21793 research outputs found
Sort by
Ouvrir les lectures de la ville : traduction, paysage linguistique et solidarité à Parc-Extension
Get PDFSis à l’intersection de la traductologie et de l’étude des paysages linguistiques, ce mémoire interroge la place et la signification de la traduction dans la création du paysage linguistique en contexte multiethnique. Avec pour toile de fond Parc-Extension, un quartier multiculturel et historiquement défavorisé aujourd’hui à la croisée des chemins, ce mémoire s’appuie sur des entrevues sur le terrain pour mettre en vitrine trois projets communautaires ou artistiques qui ont mobilisé la traduction afin d’inscrire la diversité linguistique dans l’espace urbain : le documentaire Je me souviens d’un temps où personne ne joggait dans ce quartier de la réalisatrice Jenny Cartwright; l’initiative d’entraide Aide mutuelle Parc-Extension, née dans la foulée de l’annonce de la pandémie de COVID-19; et la création d’une banque multilingue de ressources sur la pandémie par l’Alliance des communautés culturelles pour l’égalité dans la santé et les services sociaux (ACCÉSSS). S’inspirant des travaux de Sherry Simon sur la traduction dans la ville, ce mémoire démontre l’intérêt d’étudier le paysage linguistique dans une perspective traductologique. Cette approche rend manifeste l’historicité du paysage linguistique et y situe la traduction comme un acteur déterminant qui permet non seulement de donner une visibilité à des langues trop souvent invisibles, mais aussi de rassembler les communautés autour d’une action collective sur les lectures de la ville
Pavement Defect Classification and Localization Using Hybrid Weakly Supervised and Supervised Deep Learning and GIS
Get PDFAutomated detection of road defects has historically been challenging for the pavement management industry. As a result, new methods have been developed over the past few years to handle this issue. Most of these methods relied on supervised machine learning techniques, such as object detection and segmentation methods, which need a large, annotated image dataset to train their models. However, annotating pavement defects is difficult and time-consuming due to their ununiformed and complex shapes. To address this challenge, a hybrid pavement defect classification and localization framework using weakly supervised and supervised deep learning methods is proposed in this thesis. This framework has two steps: (1) A robust hierarchical two-level classifier that classifies the defects in images, and (2) A method for defect localization combining weakly supervised and supervised techniques. In the localization method, first, defects are primarily localized using a weakly supervised method (i.e. Class Activation Mapping (CAM)). Next, based on the results of the first classifiers, the defects are segmented from the localized patches obtained in the previous step. The feature maps extracted from the CAM method are used to train a segmentation network once (i.e. U-Net or Mask R-CNN) to localize and segment the defects in the images. Thus, the proposed framework combines the advantages of weakly supervised and supervised methods. The supervised modules in the framework are trained once and can be used for any new data without the need to train. In other words, to use our framework on new dataset only the classifiers should be fine-tuned. Furthermore, the proposed framework introduced an innovative method designed to calculate the maximum crack width in pixels within linear segmented defect patches, derived from the localization module of the proposed framework. This method is particularly advantageous as it provides critical information that can be further employed in the calculation of the Pavement Condition Index (PCI).
Additionally, the proposed method benefits from an asset management inspection system based on Geographic Information System (GIS) technology to prepare the dataset used in the training and testing. Thus, this advanced system serves a dual role within our framework. Firstly, it assists in the assembly and preparation of the dataset used in the model training process, providing a geographically organized collection of images and related data. Secondly, it plays a crucial role in the testing phase, offering a spatially accurate platform for evaluating the effectiveness of the model in real-world scenarios.
