39339 research outputs found
Sort by
Enhancing Legal Text Entailment: Evaluating Model Architectures, Training Approaches, and Interpretability
No full textThe legal domain is a challenge for Artificial Intelligence systems as it is characterized by its complex vocabulary, intricate reasoning, and consistency with precedents. With increasing digitization, the potential for Artificial Intelligence to become a useful tool for legal projects has grown significantly. However, adoption within the legal field lags behind other industries due to cultural resistance, limited high-quality training data, and the need for interpretability in black-box systems. To contribute to the advancement of the role of AI in the legal domain, we developed and evaluated a legal entailment classification system which determines whether a paragraph from an existing legal case supports the decision in a new case, while also providing a justification for the classification.
Leveraging advanced Natural Language Processing techniques and Explainable AI methodologies, this work integrates domain-specific pretraining, lightweight adaptation techniques such as LoRA, and an ensemble technique. A dataset of over 35,000 Canadian legal cases was used for further pretraining, while fine-tuning and evaluation were performed using the COLIEE 2023 competition dataset. Our experiments highlight the trade-offs between computational efficiency and performance, and evaluate the impact of domain-specific pretraining on smaller transformer models such as RoBERTa, compared to adaptations of larger language models for the classification task.
In addition to classification, this thesis explores the role of explainablility techniques for Artificial Intelligence in legal applications by implementing the techniques of LIME and model-generated justifications. These methods were assessed using human evaluations, as well as automated sufficiency metrics, with highest scores in the human evaluation being 82.14% for adequacy, 92.86% for understandability, and 85.71% for trustworthiness, and a peak score of 99.17% for the automated sufficiency metric.
The contributions of this research include the creation of domain-specific pretrained models, a comparative evaluation of fine-tuning and lightweight adaptation techniques for large language models, and a systematic exploration of explainability methods to improve interpretability and user trust. To the best of our knowledge, this study is the first within the Canadian legal AI context to investigate the effects of further pretraining on both small and large language models, as well as to integrate language model adaptation and explainability into a unified system for legal text entailment classification
Algorithms for Generating Cover-Free Families
No full textCover-free families (CFFs) are a combinatorial design used in many applications, including group testing and cryptography such as encryption and signature schemes. A -cover-free family, denoted -CFF(,), is a set system where the underlying set has elements, the set system has subsets, and no subset is contained in the union of any other subsets. When cover-free families are used in applications, it is generally desirable to maximize the number of subsets for a given set of underlying elements. These subsets correspond to samples in group testing, allowing more efficient testing schemes.
Currently, there are no publicly available tables of cover-free families that show the best-known cover-free family for specific parameters. This creates a challenge when implementing applications using cover-free families, since there are a variety of sources and constructions for cover-free families that would be useful. Tables for other types of combinatorial designs are publicly available, such as covering arrays, and these tables provide useful information to researchers.
In this thesis, we outline a selection of constructions of cover-free families, then create and implement algorithms to create tables of best-known cover-free families from these selected constructions. Our selection of direct constructions includes constructions from Sperner systems, packing designs, Reed-Solomon codes, and linear error correcting codes meeting the Gilbert-Varshamov bound constructed using the method of conditional probabilities. Our selection of recursive constructions include the Kronecker product, the Optimized Kronecker product, an additive construction, a doubling construction for , and an extension-by-one construction, which create larger cover-free families from smaller ones.
