Carleton University Institutional Repository
Not a member yet
20816 research outputs found
Sort by
Beyond the Chat: Abstractive Summarization and POS Analysis of Developer Discourse on Discord
Get PDFThe research entails an analysis of a dataset comprising developer dialogues from Discord, which are summarized by multiple models, followed by a comparison of POS tags with human summaries. Our findings aim to measure the difference between machine-generated and human summaries in technical contexts, thereby providing insight into the implications for improving communication efficiency among software developers. This investigation contributes to the field of natural language processing by providing valuable insights for developers and engineers seeking to streamline knowledge assimilation from extensive technical discussions
On the Interpretability and Explainability of Prototype-Based Methods and Reinforcement Learning
No full textWith the ever-growing use of AI to solve real-world problems, the need for transparency and trust in these methods has given rise to Interpretable and Explainable AI. While a growing body of research is trying to address these issues, some areas of the field can still be further improved. In the field of inherently interpretable AI methods, the embedding spaces used for the current prototype-based neural network methods have issues with prototype-query similarity. In the same area of prototype-based neural networks, another issue is the inadequacies of current schemes for evaluating the interpretability of part-prototype networks. In addition, in the field of reinforcement learning, post-hoc explainability methods are focused more on policy distillation rather than on local feature-based explanations. In this work, an inherently interpretable method is proposed to create more interpretable prototypes in a prototype-based classification scheme. In addition, a robust human-centric evaluation framework is proposed for part-prototype networks. Another method is also proposed to create post-hoc explainability in reinforcement learning by utilizing state features. Experiments show the superiority and validity of these methods compared to the previous state of the art. In the case of the proposed evaluation metric, this work also includes a comprehensive comparison of the interpretability of existing part-prototype-based neural network methods
Redefining the Limestone City Protecting Heritage in Kingston, Ontario through Modern Development
No full textKingston is commonly called the Limestone City due to the urban fabric being dominated by limestone heritage buildings. Overtime, this urban fabric has been put at risk as the City changes priorities from protecting Kingston’s unique heritage towards rushed urban intensification. As homogeneous condo towers encroach deeper into the city and developers collaborate with the city to move development into the Historic Downtown, how can architects and conservationists adapt? With the conflict between developers and heritage organizations constantly raging in and out of the courtroom for the last 10 years, architecture finds itself at the forefront of Kingston more than ever. Architecture must find the meeting point between heritage and new developments, the old and the new, in order to redefine Kingston’s unique architectural identity
Assessing Behavioural Guidance Methods to Reduce Turbine Mortality and Improve Downstream Passage of Out-Migrating Anguillid Eels
No full textHydropower infrastructure poses a great risk to fishes. The goal herein was to study how behavioural guidance could improve downstream passage of anguillid eels. First, I undertook a review of potential behavioural guidance methods for eel and found that light and sound seem most promising, and that these methods could be particularly useful if used to augment physical guidance. Second, I took part in a field assessment of light as an eel guidance mechanism using a 216 m light guidance device. I analyzed eels’ vertical movement and various correlates of out-migration. In this chapter, I found that eels exposed to above-water light exhibited diving behaviour, and observed nocturnal migration coinciding with darkening lunar phases. This work consolidates evidence on eel sensory physiology and behavioural guidance, advances our understanding fishes’ responses to anthropogenic light, and contributes to conservation and management efforts for a valuable and vulnerable family of fishes
Developing Local Currency Bond Markets: Local Currency Lending's impact on developing Government Local Currency Bond Markets
No full textThe “original sin” describes when countries cannot access financing in their local currency, exposing countries to increasing debt servicing costs from their currency depreciating (Hausmann and Eichengreen 1999, 11). Local currency bond markets (LCBM) are thought to reduce exchange rate risk and increase financial stability. Multilateral development banks (MDB) have become increasingly interested in local currency lending (LCL), often with a secondary goal of helping grow LCBMs. The thesis investigates if MDB’s LCL can increase a country’s LCBM size. The thesis uses the synthetic control method (SCM) and case study analysis. The SCM is statistically insignificant, while the East Asia and Kenya case studies show MDBs can support developing LCBMs. However, significant growth comes from comprehensive domestic-led policies. The thesis concludes analyzing whether developing LCBMs can improve countries' financial stability, arguing that relative success is based on the size and ability of countries to leverage domestic demand for LCBMs
Essays on Environmental Economics
No full textIn Chapter 1, I report the results of evaluating the hourly impact of a behavioral intervention tested in a randomized controlled trial. Under the program, a randomly selected group of households in Alberta was provided visual information on their home heat loss. I find that the households conserve the same amount of electricity during peak and off-peak electricity demand hours, i.e. the intervention has failed to target peak times, and accounting for the intraday distribution of the electricity savings is not important when measuring the social benefits of the program. As a policy recommendation, the study suggests implementing retail electricity prices fluctuating within a day coupled with information feedback on households’ electricity usage. Chapter 2 assesses realized energy and air leakage changes in homes constructed before and after new building energy code adoptions in three Canadian provinces: Ontario, New Brunswick, and Alberta. We find no electricity or air leakage reductions attributable to more stringent code requirements, and there is no evidence that natural gas consumption declined after a code change. Instead, a generalized improvement in residential electricity consumption and air leakage rates is observable several years before new code adoptions, depending on the province. These preexisting trends in electricity usage and air leakage may point to changes in building industry practice preceding new building code adoptions, though further investigation is required to assess the drivers of these changes. The estimated energy savings are also not in line with ex-ante engineering projections. In Chapter 3, we use regression discontinuity design to examine the effects of the congestion pricing policy introduced on the San Francisco-Oakland Bay Bridge on July 1, 2010. The study finds that the new road toll, which led to a decline in rush hour traffic volume on the bridge, was associated with a moderate uptake in public transit ridership, but it did not affect traffic-related local air pollution and respiratory illness incidence in the bridge vicinity, in contrast with the past work on the topic in other settings. This points to the importance of considering the heterogeneous place-based factors that drive the welfare effects of environmental policy
