18701 research outputs found
Sort by
Robust Federated Learning: Server and Client Defenses Against Poisoning Attacks
Federated Learning (FL) has been rapidly evolving and gaining popularity in recent years due to its privacy-preserving features, among other advantages. Nevertheless, the exchange of model updates and gradients in this architecture provides new attack surfaces for malicious users of the network which may jeopardize the global model performance, mainly through poisoning attacks. This dissertation tackles this security challenge by proposing three distinct defense mechanisms that safeguard FL against various types of poisoning attacks. The first part of this dissertation focuses on challenges in mitigating poisoning attacks from the client-side. This is particularly important in cases where the server is not trusted, or when the FL architecture is serverless, also known as Decentralized FL (DFL). To address this challenge, adversarial training and label noise analysis are integrated to design a new defense mechanism that effectively mitigates label poisoning attacks from the client-side. The second part of this dissertation shifts the focus to server-side defense. To mitigate poisoning attacks from the server-side, a novel evaluation mechanism is proposed that identifies and exclude such adversaries on the server prior to aggregation. This method injects random noise into client model reconstructions and analyzes their layer activations. An autoencoder, trained on the global model's activation outputs, detects abnormal patterns in client models, eliminating the need for a central test dataset for evaluation. Third, a defense mechanism is introduced for DFL setups, where aggregation is distributed across participants, and the aggregating node is not trusted. Using blockchain technology, anomaly detection, and concept drift analysis, this method scores aggregators based on their past behavior and continuously monitors the statistical dependence between client updates and aggregated models. This ensures trustworthy aggregation even in a serverless environment. In addition to the proposed methodologies, this dissertation studies an industrial application of FL, focusing on Advanced Persistent Threat (APT) detection in cyber-physical systems. FL is introduced as a solution to address the scarcity of APT data, enabling organizations to collaboratively train a robust global model for APT triage without exposing sensitive data
The Tax Effects of Intergenerational Transfers: A Quantitative Exercise.
Economists predict transfer of wealth from one generation to the next due to increased global wealth. This transfer of wealth is predicted to widen already existing poverty gaps in some economies. Some economists have suggested the taxation of intergenerational transfers in a bid to increase government revenue and reduce the widening gaps between the poor and the rich. This study captures the key mechanisms at work in assessing the labour earnings reaction to changes in unearned income, emanating from gifts and bequests that heirs receive. Three theoretical models are developed to understand the labor earnings reaction with varying unearned income compositions. The tax effects on the transfer of intergenerational wealth are then decomposed. Further, the study uses specific parameters from the Canadian economy with other assumptions to quantitatively investigate the three theoretical models and examine the differences in utility, consumption and labour supply when gift and bequest taxes are applied to unearned income. The study concludes that heirs experience high levels of utility when gift and bequest taxes are applied by the government. Heirs tend to reduce labour hours supplied upon the receipt and taxation of transfers. An optimal combination of income and transfer taxes is necessary to maintain a productive work force in an economy
Love and Difference: Refuting the ‘Risk-Free’ Conception of Romance
Love and Difference: Refuting the ‘Risk-Free’ Conception of Romance
Love, as a philosophical topic, has a convoluted history. Modern considerations of love, which inherit this history, oscillate within a spectrum that ranges from pessimistic conceptions of love as merely instrumental reproductive sexuality (attributable to Schopenhauer), to an ecstatic fusion that presents love as the harmony of two into one (expressed in the work of Simone de Beauvoir). Each of these positions can be characterized as difference-evading, escapist, and ‘risk-free’ approaches to love, which, Alain Badiou claims, denies the necessary elements that make love possible; namely, a commitment to chance, the experience of vulnerability, and perseverance and fidelity in love. By way of Badiou, this paper attempts a refutation of Schopenhauer’s pessimistic rejection of love, and, following this, attempts to think beyond the relationship of insecurity and dependency that Beauvoir associates with the plight of the ‘woman-in-love.’ As I intend to show, Badiou’s theory of love, which is existential in nature, denies the escapism that is inherent to riskless love. Badiou’s solution to both the pessimistic and fusional hypotheses is to present authentic love as a ‘truth procedure’ wherein lovers form a common subject known as “the Two” which operates beyond the non-connected and incommensurable positions Badiou calls ‘man’ and ‘woman.’ This re-invented conception of love fills up (supplements) and compensates this non-related Two, who, together, experience a truth made possible by difference—a pursuit which requires the continuous declaration and affirmation that the procedure is worth the risk.
