertain events. These results were similar for the trait and state levels. I discuss the implications of this scholarship for research on affect and emotion regulation and on mental health, specifically for those with anxiety disorders
Engineering Solid Oxide CO2 Electrolysis: From Nanoparticle-Decorated Perovskite Cathode to System-Level Modeling
No full textSolid oxide electrolysis cell (SOEC) is a promising technology for CO2 electrolysis and subsequent conversion to useful chemicals. This thesis combines the experimental development of new cathode materials with system-level simulation to enhance the performance of SOECs for CO2 electrolysis and assess their applicability for fuel production. There are two components to the work: (1) proposing nanoparticle decorated perovskite cathode material and (2) integration of DAC, SOEC and synfuel production and asses its performance with techno-economic and environmental analysis.
In the experimental section, the focus was on the cathode material of the SOEC since it is the limiting factor for CO2 electrolysis. Sr2Fe1.5Mo0.5O6-δ (SFM) has attracted much attention due to its decent performance of CO2 electrolysis. To enhance the SFM performance, it was modified by doping bismuth and nickel to make a new composition of Bi0.1Sr1.9Fe1.4Ni0.1Mo0.5O6-δ (BiSFNiM). The Ni-doping made it possible for Fe–Ni nanoparticles to exsolve in situ when the material was reduced by 5% H2/Ar. Structural characterizations like XRD and Rietveld refinement showed that, during exsolution, the material changed from a pure double perovskite structure to a mixed-phase material with both Ruddlesden–Popper (RP) and residual double perovskite phases and metallic nanoparticles. Using electron microscopy (SEM/TEM/EDS), it showed that Ni migrated to the surface of the perovskite bulk where it forms Fe–Ni nanoparticles. This material, then, was used as the cathode of SOEC and the results showed that these exsolved Fe–Ni nanoparticles significantly improved the electrocatalytic activity for the CO2 reduction reaction (CO2RR). Electrochemical performance tests demonstrated substantial improvements in current density and polarization resistance. The fabricated cell achieved a peak current density of 1.3 A/cm² at 800 °C under an applied voltage of 1.6 V, while it was 1.0 A/cm² for the non-exsolved nanoparticles sample.
The second half of this thesis was a process simulation and system-level evaluation of an integrated DAC-SOEC facility. The technology was based on capturing 250,000 tonnes of CO2 from the air each year and turning it into either methanol or synthesis fuel through downstream processes. Methanol production was 36.4 tonnes per hour, while synfuel output was 15.1 tonnes per hour. The techno-economic analysis found that the levelized production cost for methanol was 2.78 per kilogram (approximately 45% more than normal expenses). Using Ontario’s electricity grid mix, the simulated plant achieved greenhouse gas emissions of 31.1 gCO2-eq/MJ for methanol and 5.2 gCO2-eq/MJ for synfuel, the latter representing a reduction compared to conventional fossil-based pathways (40 g-CO2-eq/MJ-MeOH and 29 g-CO2-eq/MJ-synfuel). Further sensitivity analysis demonstrated that switching to fully renewable electricity sources, such as hydropower or wind, could push the synfuel production case into a net-negative emissions region.
In conclusion, this thesis contributes to both fundamental and applied aspects of CO2 electrolysis. On the material side, it offers a strong plan for boosting cathode performance by co-doping and nanoparticle exsolution. It also gives information about phase stability, exsolution behavior, and catalytic activity. At the system level, it shows that combining DAC and SOEC for sustainable fuel production is possible from a technological, economic, and environmental point of view. The dual approach shows how innovative materials and systems design can work together to help us toward carbon-neutral chemical manufacture