Using these selected constructions, we iterate over combinations of parameters for each construction to create tables of best-known cover-free families from these constructions. Furthermore, we design and implement a recursive algorithm to construct any chosen cover-free family in our tables. Our results include tables of cover-free families for up to trillion, and up to 25. The maximum appearing in any of our tables is for
Transition in care interventions for Refugee, Immigrant and other Migrant (RIM) populations: a health equity-oriented scoping review
No full textAbstract Background Transition in care involves the transfer of responsibility for aspects of patient and public health care among providers, institutions, and health and social sectors. Indeed, health systems increasingly require individuals to interact with a number of providers, in a number of health settings, and across multiple points of time. Refugees, immigrants, and migrant (RIM) individuals face several precarious transitions, language and cultural barriers, and unfamiliarity with public health systems, which may result in health inequities. A greater understanding of the interventions that facilitate effective transitions in care for RIM populations is needed to improve health outcomes in this vulnerable group. Methods This health equity-oriented scoping review aimed to report the characteristics of Transition in Care (TiC) interventions for RIM populations and identify which equity-relevant characteristics of RIM populations were targeted by these interventions. We searched MEDLINE, Embase, and Scopus for eligible studies published in English from the year 2000 onward. Two independent reviewers screened search records and extracted relevant data from included studies. We used a public health and health equity lens to identify the social determinants of health that were addressed by TiC interventions. Results Our systematic search identified a total of 42 studies, evaluating the impact of 38 unique interventions or public health programs. The delivery of interventions involved various healthcare sectors and professionals. Additionally, some programs enlisted non-medical personnel to provide health-related education and support. The most promising programs for health outcomes involved health navigation or providing public health education for RIM populations. The most common equity-relevant characteristics considered in these studies were language, cultural background, and education level. Conclusion This novel scoping review reveals a diverse range of public health interventions that are being implemented to improve national and international transitions in care for RIM populations, with the most promising from healthcare navigation and health education. Future research should target transitions to digital health technologies, public health, hospital-to-home, and pediatric to adult care gaps to ensure smoother transitions in care for equity-deserving populations navigating new healthcare systems
Refusing State Injustice: The Politics of Intentional Noncitizenship
No full textThis dissertation is about those who resist – who refuse – state injustice through citizenship, and who exclude themselves from the state and embrace associated risks of harm including statelessness. It explores the conditions that inform intentional exit and the conceptual and practical implications that follow. Recognising that liberal practices and discourses of citizenship are problematic, this project asks what are the grounds used by intentional noncitizens to justify their choices to exit the state? What does it mean to be an intentional noncitizen? Are there practical governance proposals that can respond to the normative claims made by those who intentionally exit? Despite what most residents of liberal states would claim – that intentional noncitizenship should be impermissible – the leading claim of this dissertation is that leaving the state can be justified.
This conclusion is informed by the exploration of three instances of intentional noncitizenship. The Freedom Babies movement refuses colonialism by not registering the births of their children, leaving their children at risk of statelessness. Anti-statists refuse imposed citizenship and arbitrary allegiance through renouncing their citizenship. The anti-authority movement refuses the globalised administrative state through disengagement from expectations of citizenship such as law and the legal system, public processes and symbols, and identification practices. To explore tensions between traditional preconceptions of the state and emergent plural conceptions of the good this project utilises analytical political theory as a methodology, specifically the reflective equilibrium approach. Accordingly, to engage with seemingly disparate groups of intentional noncitizens, this project relies on a conceptual typology of refusal – a form of politics distinct from resistance, conscientious objection, and civil disobedience – that embodies the key features of action, future, and relationality.
This project finds that those who refuse citizenship are aligned in their critiques of normative justifications for sovereign statehood and the goods it is said to provide, specifically security, freedom, and citizenship itself. Citizenship is a tool of refusal used specifically to exit the state: this refusal is neither absolute nor consistent but dynamic in that it responds to and negotiates its political circumstances. These findings establish a foundation from which a theory of intentional noncitizenship emerges. Intentional noncitizenship can be considered an instrumental good as it embodies the distinct features of autonomy, political action, and obligation to others, and is a viable alternative to citizenship as it does not cause ontological harm, and if sufficiently governed through legal residency status may not cause material harm. As intentional noncitizens must inevitably live within some state, this project proposes a governance mechanism in the form of an individual right to self-determination which would facilitate intentional noncitizen rights and responsibilities. By pushing the boundaries of what statelessness means and for whom this project destabilises the narrative that statelessness is an absolute harm
