1722 research outputs found
Sort by
N-Dimensional Polynomial Neural Networks and their Applications
Get PDFIn addition to being extremely non-linear, modern machine learning problems require millions
if not billions of parameters to solve or at least to get a good approximation of the solution,
and neural networks are known to assimilate that complexity by deepening and widening their
topology in order to increase the level of non-linearity needed for a better approximation. However,
compact topologies are always preferred to deeper ones as they offer the advantage of using
less computational units and less parameters. This compactness comes at the price of reduced
non-linearity and thus, of limited solution search space. This thesis proposes the N-Dimensional
Polynomial Neural Network (NDPNN) model that uses automatic polynomial kernel estimation
for N-Dimensional Convolutional Neural Networks (NDCNNs) and introduces a high degree of
non-linearity from the first layer which can compensate the need for deep and/or wide topologies.
We first theoretically formalized the 1DPNN model which can process 1-dimensional
signals and we demonstrated that its inherent non-linearity enables it to yield better results
with less computational and spatial complexity than a regular 1DCNN on various classification
and regression problems related to audio signals, even though it introduces more computational
and spatial complexity on a neuronal level. The experiments were conducted on three publicly
available datasets and demonstrate that the proposed 1DPNN model can extract more relevant
information from the data than a 1DCNN in less time and with less memory. We subsequently
extended the theoretical foundation of the 1DPNN to NDPNN which can process 2D signals such
as images and 3D signals such as videos. Also, we theoretically created a general polynomial degree
reduction formula that we used to develop a heuristic algorithm, which enables the degree
reduction of any pre-trained NDPNN. This algorithm compresses an NDPNN without altering
its performance, thus making the model faster and lighter. Following that, we used 2DPNNs
and 3DPNNs to tackle the problem of plant species recognition on a publicly available plant
species recognition dataset composed of 40,000 images with different sizes consisting of 8 plant
species. As a result, we created a novel method, called Variably Overlapping Time—Coherent
Sliding Window (VOTCSW), that transforms a dataset composed of images with variable size
to a 3D representation with fixed size that is suitable for convolutional neural networks, and
we demonstrated that this representation is more informative than resizing the images of the
dataset to a given size. We theoretically formalized the use cases of the method as well as its
inherent properties and proved that it has an oversampling and a regularization effect on the
data. By combining the VOTCSW method with 3DPNNs, we were able to create a model that
achieved a state-of-the-art accuracy of 99.9% on the considered dataset, surpassing well-known
architectures such as ResNet and Inception. Furthermore, we established that the currently
available plant species dataset could not be used for machine learning in its present form, due
to a substantial class imbalance between the training set and the test set. Hence, we created
a specific preprocessing and a model development framework that enabled us to improve the accuracy from 49.23% to 99.9%. The contributions of this thesis are the creation of a novel
generic model called NDPNN that can extract more information from data than a NDCNN with
less computational and spatial complexity, the evaluation of the performance of NDPNNs on
audio signals, images and videos, the creation of a general direct polynomial reduction formula,
the design of a heuristic algorithm for NDPNN compression that generates faster and lighter
models, the formalization of an image transformation method that circumvents image resizing
without altering fine-grained information, and the production of a state-of-the-art 3DPNN for
plant species recognition."I wish to express my thanks and appreciation to Mitacs, the NSERC, and the Faculty of Graduate
Studies of the University of Winnipeg for funding my research."Master of Science in Applied Computer Scienc
Exploring Deep Neural Networks for Plant Image Classification
Get PDFAutomatically distinguishing different types of plant images is a challenging problem relevant to both Botany and Computer Science disciplines. Plant identification at the species level is a computer vision task called fine-grained categorization, which focuses on differentiating between hard-to-distinguish object classes. This classification problem is complicated and challenging because of the lack of annotated data, inter-species similarity, the large-scale features in appearance, and a large number of plant species. A plant classification system capable of addressing the complexity of this computer vision problem has important implications for society at large, not only in public computer science education but also in numerous agricultural activities such as automatic detection of cash crops and non-crop plants (called weeds). Furthermore, successful automation of crop and weed identification will lead to the reduction of chemical compounds currently used to eliminate weeds [15]. Deep Convolutional Neural Networks (CNN) can be a solution to perform this computer vision task. In this thesis, seven different CNN models are deployed to classify 1 million images - from the TerraByte dataset - of eleven very similar plant species [13]. This robust approach divides the problem into two main steps: the first step, called the generalist, identifies similar plants and separates them into different groups that contain indistinguishable plant species. The second step, called specialist, is used to classify plants within the groups of indistinguishable plants, including five weed and seven crop species, with high accuracy. The generalist-specialist CNN network shows that the hierarchical network outperforms simple CNN models in terms of accuracy and classifying similar plant images. The contributions of this thesis are the explored different CNN models and improved performance of those models by designing and implementing the generalist-specialist CNN models for classifying similar plant images."I would ... like to thank Mitacs, George Weston Ltd, the Natural Sciences and Engineering
Research Council of Canada (NSERC), and the Faculty of Graduate Studies for their financial
support of this work."Master of Science in Applied Computer Scienc
Invisibility cloaks, prisons, and a pandemic: Did COVID-19 render the prison invisibility cloak ineffective?