Keywords: difference, fidelity, fusion, love, pessimism, risk, the Two, truth-procedure.vv
The Coastie Initiative
Coastline change is an ever-increasing area of interest with a multitude of research opportunities. The Coastie Initiative was developed with Parks Canada to crowdsource the collection of regular imagery of coastal morphology, building a large dataset for tracking shoreline change. Citizens can partake in this growing program by submitting a 'Coastie' through the web-based platform. At every site, a phone cradle, designed to standardize images, is accompanied by an informational panel and a QR code. Visitors can use the QR code to quickly access the submission wizard. After submission, the image is saved and awaits classification by researchers. The intended development of the project includes an achievement system to promote user retention, an automated classification system to eliminate manual labelling, and an ocean literacy stream as an effort to educate and inform the public. The initiative was launched during the fall of 2021 with 5 locations. The Coastie program is a collaboration that has developed from the global CoastSnap Community Beach Monitoring movement that began in Australia in 2017. Using the early prototype dataset, coastal change can be quantified from repeat geo-rectified images. These capture deposition/erosion of the beach and dune topography in response to ambient environmental conditions or higher energy storm events. This dataset will serve as a baseline to monitor how coastal systems respond to climate change, including sea level rise, change in storm activity, and reduction in sea ice coverage. The Coastie Initiative allows citizen scientists to influence coastline exploration participating alongside motivated researchers
The Effect of Exercise Training on Re-establishment of the Extracellular Matrix Following Muscle Damage
Skeletal muscle is a highly versatile tissue that has the ability to move and perform a wide range of tasks, but through these movements, can be damaged. When skeletal muscle is damaged, it has the remarkable ability to regenerate back to its original function. The most important step in skeletal muscle regeneration involves the proliferation of satellite cells, which is the rapid increase in the number muscle stem cells. These cells fuse with the damaged muscle fibre and fuse with one another to repair the damage. Part of this regeneration process involves the remodelling of the extracellular matrix (ECM), which is the coating that surrounds each muscle fibre. The ECM is essential to the muscle fibre's function, as it serves important structural and chemical signalling roles. Satellite cells also regulate the repair process of the ECM and several studies demonstrate exercise as a method to improve the remodelling of the ECM. Of the different components of the ECM, Collagen I, Collagen IV, Fibronectin, and Matrix Metalloporteinases (MMP) have been found to be the ECM proteins that respond to exercise the most. To look at these changes, 12-week-old mice were exercised for two weeks following cardiotoxin-induced damage. Changes in the ECM were be examined using immunohistochemistry techniques at several time points (3, 5, 7, and 14 days), and preliminary results reveal that a greater expression of Collagen 1 is evident in the exercise group. To date, no study has looked at how the ECM is remodelled following cardio-toxin induced damaged. Results from this study can help gain a better understanding of the effects of exercise training on regenerating skeletal muscle
Towards the Development of a Comprehensive Tissue Mass Prediction Tool for Living Men and Women
Viable, Healthy and Safe CommunitiesAccurate tissue mass estimates from living people are required to better model the response of the musculoskeletal system following impulsive impact events [1]. To date, prediction equations have been developed to estimate the soft and rigid tissue masses of all body segments from surface anthropometric measures and validated against actual tissue masses obtained from Dual-energy X-ray Absorptiometry scans (DXA) [2-5]. Good to excellent reliability has been reported for all anthropometric measures (lengths, circumferences, breadths, skinfolds) [6-7], and tissue masses obtained manually from the DXA scans [8-9]. Despite the accuracy and reliability of these equations, this process of obtaining tissue masses from living people is limited by: inconsistencies in variable names and measurement definitions between studies; susceptibility of errors when transferring written measurements to digital form, and; substantial time spent on an overly complex manual process. Therefore, the intention of the current study is to address the limitations of the current process used to estimate segment-specific soft and rigid tissue masses by developing a tool that incorporates existing equations into one location (with unified variable names and measurement definitions), automating, and linking the data recording and analysis processes. Existing tissue mass prediction equations will be compiled on a tablet device for the segments of the upper and lower extremities, as well as the head, neck, trunk and pelvis. A simple user-interface is being developed using MATLAB® (R2017a) to enable input of anthropometric measurements using an interactive human manikin. Measurement inputs will be linked to the prediction equations to streamline the analysis process. Additionally, variable names and measurement definitions from the different equations have been reconciled to increase usability. This tool will facilitate the prediction of soft and rigid tissue masses from anthropometric measurements by increasing data collection efficiency, reducing data transfer errors, and enabling large population studies of tissue masses for all body segments
Creating a Toolbox for the Asymmetric Alkylation of Isoindigo Based Polymers
['UNSDG 12: Responsible Consumption and Production (https://sdgs.un.org/goals/goal12)']Viable, Healthy and Safe CommunitiesDue to their low cost, easy modification, and solution processability, organic semiconducting polymers have gained increasing interest for their applications in a wide variety of organic electronic devices including; organic field effect transistors (OFETs), organic light emitting (OLEDs) diodes, and organic photovoltaics (OPVs). Side chain engineering is an important factor in modulating the intrinsic properties of these semiconducting polymers in terms of molecular stacking, and intermolecular interactions, as well as thin film morphology, all contributing to device performance. A relatively new area of interest for polymer modification is asymmetric side chain synthesis. This work focuses on creating a toolbox for the asymmetric alkylation of Isoindigo-Bithiophene donor-acceptor type polymers. The effects of asymmetry will be determined by keeping one alkyl chain constant and then successively moving the branching point of the second alkyl chain further from the polymers' backbone. To determine polymer performance, top-gate bottom-contact devices will be fabricated to test asymmetry-mobility relationships
Recreational Cannabis in Ontario: An Evaluation of the First Year of Legalization
Recreational cannabis was legalized in Canada on October 17, 2018 and has created new business and income for the country. Focusing on Ontario, this paper analyzes how the province failed to benefit from legalization during its first year. As a result of the change in provincial government—which led to ineffective sales models, delays, and financial losses—Ontario has failed in cannabis legalization. Using the theory of federalism, this paper explores the division of power within Canada’s levels of government and its impact on cannabis legalization and sales. Further, this paper includes a case study on Alberta and its free-market success with alcohol and cannabis to assess Ontario’s rollout of cannabis sales and argues that, despite the failures of its first year, Ontario has potential for success in the future
Educational Outcomes of Program Effectiveness of a Postsecondary Concurrent Education (Pre-Service Teaching) Student-Mentoring Program
Presenters: Kendra Hart (live participation tentative on B.Ed. practicum scheduling) Holly Mariconda Deborah LazeThe submission is attached in the Digital Files section
Responsible Conduct of Research & Tri-Council Policy Statement 2 Update
Update on the new policy and mandate from the Secretariat on Responsible Conduct of Research regarding research ethics for the three federal research granting agencies