Targeting Viral Epigenetics for the Control of Human Adenovirus Replication
No full textHuman adenovirus (HAdV) causes minor illness in healthy individuals, but can cause severe disease in at-risk individuals, such as pediatric, geriatric, and especially immunocompromised individuals. This is problematic, as no approved therapeutic exists for treating HAdV infection, worsening the burden these individuals face. During infection, HAdV exploits epigenetic regulation of the host cell by modulating chromatin dynamics, promoting expression of cellular genes productive to infection. Epigenetic regulation of viral DNA is also important, as HAdV DNA is chromatinized during infection. As such, we hypothesized that treatment with compounds capable of regulating viral epigenetics during infection would be an effective strategy in combating HAdV infection. Though treatment with curcumin and CBL0137 have diverse effects within the cell, these compounds are each capable of affecting epigenetic processes, by modulating the expression of epigenetic machinery as well as inhibiting histone chaperones. Thus, we aimed to evaluate the effect of curcumin and CBL0137 treatment, as well as siRNA-mediated depletion of the histone chaperone facilitates chromatin transactions (FACT) complex on HAdV infection. Treatment with curcumin inhibited expression of the viral early 1A (E1A) proteins as well as the late capsid protein hexon. E1A is the first region expressed during infection, and the E1A proteins are vital for modulating cellular gene expression and transactivating other viral genes, and the lack of E1A proteins following curcumin treatment subsequently impacted viral DNA replication and progeny formation. CBL0137 treatment also prevented E1A protein production, which was attributed to CBL0137-induced degradation of the catalytic subunit of RNA polymerase II. However, we noted that other early viral genes contribute to the stabilization of RNA polymerase II during CBL0137 treatment. Finally, the role of the FACT complex in HAdV infection was explored. Depletion of the FACT complex reduced E1A production by lowering the amount of E1A transcripts within the cell. The FACT complex was also observed co-localizing with viral replication centers, with the FACT complex associating with the viral DNA-binding protein, a key protein in viral replication centers. Taken together, these results indicate targeting E1A production by interfering with viral chromatin is a viable strategy in treating HAdV infection
Towards Compostable Pressure Sensitive Adhesives
No full textPolymers play a crucial role in modern society, yet their widespread use has led to environmental challenges, particularly due to their persistence and reliance on petroleum-based feedstocks. The transition toward sustainable polymer reaction engineering requires innovations in polymer synthesis, including water-based polymerization, bio-based monomers, and degradable structures that enable responsible end-of-life disposal. Among polymeric materials, pressure-sensitive adhesives (PSAs) are widely used in packaging, medical, and industrial applications, but conventional PSAs are synthesized via solution polymerization using petroleum-derived monomers, resulting in non-biodegradable materials that contribute to long-term waste accumulation.
This study develops sustainable, compostable PSAs via emulsion polymerization, incorporating bio-based monomers and renewable nanomaterials to achieve a balance between adhesive performance and environmental degradability. A key innovation is the integration of 2-methylene-1,3-dioxepane (MDO) into butyl acrylate (BA)/vinyl acetate (VAc) terpolymers, introducing hydrolyzable ester bonds to enhance degradability. The reactivity ratios of MDO, BA, and VAc were estimated using the Error-in-Variables Model (EVM) to provide crucial insights into monomer distribution and polymer structure. A major challenge in emulsion polymerization was MDO’s hydrolysis sensitivity and ring retention, which was mitigated through optimized reaction conditions, including pH control (7.8–8.8) and reaction temperatures (40–50°C). These optimizations minimized ring retention, ensuring the effective incorporation of degradable linkages.
To further enhance PSA performance, carboxylated cellulose nanocrystals (cCNCs) were incorporated via post-polymerization blending, reinforcing the polymer matrix and simultaneously improving tack, peel strength, and shear adhesion without compromising sustainability and biodegradability. The biodegradability of these formulations was evaluated under controlled composting conditions following ASTM D5338, using a lab-scale composting setup. Among the tested formulations, BMV10-NEW-cCNC (containing 10 wt% MDO and cCNC) exhibited the highest degradation rate. The elevated polymerization temperature (50°C) led to nearly complete MDO ring opening, as confirmed by 13C-NMR, with CO₂ evolution indicating 12.49 wt% biodegradation over 60 days. While these findings demonstrate the potential of MDO-based PSAs for compostable adhesive applications, further improvements in MDO content and polymerization strategies are needed to achieve complete degradation under industrial composting conditions.
