21793 research outputs found
Sort by
Outliers detection based on quantiles and depth functions
Get PDFAbstract
Outliers detection based on quantiles and depth functions
Outlier detection plays a crucial role in data analysis and is employed in various domains such as finance, healthcare, and anomaly detection. This thesis presents a novel approach for detecting outliers using quantiles and depth functions, and we apply it to an air quality dataset. Quantiles provide a statistical measure of the distribution of data, while depth functions assess the centrality of observations relative to the entire dataset. Combining these two techniques, we propose a robust
and effective method to identify outliers in multidimensional datasets. Our approach is particularly useful in scenarios where traditional outlier detection methods may be inadequate or fail to capture the complex patterns present in the data. By considering multiple quantiles, we can identify outliers that deviate from different aspects of the data distribution.
Additionally, we incorporate depth functions, which measure the centrality of observations within a dataset, to further refine our
outlier detection process. To evaluate the effectiveness of our approach, we apply it to a real-world air quality dataset.
The data is about the New York Air Quality Measurements of 1973 for five months from May to September recorded daily. It contains 153 observations of 6 variables. By applying our method, we can identify outliers representing unusual air quality patterns, potentially
indicating anomalies or errors in the data collection process.
Our experimental findings support the proposed approach and effectively detect outliers in the air quality dataset.
Compared to traditional outlier detection techniques, our method achieves higher accuracy and provides more detailed in sights into the nature of the outliers.
Furthermore, we show that the identified outliers can be
valuable in understanding the factors contributing to air pollution and in improving the quality of air quality monitoring systems.
The findings of this research contribute to the advancement of
outlier detection methodologies and provide valuable insights for practitioners in identifying and handling outliers in real-world applications
Idle Cinema: A Crisis of Knowledge in Narrative Film
Get PDFAmplified by their use of long takes, slow films dwell on the minutiae of everyday life. Their time is drawn out, and their story-telling inefficient. But as the term “slow” suggests, it has come to represent a manner of storytelling predicated on a delayed, but eventual arrival of an event – an event that ultimately meets the seasoned viewer’s expectation of a “pay off” and in retrospect, justifies the film’s slow build. Even still, the category has become synonymous with the idea that in these films, “nothing happens.” Slow cinema has thus devolved into an often pejorative label that accounts for any and every film that makes use of the long take and tests our patience. As this contentious discourse surrounding slow cinema has clouded a sufficient inquiry into its breadth of aesthetic approaches, the purpose of this thesis lies in identifying what I am calling “idle cinema,” a style of filmmaking that has grown out of the tradition of slowness, but takes its resistance to the narrative efficiency of classical narrative storytelling in new aesthetic directions. In idle cinema, ellipsis gains equal importance to the long take; like slow films, idle films make time visible, but they also make it disappear. As moving images are freed from the economy of “adding up to something,” idle cinema becomes less of a resistance to speed than an expression of a crisis of knowledge
Surrogate Models for Diffusion on Graphs: A High-Dimensional Polynomial Approach
No full textGraphs provide a powerful abstraction for modeling real-world systems, such as social and transportation networks. Of particular significance is the study of diffusion processes on graphs, which is crucial to disciplines spanning biology, engineering and computer science, and to the understanding of phenomena such as disease propagation or the flux of goods and/or people through a transportation network. In this work, we formulate the solution of the diffusion equation on graphs as a parametric model, which gives us insight into the behavior of the diffusion processes by studying how the diffusivity parameters influence the model output. However, accurately simulating the model output of these diffusion processes can be computationally demanding since it involves solving large systems of ordinary differential equations.
