21793 research outputs found
Sort by
Gaby Says | Gaby Dit
No full textThis poster presents a study on deploying and evaluating a conversational agent using Retrieval-Augmented Generation (RAG) and Large Language Models (LLMs) in an academic setting. The objective was to implement a RAG-based system capable of answering reference questions and develop an evaluation protocol to measure the "usefulness" of the chatbot, comparing multiple models. The system follows a two-step approach: it first retrieves relevant documents from a curated knowledge base and then generates accurate, context-aware responses using LLMs. The evaluation protocol involved grading responses across five dimensions: accuracy, groundedness, elicitation, completeness, and further assistance. The results highlighted that while RAG significantly reduced hallucinations, challenges remained in preventing them completely. The evaluation rubric effectively differentiated between the performances of various models, despite the subjective nature of the grading process. This work emphasizes the integration of RAG and LLMs in academic reference services to provide responsible, reliable, and user-centered AI responses
Speech and language characterization for neurodegenerative diseases detection
No full textInvestigating voice abnormalities as potential biomarkers for neurodegenerative disorders may
present a cost-effective substitute for other neuro imaging techniques, bringing fresh perspectives
and enhancing diagnostic precision. The objective of this research was to pinpoint voice biomarkers
that can differentiate between healthy individuals and those suffering from dementia, as well as to
create a predictive model for Mini-Mental State Examination (MMSE) scores. Our analysis involved
studying voice recordings from 172 French speakers, who were divided into dementia, mild cognitive
impairment (MCI), and healthy control (HC) groups. By utilizing Praat and Matlab, we were able to
extract para-linguistic and linguistic speech features. Through various statistical analyses, including
ANCOVA, partial correlations, and stepwise linear regression, we were able to identify discourse
complexity, noun ratio, and average syllables per word as significant predictors of MMSE scores. Our
model, which incorporated these features, outperformed a baseline model that included covariates
such as sex, age, and education level. This demonstrated strong predictive capabilities, with an RMSE
of 3.65 and 3.7, and an R-squared of 0.35 and 0.31 on the training and test sets, respectively. This
study underscores the potential of linguistic speech features in the detection of cognitive impairment,
and emphasizes the importance of considering gender separately in the analysis to avoid bias and the
type of vocal task used in data collection
Understanding Test Code Quality from Perspective of Test Code Design and Maintenance
Get PDFSoftware testing is vital for ensuring software reliability and robustness. It involves executing a program and verifying it against developer-defined criteria to identify and fix deviations, aiming for fault-free software. Despite extensive research on automated test prioritization, fault localization, and program repair, test design and maintenance remain under-explored. This dissertation aims to understand test code quality from the perspective of design and maintenance practices, exploring various aspects of test design through manual classification, quantitative methods, and automated approaches.
The first part of the dissertation examines test smells, which are design issues that impact test code comprehension and maintainability. Although widely accepted in academia, it is unclear whether developers address test smells in practice. The thesis investigates developers' awareness of test smells and their impact on defect-proneness. Findings reveal that many proposed test smells persist and are removed incidentally through code deletion and refactoring, with minimal effect on software defect density. This dissertation aims to provide empirical support for re-ranking current test smell catalogs.
As test automation grows, modern frameworks like JUnit and TestNG are increasingly used in Java-based systems. These frameworks introduce annotation-driven development, which manages crucial aspects of the test execution lifecycle. Our second aim is to use mining software repository techniques to provide insights into how modern testing frameworks are used to improve test code design and maintenance. First, we investigate annotation usage in test code, providing a manual classification of their API usage and misuses (e.g., test smells). Second, we conduct a detailed empirical analysis of one controversial practice: disabling tests using JUnit and TestNG's @Ignore annotation. While this alleviates maintenance challenges, it introduces technical debt as it does not validate software quality. Finally, we examine the tradeoffs between reusability and redundancy in test code practices, particularly through inheritance, highlighting issues in test increasing test execution time.