A dataset from Georgia State in the USA was used in the case study. The proposed framework obtained high precision of 97%, 88%, 92% and 97% for localizing the alligator, block, longitudinal and transverse cracks, respectively. Considering all factors, such as annotation cost, and performance on the test dataset, the proposed localization method outperforms the supervised localization methods, such as instance segmentation and object detection for localizing road pavement defect
SOLIDARISMO: The Rise and Resilience of the Costa Rican Solidarist Movement
Get PDFSolidarismo (Solidarism), is an economic and labour movement presented in Costa Rica in 1947. It has become a point of contention among labour activists and scholars for decades, creating the long-standing Solidarist-Syndicalist divide in Costa Rica. Despite evidence that Solidarism in Costa Rica has been used as a weapon against trade unionism, there exists the reality of its predominance in the country. In this dissertation, I offer an analysis that explains the prevalence of the Costa Rican Solidarist Movement (MSC), and its sustenance over time. I rely on both a historical and empirical analysis to provide an account of the rise and resilience of the MSC. I argue that the MSC was strategically founded at a pivotal political moment in Costa Rican history, during the anticommunist climate of the Costa Rican Civil War. As such, the movement presented itself as a middle ground between the individualism of liberalism and the collectivism of socialism. Furthermore, the movement owes its ubiquity to having deliberately forged its identity in line with the collective identities of nationalism (specifically, Costa Rican exceptionalism) and religion in the country. Over the years, the movement has seen continued support. I use the case of Del Monte Foods Inc.’s subsidiary, Pineapple Development Corporation (Pindeco), in Volcán de Buenos Aires, as a case study, to provide an ethnographic representation of community members’ lived experiences with solidarist organizations such as Solidarist Associations of Employees (ASE) and Permanent Worker Committees (CPT). I tell the story of the significance of Solidarism in the lives of workers in Volcán. I use the ethnographic data collected over 12 years in Volcán, to represent the experience of Volcanians and the history that has shaped the prevalence of the movement in the country
Exploring the neuro-computational mechanisms underlying age-related changes in complex decision-making
Get PDFOver the last decade, research in decision-making has made remarkable advancements in understanding how the relative engagement in model-based and model-free decision-making changes with healthy aging. While we are beginning to understand the factors that affect older adults’ shift away from model-based decision-making, the exact mechanisms at play are still poorly understood. This dissertation presents findings as well as a novel theory which aims to advance our understanding of these neuro-computational mechanisms. Chapter 2 demonstrates
that, in contrast to younger adults, older adults do not benefit from more distinct probabilistic transitions between stages in a two-step decision-making task. By examining trial-by-trial neurocomputational dynamics, this first empirical paper provides evidence for age-related deficits in the ability to represent probabilistic transitions, and predict the value of upcoming choice options. Chapter 3 presents a novel theory: the diminished state space theory of human aging. This theoretical contribution proposes that older adults’ deficits in model-based learning
are due to their underlying difficulties in representing state spaces. Chapter 4 examines one of the computational explanations brought forward in this theoretical paper. Namely, that older adults’ diminished state spaces may be explained (at least in part) by their difficulties updating their internal task representation. In line with this hypothesis, results demonstrate that in contrast to younger adults, older adults show difficulties identifying outcomes that signal the need to update their internal model. Together, these findings suggest that older adults’ deficits in model-based decision-making can be explained by their diminished state space representations, which in turn may in part result from their difficulty updating their internal model during cognitive tasks. Ultimately,
this dissertation provides important insights regarding older adults’ deficits, and opens future directions for the study of age-related changes in representational abilities
Embedded spherical probabilistic modeling for topic discovery and text representation learning
Get PDFEvery day, large amounts of text data are generated on the web. Taking advantage of such data necessitates good methods of retrieval, exploration, and analysis to extract hidden knowledge from these voluminous unstructured texts. In this context, probabilistic topic modeling is regarded as an effective text mining technique that uncovers the main topics from an unlabeled set of documents. Topic models have been successfully used in various domains to exhibit hidden topics, e.g., marketing, medicine, and political sciences. However, the inferred topics by conventional topic models are often unclear and not easy to interpret, because they do not account for semantic structures in language. Recently, several topic modeling approaches have been proposed to leverage external knowledge to enhance the quality of the learned topics, but they still assume a Multinomial or Gaussian document likelihood in the Euclidean space, which often results in information loss and poor performance. In this thesis, we introduce a set of probabilistic embedded spherical topic models designed to address several challenges, including lack of topic interpretability, high-dimensionality, and sparsity. Our approaches involve integrating knowledge graphs and word embeddings within a non-Euclidean curved space, namely the hypersphere, to enhance topic interpretability and generate discriminative text representations. The proposed models effectively handle a wide range of scenarios, encompassing unsupervised and supervised learning tasks. Experimental results demonstrate the effectiveness of the proposed algorithms in discovering coherent topics and learning high-quality text representations, which prove valuable for common Natural Language Processing (NLP) tasks across diverse benchmark datasets. These findings further highlight the advantages of modeling textual data on the surface of the unit-hypersphere using directional distributions while incorporating word and knowledge graph embeddings
Passive IoT Device-Type Identification Using Few-Shot Learning
Get PDFThe ever-growing number and diversity of connected devices have contributed to rising network security challenges. Vulnerable and unauthorized devices may pose a significant
security risk with severe consequences. Device-type identification is instrumental in reducing risk and thwarting cyberattacks that may be caused by vulnerable devices. At present, IoT device identification methods use traditional machine learning or deep learning techniques, which require a large amount of labeled data to generate the device fingerprints.