Get PDFThis thesis acknowledges the importance of examining news media representation of prisons, and more specifically, news media representation of Canadian prisons during the COVID-19 pandemic. A thematic qualitative and quantitative content analysis of news coverage of prison before and during the COVID-19 pandemic was undertaken to analyze how news media organizations communicate meanings and messages about punishment and prison to the general public. Utilizing a social constructionist approach, I examined how the news media frames coverage of correctional institutions, consequently shaping public understanding of punishment and prison which may impact correctional policy. This thesis addressed the following questions: 1) Was there an increase in news media coverage of prison during the COVID-19 pandemic? Has coverage during the COVID-19 pandemic made the prison more visible? 2) Does the news media coverage of correctional institutions during the COVID-19 pandemic reinforce traditional myths and stereotypes surrounding punishment and prison? Or challenge them? 3) Is the news media representation of correctional institutions during the COVID-19 pandemic consistent with coverage of traditional prison newsworthy items which focus on discrete incidents? Or does the coverage reflect newer, broader systemic newsworthy issues, namely, reform? The findings demonstrate that the COVID-19 pandemic did not quantitatively bring more visibility to prisons as assessed by the amount of news items, however, qualitatively it appears COVID-19 brought more visibility to prison issues. While some traditional prison stereotypes are still present in the news media and were reinforced during the pandemic, other myths and stereotypes were challenged, or were rare. Lastly, the results demonstrate that although traditional prison newsworthy items were still often reported in the news, discussions of prison reform were prevalent in the sample.Master of Criminal Justic
An Examination of the Effects of Elevated CO2 on Juvenile Salmonids
Get PDFThis thesis explores the behavioural and physiological effects of elevated carbon dixoide (CO2) on freshwater, juvenile samonids. Arctic Charr, Rainbow Trout, and Brook Charr were each exposed to elevated CO2 over a two week period. During this time, metabolic rate, growth, and behaviour were monitored. Feeding and tissues were examined after the exposure period. Generally, few effects of CO2 were observed; however, Arctic Charr had higher standard metabolic rate as a result of the CO2 exposure. Additionally, Arctic Charr gills showed signs of tissue damage that may have been due to the CO2 exposure. Overall, my findings suggest that short-term elevated CO2 may not be harmful for juvenile salmonids in freshwater. Future studies should explore longer and higher CO2 exposures, and the relationship between elevated temperature and CO2.NSERC Discovery Grant (10301); Research Manitoba Master's StudentshipMaster of Science in Bioscience, Technology, and Public Polic
Leveraging Collective Efficacy in The Dangerous Space Between Good Intentions and Meaningful Interventions: A Study on the Use of School Suspensions in Manitoba; A Review of Literature for Manitoba Education and Early Childhood Learning
No full textShannon Dicker on Professional Learning for Culturally Nourishing Pedagogies: Limitations, Applications, and Next Steps
No full textThis video was produced by Shannon Dicker as part of an Inuit Studies 2022 presentation. In it, she discusses her role as Indigenous Specialist with the Newfoundland and Labrador English school district, her role in professional learning for culturally nourishing pedagogies, and her reflections on successes of the SSHRC-funded project "Professional Learning for Culturally Nourishing Pedagogies in Nunatsiavut Area Schools"
Unsupervised Domain Adaptation using Satellite Images for Significantly Different Infrastructure Objects