To address this challenge, we propose to construct surrogate models able to approximate the state of a graph at a given time from the knowledge of the diffusivity parameters. In particular, we consider recently introduced techniques from high-dimensional approximation based on sparse polynomial expansions, which are known to produce accurate and sample-efficient approximations when the function to be approximated has holomorphic regularity. Hence, to justify our methodology, we present theoretical results showing that solution maps resulting from a certain class of parametric graph diffusion processes are holomorphic. Then, we demonstrate numerically that it is possible to efficiently compute accurate sparse polynomial surrogate models from a few random samples, hence empirically showing the validity of our approach
Tackling the Electric Vehicle-Related Threats to Power Grid Stability
Get PDFDue to the growing threat of climate change, the world’s governments have been encouraging the adoption of Electric Vehicles (EVs). Thus, EV numbers have been growing exponentially and gaining a significant market share. As a result, EV Charging Stations (EVCSs) are being rapidly deployed to satisfy the rising charging demand. These EVCSs are connected to a complex and interconnected system of cyber and physical components. However, without proper security in place, the EV charging load can become a weapon yielded by adversaries to destabilize the power grid. To this end, this thesis examines the security of the EV ecosystem and the impact of EV-based attacks against the grid. The thesis starts by examining the different components and technologies of the EV ecosystem before examining their vulnerabilities. We then demonstrate the greater impact that EV-based attacks can have on the grid as compared to traditional high-wattage smart loads attributed to their bi-directional power flow capabilities and their non-linear nature. We then propose a novel dynamic Load altering (LA) attack strategy that takes advantage of feed-back control theory to induce large frequency instabilities on the grid. To address the serious consequences of such attacks, two district detection methods are devised. The first is a two-tiered detector tailored specifically to be deployed on the EV ecosystem’s Central Management System (CMS) and EVCSs to detect attacks emanating from the EV ecosystem. The second method is a detection scheme from the perspective of the utility aimed at detecting all kinds of LA attacks against the grid. This detector utilizes the grid’s mathematical model to preprocess the collected data and feed it to a Feature Fusion Neural Network (FFNN) that achieves 99.9 % detection accuracy while remaining robust against data poisoning. Finally, we show case the potential strength EVs can introduce into the grid once secured, by developing a robust LA attack mitigation scheme. This mitigation scheme utilizes the EV loads/injections to mitigate the impact of the three types of LA attacks. Additionally, we mathematically model the possible real-life uncertainties that hinder the operation of this mitigation scheme to achieve a robust performance
TASKS Framework for Personalized Task Implementation
Get PDFThis thesis addresses the complexities of task implementation, focusing on personalized barriers encountered in diverse contexts. It introduces the TASKS framework as a novel deductive approach to analyze and overcome these barriers. The framework, grounded in the interplay between tasks and an implementer’s Affect, Skills, Knowledge, and Stress, offers a structured method to identify and address personalized implementation barriers. The thesis validates the framework through three distinct case studies: enhancing designer creativity in design processes, identifying personalized barriers in hypertension self-management, and understanding stakeholder behavior in sustainable product design. Each case study illuminates the framework’s efficacy in different scenarios – from creative design practices, healthcare challenges, to environmental sustainability in product design. The findings demonstrate the framework's versatility in categorizing barriers into emotional, logical, knowledge, and resource categories, and its effectiveness in providing tailored solutions. This research contributes to implementation science by offering a comprehensive tool for understanding and tackling personalized barriers in various task implementations, emphasizing the importance of customizing strategies to individual needs and contexts. The thesis not only enriches our understanding of task implementation but also sets the stage for future research directions, including developing tools for streamlined barrier analysis, exploring dynamic problem-solving methods, and team design and healthcare systems, aiming to enhance the practical applicability and scalability of the TASKS framework
Developments in Thieno[3,2-c]isoquinoline Analogues as Potential Anti-Cancer Agents
No full textChemotherapies are the leading treatment for late-stage cancers. Despite their efficacy, these treatments are known to have harsh side effects caused by poor selectivity for cancerous cells over healthy cells. For this reason, the exploration of molecular targets specific to cancer cells is crucial for improving chemotherapies.
In preliminary studies, thienoisoquinolines have shown selective biological activity against lung cancer cells. Microscopy studies of cells treated with a leading thienoisoquinoline derivative exhibited arrested mitosis and disruptions within the microtubules. Additional experimentation via an in vitro microtubule polymerization assay and competition studies indicate that this lead compound has high binding affinity to microtubules at the colchicine site.
With these results, we aimed to further optimize the lead compound 75 (C75) to generate more drug-like molecules and confirm its mode of action through substrate alteration. In order to facilitate candidate modification, new synthetic approaches have been developed for enhanced modularity. This has generated new derivatives for structure-activity relationship studies, as well as provided access to wider range of derivatives for future pull-down assays, with the aim of identifying the specific biomolecular target of thienoisoquinoline derivatives with anti-cancer activity. Notably, the benzylic position is a major contributor to degradation of these molecules in biological assays. A newly devised synthetic route overcame the challenges in functionalization of this position, allowing for the production of more drug-like derivatives which will hopefully have increased biological activity. This was done in conjunction with kinetic solubility measurements and degradation studies in order to streamline the production of more efficacious compounds
Strategizing Crude Oil Market Dynamics: Using Deep Learning Predictive Models and the Influence of Parameter Optimization
No full textCrude oil, originating from organic material deposited millions of years ago, serves as the raw material for products like gasoline, diesel, and jet fuel, highlighting its crucial role in modern industry and daily life. Predicting crude oil prices is vital for supply chain managers making operational decisions such as purchasing, production, and transportation. This research aims to predict oil prices to reduce operational costs, increase profit, and enhance competitive advantage. We employed deep learning models to capture the complex, nonlinear characteristics of crude oil price dynamics, utilizing a hyperparameter optimization framework with Bayesian optimization (Optuna) for improved convergence, reduced overfitting, and higher accuracy. In a world shaped by technological breakthroughs, geopolitical intricacies, and economic pressures, precise prediction of oil prices is challenging but essential. Advanced machine learning techniques like Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU) were utilized. This study used datasets including WTI and BRENT to explore neural network’s ability to understand complex market linkages. Three comparative metrics—RMSE, MAD, and MAPE—ensured result reliability and application across various domains. Forecasts were conducted across three time horizons: daily, weekly, and monthly, each crucial for different stakeholders such as day traders, logistical planners, and strategic decision-makers. Daily forecasts navigate immediate market volatility, weekly forecasts inform logistical and operational adjustments, and monthly forecasts align with long-term strategic planning.