Development and Validation of an Optimal Method for Early Release Decision-Making for the California Department of Corrections and Rehabilitation
No full textIn response to concerns about COVID-19 spreading through prisons, the California Department of Corrections and Rehabilitation (CDCR) implemented a series of proactive and reactive measures to reduce prison populations. The present research sought to learn from these measures and develop a decision-making tool to aid in future prison population reduction efforts. In Study One, I examined outcomes for individuals released from CDCR custody between April 1 and December 31, 2020 (N = 30,562) with follow-up at one and two-years. Rates of return were significantly lower in the pandemic (4.2 - 13.7%) than the pre-pandemic cohort (6.2 - 16.3%), but otherwise followed expected trends across demographic and offense type subgroups. This study also evaluated considerations used for early release. Results support early release for those within 180 days of sentence expiry compared to those with more time to serve (12.3% vs. 15.5). Individuals displayed significantly higher rates of return if they were released early with COVID-19 considerations (16.9% vs. 14.6%), even after controlling for risk. Study Two fills a need for a structured approach to make early release decisions in large-scale prison reduction efforts. Within CDCR, early release decisions are made for a subset of individuals by the Board of Parole Hearings, who uses two structured professional judgment tools to guide decision-making: the Comprehensive Risk Assessment and the Structured Decision-Making Framework. It would not be feasible to rate the entire CDCR custody population with these instruments, as they are labor intensive. I developed an automated tool designed to aid CDCR in release decision-making in future large-scale prison reduction efforts that incorporate information used in parole practice. I compared five methods for tool development, from traditional to machine learning, for optimized item selection and performance using a construction (N = 15,244) and validation sample (N = 15,318). The Reduction in Capacity Evaluation (ReduCE), discriminated well in predicting any return to CDCR custody at one and two years (AUC = .69 - .69) and return for new felony (AUC = .75 - .77), and calibrated well across race and gender. There was a lack of marked improvement when using machine learning methods
Representation of Atmospheric Gravity Wave Breaking and Saturation under the Anelastic Approximation
No full textIn this thesis, we present a mathematical model that represents internal gravity waves in the atmosphere using nonlinear partial differential equations and then use analytical and numerical methods. We consider a two dimensional, monochromatic, nonlinear, non-hydrostatic and viscous flow configuration with two types of approximations; Boussinesq and anelastic. In the real atmosphere, gravity wave amplitudes increase with altitude and this can lead to wave breaking. Most previous studies were based on the Boussinesq approximation in which the linear wave solutions do not increase with altitude. We consider a more realistic representation based on the anelastic approximation in which the wave amplitude increases with altitude. We carry out a weakly-nonlinear analysis to derive equations describing then time evolution of the wave-induced mean flow and potential temperature due to the nonlinear interactions. We carry out weakly and fully nonlinear numerical simulations of Boussinesq and anelastic equations. We also carry out numerical simulations to study the anelastic gravity wave propagation, growth, breaking, overturning and its impacts on the mean flow in the atmosphere. Results demonstrate that upward-propagating waves grow to sufficiently large amplitude and eventually break. Results demonstrate that the inclusion of viscosity and heat conduction allows us to investigate the long-term evolution of the gravity wave after wave breaking. It is found that the breaking of the wave field may occur due to two main mechanisms, namely, convective instability and dynamical instability. Convective and dynamical instabilities are examined by the local Richardson number such that convective instability corresponds to situations where the Ri ≤ 0 and dynamical instability corresponds to 0 < Ri < 0.25. The local Richardson number may drop below the stability criterion due to the negative potential temperature gradient, large shear of the velocity perturbation or a combination of both. In general, dynamical instability is more likely to occur at an earlier stage than convective instability. Convective instability is more likely to occur when there is very strong stratification
Transient Havens: Building Refugees Housing through the Notion of Villages
No full textThis study explores an innovative approach to transitional housing, designed to address the pressing needs of vulnerable populations in times of crisis. The proposal involves overnight relocation strategies, and providing temporary accommodation until more permanent solutions can be secured. Beyond providing mere shelter, the approach emphasizes the creation of inclusive communal spaces, fostering a dynamic environment for social interaction and mutual learning. The Transient Havens is particularly tailored to serve a diverse population, including elderly individuals seeking support, refugees requiring temporary dwellings, and professionals looking for more affordable housing solutions. By prioritizing the principle of adaptability within community-centred spaces, this prototype not only offers immediate relief but also empowers users to forge connections, share experiences, and acquire new skills. By melding aspects of shelter, community building, and skill-sharing, Transient Havens presents a holistic framework that transcends traditional notions of temporary housing
(Re)Integrating Rammed Earth Construction : Regulatory and Material Challenges and Opportunities of Rammed Earth Construction in Ontario
No full textRammed earth construction is a promising, bio-based alternative to conventional carbon-intensive building materials. Despite this, it faces many obstacles to mainstreaming, such as a lack of integration into the regulatory framework in Ontario, concerns with growing embodied carbon in modern rammed earth, and an overall lack of awareness, acceptance, and education. Based on a literature review of international codes and standards, material explorations, and, guided by interviews with industry leaders (builder, engineer, consultant, and policy analyst), this thesis attempts to contribute to the mainstreaming of rammed earth construction in Ontario through a series of ‘micro-interventions’ addressing gaps in the field identified in the interview process. They include building digital models, submitting a Code Change Request for the NBCC, proposing a Rammed Earth Standard for Ontario, and creating a Rammed Earth Student Guidebook