Through case studies and experiments, the techniques proposed in this dissertation offer new insights into test code quality that may guide the development of automated tool support in the future, which may help developers improve test maintenance
On the Syntax-Morphology Divide: Towards a Unified Analysis of Causatives. The Case of Hungarian and Japanese
Get PDFThis thesis argues for a unified syntactic analysis of causatives. Previous literature has taken contrasts between Hungarian and Japanese morphological causatives as evidence that Hungarian causatives are derived in the lexicon via arity operations, while their Japanese counterparts are derived in the syntax via Merge and Agree. It is shown that the contrasts between Hungarian and Japanese causatives can be accounted for within the syntax, without needing to posit a separate computational component in the lexicon. Specifically, I argue that the locus of variation has to do with the size of the complements taken by Hungarian and Japanese causatives. Following Pylkkänen (2008)’s causative typology, I assume that Hungarian causatives embed little-v (i.e. they are ‘verb selecting’ in Pylkkänen’s terminology), while Japanese causatives embed Voice (i.e. they are ‘Voice-selecting’). Furthermore, I demonstrate that the syntactic analysis I propose accommodates empirical evidence that cannot be accounted for under a lexicalist analysis
“Your Why for Life”: Understanding the Benefits, Mechanisms, and Maintenance of Purpose in Life
Get PDFIt was been suggested that “if you have your why for life, you can get by with almost any how” (Nietzsche, 1889/1997, p. 6). Purpose in life is the “why”, the viewpoint that your life has personal meaning and direction (Ryff, 2014). Numerous studies have found empirical support for the abundance of benefits that are associated with having a sense of purpose, including its positive effects on mental and physical health (e.g., Boyle et al., 2012), as well as its ties to other adaptive traits, such as gratitude, compassion, and grit (Damon & Malin, 2020). However, gaps in the research literature remain in terms of particular populations that may benefit from purpose, when it might be effective, how it functions, and how it is maintained. The current investigation into purpose in life was done across three studies, with a particular focus on the emerging adulthood period of the lifespan, which some have argued is a time of inherent purposeful exploration and development (Pfund et al., 2020). Using longitudinal multilevel modeling in study 1, the benefits of purpose in life were explored during a transitional period in emerging adulthood, namely the passage from postsecondary studies to employment. It was found that a greater sense of purpose in life helps emerging adults during this often challenging transition, by allowing them to appraise their employment situation more positively. In study 2, with a cross-sectional design, results indicated that a greater sense of purpose was associated with lower rates of distress among university students in the face of a global stressful event (i.e., the COVID-19 pandemic). Notably, a stronger purpose in life allowed emerging adults to engage in adaptive strategies such as positive reframing and appraising a stressful event as less threatening to oneself, thus reporting lower distress. Lastly, in study 3, a longitudinal structural equation model showed that, by engaging in acceptance-based meaning making strategies, university students were able to sustain their sense of purpose during an unprecedented global health crisis. These findings aim to inform future interventions and research into how individuals may develop a purpose in life and experience its benefits
Working in the “New Normal”: Exploring the Link Between Hybrid Work and Need Fulfillment
Get PDFThis study investigated the mechanisms through which daily work modalities (i.e., work-from- home (WFH) and office-based work) influence daily fulfillment of employees’ basic psychological needs. Grounded in Self-Determination Theory (SDT), I hypothesized that daily fulfillment of the needs for autonomy and competence is higher on days when an employee is in the WFH modality than on days when they are in the office modality, and that daily fulfillment of the need for relatedness is higher for days in the office modality than for days in the WFH modality. I further hypothesized that perceived monitoring, perceived locational autonomy, workplace interruptions, and social interactions are mechanisms through which these effects occur. I also proposed that shared modality prevalence (the extent to which colleagues share the same daily work modality) moderates the effects of daily work modality on these mechanisms. A longitudinal daily diary methodology was employed to capture within-person variability and to examine how differences in work modalities affect need satisfaction at the within-individual level. Survey data were collected from 142 participants over five workdays (N = 655 observations). A multilevel modeling approach was used to test hypotheses, accounting for both within-person and between-person variability. In line with my hypotheses, I found that the WFH modality was associated with enhanced fulfillment of the need for autonomy, mediated by reduced perceived monitoring and increased perceived locational autonomy. I also found that the office modality was associated with enhanced fulfillment of the need for relatedness, mediated by positive social interactions. In contrast to my hypotheses, no significant effects of work modality on fulfillment of the need for competence were found. This research advances the application of SDT to hybrid work environments, contributing to the literature on the impact of daily work modalities on psychological need satisfaction. It also establishes a foundation for future studies to investigate the nuanced effects of hybrid work arrangements on employee motivation. The practical implications of these findings are noteworthy. Organizations can enhance employee motivation by offering flexibility in work location and minimizing surveillance practices, thereby fostering a sense of autonomy. A balanced hybrid work model that combines flexible WFH options with coordinated, mandatory in-office days is recommended. This approach ensures that employees can enjoy the autonomy benefits of working from home while still reaping the social and relational advantages of in-person interactions, thereby optimizing both individual well-being and organizational cohesion
Relationship between Movement Competence and Degree of Sports Specialization in 8-to-12-year-old Football Players
Get PDFAn increase in youth sport specialization prevalence has been associated with an increase injury rate and a decrease in movement competence. However, movement competence has not been compared between the degrees of sport specialization in 8- to 12-year-old football players. The purpose of the study is to primarily observe the relationship between movement competence and the degree of youth sports specialization in 8- to 12-year-old football players using the Child Focused Injury Risk Screening Tool (ChildFIRST). Secondly, the study aims to observe the differences amongst positions and the association for injury prevalence. We hypothesize that youth football players with a higher youth sport specialization categorization will have a lower movement competency. We also hypothesize that there with be a difference in movement competency amongst football positions.