This study highlights the potential of integrating bio-based polymer chemistry, nanomaterials, and controlled water-based polymerization techniques to develop high-performance, compostable PSAs, advancing the field of sustainable polymer reaction engineering and contributing to a circular economy
A Multimodal Approach to Restoring Motor Function After Spinal Cord Injury: Exploring the Use Of Plant-Based Biomaterials & Stimulating Propriospinal Interneurons
No full textSpinal cord injury (SCI) presents a significant challenge in regenerative medicine due to its complex pathophysiology & extensive molecular barriers that hinder repair. Overcoming these challenges requires interdisciplinary strategies that combine molecular biology, bioengineering, neuroscience, and rehabilitative therapy. In particular, the neural tissue engineering approach aims to produce an implantable scaffold that can support regeneration of spinal cord tissue in the hopes of forming relay circuits or promoting the regrowth of native axons. These biomaterials can be engineered to create a favorable environment that facilitates axonal infiltration into the damaged regions of the central nervous system. This thesis explores the use of decellularized plant tissue as a biomaterial for 3D cell culture & neural tissue engineering. We first demonstrate the ability of plant-derived scaffolds to support proliferation of neural stem cells in vitro and guide their differentiation into the neuronal lineage. Subsequently, we assess the scaffold's potential to support motor recovery in a rat model of complete spinal cord injury. Our findings highlight a potential therapeutic application for plant-based biomaterials and show their ability to be functionalized with peptide coatings. Finally, we identify dI3 neurons as key contributors to the re-activation of spinal locomotor circuits in response to cutaneous sensory input below the injury. Together, these results highlight the promise of multimodal therapeutic strategies for motor recovery after spinal cord injury
Examining the Permeation of Toxic Chemicals Through Barrier Materials
No full textThe wide usage and availability of hazardous substances in many facets of todays life necessitate people must be in close proximity to, and risk contact with, hazardous chemicals during the conduct of their work. One layer of risk mitigation for exposure to hazardous substances is the use of protective equipment. In the case of hazardous chemicals, this is often a polymer layer worn to balance protection with mobility. However, modelling of the diffusion of chemicals through rubbery polymers is difficult and common test methods may contain sources of error inherent in the experimental design. Through numerical simulations, this thesis examines the accuracy of current standard test methods, seeks to determine which factors have the most impact on errors and suggests mitigations. Experiments were conducted to examine the permeation and determine transfer properties using the time-lag method. This work suggests that some common testing protocols have sources of error inherent to the design and experimental conditions must be carefully considered when selecting test protocols
Familiar Faces: Evaluating Outcomes of a Community Mental Health Program for Frequent Emergency Department Users
No full textThis dissertation is composed of three studies that evaluated outcomes of a novel community-based mental health care program providing services to frequent emergency department users presenting with mental illness or addiction in Ottawa, Ontario, Canada. The publicly-funded program, known as Familiar Faces, provided a stepped-care intervention to clients, including system navigation and intensive case management. The program was designed to support client health and social needs while alleviating resource burden on hospital emergency departments.
Study 1 used a pre-test/post-test design to prospectively assess program outcomes. Quantitative data were obtained from self-report measures of 63 program clients regarding their experiences of psychosocial functioning, mental illness symptoms, and addiction symptoms, at baseline program intake and follow-up at least six months later. Results indicated improved client outcomes over time related to overall functioning and severity of anxious and depressive symptoms.
Study 2 used a qualitative design and pragmatic thematic analysis to retrospectively assess program outcomes. Data were obtained from interviews with 15 program clients and focus groups with six program case managers regarding their perceptions of program outcomes, mechanisms of change, and barriers to change. Results indicated perceptions of improved client outcomes related to quality of life and severity of mental illness symptoms. Perceived mechanisms of change included the importance of fostering positive working relationships between program clients and case managers, as well as the program's focus on supporting clients to develop practical skills.
Study 3 used a pre-test/post-test, nonequivalent groups design to retrospectively assess program outcomes. Quantitative data were obtained from longitudinal population-level public health care records for 278 program clients and three matched comparison groups regarding their frequency of emergency department visits, hospital admissions, days in hospital, general practice outpatient visits, and psychiatric outpatient visits, in the one year before and two years after program intake. Results indicated that both program clients and comparison participants improved on outcomes over time, including reduced frequency of emergency department visits and hospital admissions. No interaction effects were found on outcomes between groups over time, other than for psychiatric outpatient visits, with program clients experiencing greater increases in psychiatric outpatient visit frequency over time than comparison participants.
These dissertation studies employed varied research designs, different methods, multiple data sources, and diverse participant pools, enabling a comprehensive and rigorous evaluation of program outcomes. The dissertation findings contribute empirical evidence and practical insights to inform mental health care program and policy development for frequent emergency department users with mental illness or addiction and the health care systems that support them
Exploring How Well Llama3 can Generate State Machines Represented in Umple
No full textModelling a system is an important part of design which can be time consuming and difficult. A common type of model is a state machine, describing a system or component's behaviour. Multiple languages have been created to make this process smoother, one of them being Umple, which enables describing state machines both textually and graphically, as well as embedding them in multiple programming languages and generating code from them. Although tools such as Umple have made the process easier, developers or business analysts still have to translate requirements into state machines. In this thesis, we investigate how well this step can be automated with the application of artificial intelligence. We show that using modern large language models, Llama 3 in our case, we can allow a user to generate a state machine by only providing a short description of the system requirements. These state machines generated by large language models (LLMs) can be used as a model for the system as is if they meet the requirements or a base that can be improved on. We found that for simple systems using a large language model along with techniques such as retrieval augmented generation and multi-shot learning, can save users large amount of time compared to coding state machines from scratch