Moreover, these techniques require building a new model whenever a new device is introduced. To address these limitations, we propose a few-shot learning-based approach on
siamese neural networks to identify IoT device-type connected to a network by analyzing their network communications, which can be effective under conditions of insufficient labeled data and/or resources. We evaluate our method on data obtained from real-world IoT devices. The experimental results show the effectiveness of the proposed method even with a small amount of data samples. Besides, it indicates that our approach outperforms IoT Sentinel, the state-of-the-art approach for IoT fingerprinting, by a margin of 10% additional accuracy
Feature-Based Polynomial Adaptation for High-Order Methods Applied to Martian Aerodynamics
Get PDFThis work presents a comparative study of different feature-based polynomial adaptation strategies for high-order unstructured methods applied to unsteady turbulent flows. Recently, the Flux Reconstruction (FR) approach has been introduced as a unifying framework for high-order unstructured spatial discretizations. To achieve high-order accuracy, FR utilizes an element-wise polynomial representation of the solution. In the current work, we consider three indicators for local adaptation of this polynomial degree. These are based on non-dimensional indicators that track the maximal vorticity norm, Frobenius norm of the velocity gradient, or eigenvalue modulus of the velocity gradient with respect to the effective maximal local grid spacing and free stream velocity. These feature-based methods are simple to implement and have the potential to track small-scale turbulent structures that arise in scale-resolving simulations, such as Direct Numerical Simulation (DNS) and Large Eddy Simulation (LES). The vorticity, gradient, and eigenvalue-based polynomial adaptation strategies with the FR approach are used to solve the compressible Navier-Stokes equations. DNS simulations are performed for unsteady laminar flow over a two-dimensional circular cylinder, turbulent flow over a three-dimensional sphere, and massively separated flow over a Martian helicopter rotor airfoil section. The results show a reduction in computational cost, with approximately one-quarter of the number of degrees of freedom relative to a non-adaptive case. The gradient-based method remains consistent for each numerical application, concluding to be the most well-suited feature-based indicator among the three considered
Deep Learning Methods for Hand Gesture Recognition via High-Density Surface Electromyogram (HD-sEMG) Signals
Get PDFHand Gesture Recognition (HGR) using surface Electromyogram (sEMG) signals can be considered as one of the most important technologies in making efficient Human Machine Interface (HMI) systems. In particular, sEMG-based hand gesture has been a topic of growing interest for development of assistive systems to improve the quality of life in individuals suffering from amputated limbs. Generally speaking, myoelectric prosthetic devices work by classifying existing patterns of the collected sEMG signals and synthesizing intended gestures. While conventional myoelectric control systems, e.g., on/off control or direct-proportional, have potential advantages, challenges such as limited Degree of Freedom (DoF) due to crosstalk have resulted in the emergence of data-driven solutions. More specifically, to improve efficiency, intuitiveness, and the control performance of hand prosthetic systems, several Artificial Intelligence (AI) algorithms ranging from conventional Machine Learning (ML) models to highly complicated Deep Neural Network (DNN) architectures have been designed for sEMG-based hand gesture recognition in myoelectric prosthetic devices. In this thesis, we, first, perform a literature review on hand gesture recognition methods and elaborate on the recently proposed Deep Learning/Machine Learning (DL/ML) models in the literature. Then, our utilized High-Density sEMG (HD-sEMG) dataset is introduced and the rationales behind our main focus on this particular type of sEMG dataset are explained. We, then, develop a Vision Transformer (ViT)-based model for gesture recognition with HD-sEMG signals and evaluate its performance under different conditions such as variable window sizes, number of electrode channels, and model's complexity. We compare its performance with that of two conventional ML and one DL algorithm that are typically adopted in this domain. Furthermore, we introduce another capability of our proposed framework for instantaneous training, which is its ability to classify hand gestures based on a single frame of HD-sEMG dataset. Following that, we introduce the idea of integrating the macroscopic and microscopic neural drive information obtained from HD-sEMG data into a hybrid ViT-based framework for gesture recognition, which outperforms a standalone ViT architecture in terms of classification accuracy. Here, microscopic neural drive information (also called Motor Unit Spike Trains) refers to the neural commands sent by the brain and spinal cord to individual muscle fibers and are extracted from HD-sEMG signals using Blind Source Separation (BSP) algorithms. Finally, we design an alternative and novel hand gesture recognition model based on the less-explored topic of Spiking Neural Networks (SNN), which performs spatio-temporal gesture recognition in an event-based fashion. As opposed to the classical DNN architectures, SNNs are of the capacity to imitate human brain's cognitive function by using biologically inspired models of neurons and synapses. Therefore, they are more biologically explainable and computationally efficient
Driving force computation for fatigue crack growth based on the integration of fracture mechanics with artificial neural networks
Get PDFFracture mechanics principles play a crucial role in characterizing fatigue crack growth (FCG) rates based on the concept of driving force. Two well-known and promising driving forces in fracture mechanics are the stress intensity factor range (∆K) and cyclic J-integral (∆J). While ∆K is a linear elastic fracture mechanics (LEFM) parameter, ∆J is an elasto-plastic fracture mechanics (EPFM) parameter. However, both driving forces have limitations when it comes to FCG characterization. ∆K fails to account for relatively large-scale plasticity, rendering it inadequate for describing the short crack (SC) regime. On the other hand, ∆J inherently has the potential to consider large-scale plasticity, but its application on real engineering problems is challenging. The difficulty arises from the need to perform complex and time-consuming elasto-plastic analyses to compute the actual elasto-plastic stress, strain, and displacement fields near the crack tip for the calculation of ∆J. This study explores the integration of artificial neural networks (ANNs) with fracture mechanics principles to overcome these challenges. The research is carried out in three phases:
Phase 1 focuses on integrating ANN with ∆K as a LEFM parameter. Unlike ∆K-based models that solely formulate FCG rate based on the maximum stress intensity factor (Kmax) and ∆K, this approach incorporates other controlling parameters. FCG rate is considered as a function of ∆K and stress ratio (R) in the long crack (LC) regime, and as a function of stress level (σ) in addition to ∆K and R in the SC regime. ANNs are developed to reveal these non-linear and complex functions in both regimes, using experimental FCG data sets from Ti-6Al-4V titanium alloy, 2024-T3, and 7075-T6 aluminum alloys for training and verification. Although this phase shows potential, the reliance on limited FCG data sets due to costly procedures remains a challenge. Moreover, ∆K as a LEFM parameter inherently cannot handle large-scale plasticity in the SC regime.