During practices in the 2023 football season, 8- to 12-year-old football players from the Montreal Regional Football League were asked to complete an injury and youth sport specialization questionnaire. Participants were then assessed using the ChildFIRST. There was no significant association between ChildFIRST composite score and youth sport specialization score. When looking at the differences amongst positions, linemen had a significantly lower ChildFIRST composite score mean than other positions. No association with injury and movement competence was observed. Future studies should continue observing the movement competency in 8- to 12-year-old football players differentiating by their playing position. Such findings could contribute towards the development of an evidence-based injury prevention program for youth football players
Comparative Study of the Effects of Device Geometry on the DC Characteristics, Linearity and Low-Frequency Noise Performance of Lattice-matched InAlN/GaN HFETs
Get PDFThe novel lattice-matched HFETs realized on an epilayer consisting of a thin In0.17Al0.83N barrier layer grown on top of an undoped GaN channel have been demonstrated over the past decade to enjoy improved stable high-frequency power characteristics compared to their famous AlGaN/GaN counterparts. This is specially thanks to employing a lattice-matched barrier enjoying substantial spontaneous polarization-induced sheet charge density.
An extensive body of research implies a significant correlation exists between electronic device technology, the level of 1/f noise, and the manifestation of generation–recombination (G-R) bulge signatures. This correlation has been shown to offer a highly sensitive foundation for reliability and further performance optimization of electronic devices. A decrease in low frequency noise (LFN) has a significant impact on oscillator phase noise and the performance of intermediate frequency (IF) amplifiers and mixers Thus, in this thesis, the low frequency drain noise-current characteristics of metallic-face InAlN/AlN/GaN heterostructure field effect transistors (HFETs) having fin structures only under the gate, while maintaining a planar structure in the access regions, are compared to those of the HFETs having fin structures stretched from source to drain. This work aims to address the possible difficulties in the performance of these devices. Evidence indicates that both device types follow the trends of carrier number fluctuation (CNF) with correlated mobility fluctuation (CMF) model of 1/f noise. Devices having fin structures just under the gate are exhibiting improved 1/f noise performance with lower drain noise-current spectral density.
Since a good gate-transconductance (Gm) linearity, specially at high gate over-drives, is essential to linear high-frequency amplifiers intended for use in modern telecommunication system (such as 6G networks), enhancing the linearity of the deeply scaled HFETs implemented on such epilayers is very much in demand. Moreover, the investigation of the on-state breakdown mechanism is beneficial for the definition and design of the safe operation area (SOA) in GaN-based high power amplifiers and switches. In this thesis, I have investigated the impact of the scaling of the gate-source length (LGS) and gate length (LG) on the DC characterises and gate-transconductance (Gm) linearity of metallic-face InAlN/AlN/GaN heterostructure field effect transistors (HFETs) having fin structures only under the gate and those having them stretched from source to drain. Evidence for both device types suggests that the downscaling of LGS and LG augments the electron velocity in the source-access region, as a result of which the higher carrier density under the gated-channel improves the maximum drain-current density but not necessarily the Gm linearity of the device. It is shown that the devices having a planar and longer source access region are exhibiting relatively improved gate-transconductance linearity. In addition, the downscaling of the LG is observed to have a positive influence on device linearity. However, the negative impact of the downscaling of the LGS and LG on the on-state breakdown voltage has been observed.