Finally, we compared traditional statistical models with the best deep learning models for both Brent and WTI crude oil using RMSE, MAD, and MAPE to assess the robustness of the deep learning approaches
Produire l’archive sensible grâce au dispositif cinématographique : Recherche sur les technologies de la mémoire
No full textEn tressant réflexions théoriques, anecdotes personnelles et pratique artistique, ce mémoire en recherche-création explore les tensions entre la mémoire humaine et la mémoire extériorisée, dite exo-somatique. L’auteure y interroge les limites des systèmes d’indexation de la mémoire numérique ainsi que les enjeux éthiques et affectifs liés à l’extériorisation de la mémoire, en s’appuyant sur les archives de la famille « choisie » dans laquelle elle a grandi.
Inspirée par la transgression de la linéarité des récits et par les pratiques d’archivage analogique, l’auteure s’attache à développer une méthodologie d’archivage vernaculaire et sensible, capable de réactiver les anecdotes intimes et de mettre en lumière les liens affectifs souvent invisibles dans les structures institutionnelles de la mémoire collective. Cette démarche invite à repenser la quête de permanence propre à la mémoire numérique et exo-somatique, sans pour autant freiner l’aspect transformateur de nos constructions identitaires.
Accompagné d’un catalogue filmique explorant les thèmes du deuil, de la maternité et de la sororité, ce mémoire propose une revalorisation de la subjectivité et de la transformation comme réponses aux dérives des pratiques mnémoniques contemporaines
Minimizing Communication Costs and Dropped Tasks via Dynamic Human Digital Twin Placement in Mobile Edge Computing
No full textWith the advent of 6G networks, digital twin (DT) systems have become critical for real-time monitoring, decision-making, and control across various sectors. A digital twin is a virtual representation of a physical twin (PT), enabling continuous interaction and data exchange. However, real-time communication between the DT and PT incurs variable costs as users change locations, significantly affecting system efficiency and user experience. In mobile environments, minimizing these communication costs is essential for maintaining DT performance, as increased latency and resource demands arise with user mobility. To address this, we consider three communication strategies between the user and their digital twin: direct communication, multi-hop communication, and digital twin migration. Consequently, an optimal dynamic placement strategy for DTs on edge servers is crucial to reducing communication overhead while ensuring responsiveness. This work introduces an optimization framework leveraging Lyapunov optimization to model and minimize communication costs between the Human digital twin (HDT) and PT, considering task drops during twin migration. The proposed solution dynamically adapts to user movements and network conditions, ensuring efficient real-time interactions with minimal costs. Evaluation results demonstrate
the effectiveness of our approach in reducing communication costs and task drops while maintaining data exchange quality and reliability in 6G-enabled DT systems
Mapping Cadastral Records as Evidence of Colonial Land Theft: A Kanehsata'kehró:non-backed Investigation into the Archives of Quebec’s Sulpician Priests
Get PDFDuring the 18th and 19th centuries, the Saint Laurent valley, Quebec, became increasingly surveyed as France, England, and later Canada, claimed land and removed the Indigenous people who inhabited it. In Kanehsatà:ke (also known as Oka), the rights and interests of the Kanehsata'kehró:non were never included in the mapping process and they were continually denied rights and recognition. This research aims to reclaim land surveying by repurposing cadasters and land patents, originally used by religious and governmental institutions, for the Kanehsata'kehró:non to talk about land that was historically taken from them. While cadastral mapping has been critiqued as a tool of colonial surveillance and control, I flip the traditional power dynamic by mapping where settlers live and how they acquired land. This shift demonstrates that cadastral maps and archives can function not only as instruments of control but also as tools for liberation.
To achieve this, I outline my mapping process in three parts. First, I review how maps and archives have been reshaped through counter-practices. I interview an art historian, a Citizen Potawatomi cartographer, a Kiowa journalist, and a historian recycling colonial land records in critical Indigenous mapping projects. These conversations provide a broad perspective on the intersection of maps and archives, highlighting unique insights, ethical considerations, and challenges in studying them together. Second, with the guidance of a Kanehsata'kehró:non Land Defender, I designed and led an investigation into the archives of the Sulpician priests. I developed a methodology to repurpose these “counter-archives” into a geospatial database for use in Geographic Information Systems (GIS). Finally, I explored future uses of my mapping project by tracking how various stakeholders plan to use the map to advance their own intentions and priorities.
This approach not only demonstrates how colonial maps and archives provide different perspectives on historical events but also how to leverage these differences to amplify the voices of those who have been marginalized or silenced. By turning my attention towards the increasing Indigenous led research in colonial archives, I trace the emergence of new kinds of mapping projects and explore innovative uses of cadastral data to support political change. Recognizing colonial archives as a mappable dataset for Indigenous benefit broadens the scope of settler colonial studies in Quebec, a field that often plays second fiddle compared to other provinces. Ultimately, while mapping archives can serve as a basis for contemporary debates on land rights, reparations, and Indigenous sovereignty, the true essence of the work lies in cultivating and maintaining the relationships that bring these maps to life