To address these issues, a novel approach is suggested and investigated in Phases 2 and 3. Phases 2 and 3 propose replacing ∆K with ∆J as a promising EPFM driving force and combining finite element (FE) analyses with ANN algorithms. Firstly, the implementation of FE models provides ample datasets for training the ANNs. Secondly, this integration allows for the determination of ∆J through a linear elastic solution rather than complex elasto-plastic analyses.
Phase 2 involves FE analyses to determine stress, strain, and displacement fields under elastic and elasto-plastic states near a crack tip for a notched specimen made of stainless steel (SS304) under monotonic loading. Hypothetical elastic stress, strain, and displacement fields around the crack tip are used as input data for the developed ANNs. The corresponding actual elasto-plastic stress, strain, and displacement fields are the output of the ANNs. Well-trained ANNs successfully establish relationships between the elastic and elasto-plastic fields, enabling predictions of elasto-plastic stress, strain, and displacement based on hypothetical elastic data. An in-house model based on the equivalent domain integral (EDI) method is developed to determine J-integral as a function of stress, strain, and displacement fields around the crack tip. This model can be served as a post-processing step after elasto-plastic FE analyses. In addition, it can be employed to determine J-integral based on ANN predictions. The accuracy of the in-house model is verified by the J-integral data in the literature. ANN predicted elasto-plastic stress, strain, and displacement fields are compared and verified with those obtained from elasto-plastic FE analyses. The proposed method demonstrates significant accuracy in determining J-integral values.
Phase 3 extends the approach to cyclic loading conditions. The developed ANNs are trained on cyclic stress, strain, and displacement fields. The in-house model is upgraded to determine ∆J under cyclic loading. The accuracy of cyclic ANN-predicted elasto-plastic stress, strain, and displacement fields is compared with those obtained from elasto-plastic FE models, resulting in significant agreement. The in-house model is verified by the ∆J data in the literature. Moreover, ∆J values predicted by the proposed model are comparable to those directly determined by elasto-plastic FE analyses.
The integration of artificial neural networks with fracture mechanics principles provides valuable insights into overcoming traditional driving force limitations in FCG characterization. This research offers a promising avenue for future research and practical applications in the field of fatigue crack growth analysis
Baby Boy Cousins: Looking for Roots, the Essay Documentary
Get PDFHow does the essay documentary curate truth, and who benefits from the personal transparency required of the essay film – the audience or the filmmaker? This research creation examines the thematic and conceptual research and created methodologies employed by Montreal filmmaker Adrian Wills to create his first essay film “Baby Boy Cousins”. Adopted as an infant, Wills embarked on a two-year journey in search of his birth family in Newfoundland for this research creation. Through real-time documentation, he uncovered the heartbreaking truth that his birth mother had taken her own life, amongst other family secrets. “Baby Boy Cousins: Looking for Roots, the Essay Documentary” investigates the essay film’s unique ability to authentically explore subjective experiences and emotions while acknowledging the limitations of objective truth. It explores the work of acclaimed essay filmmakers Sarah Polley, Alan Berliner, Chantal Akerman, and Deann Borshay Liem in relation to the construction of “Baby Boy Cousins”. Moreover, this research creation examines the intricate ethical considerations surrounding personal transparency. It investigates the delicate balance between meeting audience expectations and safeguarding the privacy of the filmmaker. “Baby Boy Cousins: Looking for Roots, the Essay Documentary” concludes with profound insights into the transformative power of personal filmmaking while acknowledging the essay filmmaker’s need for psychological self-care when engaging in personal vulnerability