In addition, in this thesis I have investigated the effect of partially etching of the gated barrier on the Gm linearity of lattice-matched InAlN/GaN HFETs. Simulation results show an improvement in linearity observed through broadening the gate transconductance characteristics of Vth-modulated HFETs over the non-recessed and an alternative recessed HFET, for which gated barrier was uniformly recessed
License Plate Detection and Character Recognition using Deep Learning and Font Evaluation
Get PDFLicense plate detection and character recognition pose challenges due to environmental
sensitivity, such as lighting, dust, and the impact of the chosen font type on recognition
tasks. Automatic License Plate Detection and Recognition (ALPR) are crucial
in practical applications such as traffic control and parking, vehicle tracking, toll
collection, and law enforcement. While much research has been done using image
processing and machine learning algorithms, deep learning methods need further
exploration due to their recent advances in reliable performance in various scenarios.
Moreover, current proposals are limited to specific regions and dataset applicability.
This study has a dual focus: firstly, we suggest utilizing a Deep Learning technique,
specifically using Faster R-CNN for the license plate detection task and a CNN-RNN
model with CTC loss, and a MobileNet V3 backbone for recognition task. We also
utilized You Only Look Once (YOLO) for license plate detection and recognition tasks.
Secondly, we aim to assess font features within the LP context. This work uses Brazilian
dataset and datasets from two different provinces in Canada and two different states in
the United States of America, including Ontario, Quebec, California, and New York
State. We suggest employing an adaptive algorithm based on Faster R-CNN and CTC
network along with YOLO, fine-tuned with optimized parameters to improve its
effectiveness using two different approaches, including domain generalization.
Alongside presenting the recall ratio findings, this study will perform a thorough error
analysis to gain insights into the nature of false positives. The proposed model
demonstrated a commendable recall ratio of 94% using a single YOLO network.
Specific fonts pose readability challenges for humans, while others present difficulties
for computer systems regarding recognition. In this study, we provide five sets of
outcomes for font assessment: results about font anatomy and those related to the
recognition of commercial products. The font anatomy analysis focuses on five specific
fonts: Driver Gothic, Dreadnought, California Clarendon, Zurich Extra Condensed, and
Mandatory. Additionally, we assess the impact of these fonts in the context of a dataset
made of five different license plates using a commercial product, OpenALPR. The font
anatomy findings unveil significant confusion cases and quality features associated with
chosen fonts
Beyond the Hype. Deploying and Evaluating a Conversational Agent Using LLMs in an Academic Setting
Get PDFThis presentation will cover our ongoing work investigating and deploying generative AI technology in the context of libraries and memory institutions. It’s not novel that libraries provide online human or machine-based chat services, but taking advantage of generative AI requires new technical approaches and considerations around the ethics and usefulness of conversational agents. We will discuss our development of a chatbot configured for delivering academic library information services. This includes defining a protocol for assessing and guiding implementation decisions as well as evaluating the tool’s utility.
Our initial step in developing the chatbot involved building a knowledge base (stored on an in-house metadata management system), which could be connected to generative AI technology. Next, we experimented with a variety of open source and proprietary language models to understand how each performs. We are testing the following approaches: A closed source large language model (Bing Chat / Gemini / ChatGPT) prompted to act as reference personnel; a context-aware closed source LLM (OpenAI GPT); and a context-aware open source LLM (Llama). We are testing with questions that a useful chatbot should be able to answer. The chatbot’s responses for each approach are evaluated comparatively.
A key objective of this project is the testing protocol and evaluation framework. Reference questions often require a dynamic conversation, iterating on the direction of inquiry. This makes it challenging to evaluate outputs as merely accurate or inaccurate. Our study builds on Lai (2023) to develop a testing protocol, incorporating multiple dimensions of user interactions. Our protocol will support the interrogation of ethical concerns around these technologies and their application. We are operationalizing aspects of the LC Labs AI Planning Framework (Library of Congress, 2023) to define use cases for generative AI in information services and ethical criteria