Get PDFDeep learning has become one of the most efficient computer vision tools in recent years. The success and variety of deep learning semantic segmentation models inspired scientists in the remote sensing domain to apply them to satellite imagery. Here, these models can produce reliable land use land cover maps in a short amount of time. However, porting these models to new sensors or domains is still limited by the amount of labelled data for training the network. The image labelling process is time-consuming and expensive because it is often manual (or semi-automated) work and requires assigning a label to each pixel in a satellite image. One solution is to apply the semantic segmentation model trained on a domain with known labels to a domain where labels are missing. For this to work, the discrepancy between domains must be narrowed to produce acceptable results. However, in practice, domain discrepancy can be significant. Developing domain adaptation models to bridge this discrepancy is the problem considered in this thesis, and it is important because semantically similar objects can look different from one geographical area to another. Therefore, several state-of-the-art domain adaptations were considered and validated using GeoEye-1 and WorldView-2 satellite imagery. The GeoEye-1 images represented a Canadian land cover, and WorldView-2 represented the African continent; thus, the domain discrepancy was significant. The CyCADA model with adapted noisy labeller showed the highest performance among all the considered models and achieved 32.6% of mean intersection over union, which is 7.5% higher compared to the model without adaptation. The contributions of this thesis are an attempt at domain adaptation across domains with the significant structural discrepancy, structural improvements to the CyCADA and DAugNet models, and quantitative and qualitative analysis of model performance on domain adaptation with significant structural discrepancy.Manitoba Hydro, Arctic Gateway Group, Mitacs, NSERC, the Faculty of Graduate StudiesMaster of Science in Applied Computer Scienc
Clive Thomson, On croit comprendre le monde avec ça ! Entretiens mémoriels avec Henri Mitterand
No full texthttps://www.persee.fr/doc/lepec_0183-4681_2022_num_111_1_1029_t15_0257_0000_
It's JustiFlied: The Endogenous Retrovirus K Integrase Induces Motor Disturbances in Transgenic Drosophila
Get PDFAmyotrophic Lateral Sclerosis (ALS) is an incurable neurodegenerative disease characterized by the loss of cortical and spinal motor neurons. Endogenous retrovirus K (ERVK) is a genomic viral symbiont that has been associated with motor neuron loss in ALS. The ERVK integrase (IN) is an enzyme with a role in driving neuropathology and motor deficit. The primary role of the viral IN enzyme is to insert viral DNA into the host cell genome. Accumulating evidence also points to ERVK IN activity causing DNA damage and genomic instability in the host. In Drosophila, retroelement activity contributes to deregulation of the ALS risk gene TARDBP (TDP-43, TBPH in Drosophila) via DNA damage-mediated cell toxicity. This suggests a dynamic interaction between TDP-43 biology, DNA damage and retroelements. I have determined that motor disability in ERVK IN expressing Drosophila correlates with neuropathological evidence of DNA damage, inflammation, and TDP-43 aggregation. Viability and behavioral assays and the Trikinetics DAM5H monitor were used to assess motor impairment in ERVK IN expressing flies. Two FDA approved HIV integrase inhibitors were administered to determine if the progression of motor impairments could be limited. Western blot analysis was used to monitor changes in ERVK IN, ãH2AV (DNA damage marker), TDP-43, PARP1 and other related proteins over time. Pathological molecular markers were correlated with behavioural assays for motor function, to identify potential biomarkers. Establishment of this model allowed me to assess the association between ERVK IN-driven motor impairment and neuropathological outcomes. Determining the effect of integrase inhibitors in ERVK IN expressing Drosophila is a crucial step towards evaluating antivirals as a novel therapeutic strategy for the reversal of motor neuron damage and motor deficit in ALS.Master of Science in Bioscience Technology and Public Polic
