21793 research outputs found
Sort by
Critiquing Canadian Multicultural Discourse Through the Art of the Filipino Diaspora
Get PDFCanada is touted to be a model of harmonious multiculturalism vis-a-vis other multicultural societies. However, implicit in Canadian multicultural policies are heteronormative and inequitable frameworks that shape the prevailing policies toward diasporic groups. Through discourse analysis and decolonial aesthetic observation, this study explores how Canadian multicultural discourse can be challenged through the art of Filipino-Canadian artists, and what the latter reveals about the experiences of Filipino immigrants in Canada. By performing a discourse analysis on the 1988 Canadian Multiculturalism Act and the 2017 Canada Council for the Arts Equity policy, in combination with the examination of the work of Toronto-based Filipino-Canadian artist Azia Jonelle, this study will seek to answer the following question: how does the work of Filipino-Canadian diasporic artists challenge normative ideas of Canadian multiculturalism? In the end, this paper will identify the predominant issues of Canadian multiculturalism by examining the artist’s particular representations vis-a-vis normative multicultural discourses found within the aforementioned policies, and thus determine how the particular experience of the Filipino-Canadian diasporic subject is disarticulated and rearticulated amidst Canada’s multicultural landscape
Deep Reinforcement Learning-based Automated Penetration Testing for Active Distribution Networks
Get PDFThe smart grid is a highly complex cyber-physical system of heterogeneous components with sensory, control, computation, and communication. Due to its complexity, dimensionality, uncertainty, and strong cyber-physical coupling, manually identifying critical vulnerabilities against cyberattacks at infrastructure levels has proven to be challenging. In the information and communication technology (ICT) industry, penetration testing (PT) has demonstrated its efficacy in pinpointing vulnerabilities within information systems through authorized cyberattacks. Building upon the principles of PT, this study delves into exploring effective and efficient PT approaches to discover vulnerabilities for active distribution networks (ADNs) of smart grids based on deep reinforcement learning (DRL) methods.
To overcome the poor efficiency and non-comprehensiveness of common PT in identifying vulnerabilities for an ADN caused by its complex structure and strong cyber-physical coupling, we first propose a DRL-based PT framework and formulate the PT as a Markov decision process (MDP) specifically for the industrial control networks of the ADN. This framework comprehensively considers cyber-physical coupling, realistic cyberattacks, and the physical impacts of ADNs. The framework is applied to model a replay attack scheme on an ADN as the study case, which aims to identify critical attack paths that lead to system voltage violations. Additionally, a co-simulation platform named GridBattleSim was developed specifically for DRL-based PT on ADNs, integrating dedicated simulators for different parts of the ADN. The simulation results show the efficacy of DRL-based PT in learning optimal attack paths under varying system conditions and different levels of attack difficulty.
To overcome the limited observability in practical PT scenarios, a partially observable Markov decision process (POMDP) formulation is proposed, which allows the PT agent to learn PT policies under partially observable conditions. To solve the POMDP and obtain the optimal PT policy, we apply the physical model of the ADN to estimate its full state based on the local observable data captured by the PT agent and then transform the POMDP to an MDP that can be solved by DRL.
Furthermore, to address the sparse reward issue, improve the generalization of reward functions, and improve the interpretability of DRL-based PT, a knowledge-informed AutoPT framework (RM-PT) is proposed, which incorporates cybersecurity domain knowledge based on Reward Machine (RM). We use the lateral movement of PT on ANDs as a case study, where two RMs are designed based on MITRE ATT&CK knowledge base as two PT guidelines. Finally, the deep Q-learning with RM (DQRM) algorithm is applied to train the PT policies. The proposed RM-PT is evaluated under the CyberBattleSim platform. The experimental results show that the knowledge-informed PT exhibits a higher training efficiency compared to the PT without knowledge embedding. Furthermore, RMs that incorporate more detailed domain knowledge exhibit superior PT performance compared to RMs with simpler knowledge.
Finally, we also discuss the future directions of this study in terms of domain knowledge integration for AI-powered PT. We anticipate that the methodologies and findings presented in this study can inspire efforts in securing critical infrastructure and closing research gaps for the cybersecurity of smart grids
Detecting Man-in-the-Middle Attacks in Cellular Networks Using Generative Machine Learning Based Indistinguishable Probing
No full textMan-in-the-Middle (MiTM) attacks launched against the user equipment (UE) and the core networks represent a well-known security threat to the proper operation of telecommunication networks from earlier generations. Such attacks can potentially downgrade the security capabilities or degrade the quality of service to either prepare the ground for subsequent attacks or cause denial of service to the end users. Existing solutions focus on detecting specific MiTM attacks (e.g., utilizing false base stations or malicious UEs) while relying on the Radio Access Network (RAN) and considering it as a trusted entity. With the advent of virtualization of the RAN (i.e., vRAN) and the opening of the interfaces ( i.e., Open RAN), an attacker can potentially infect the vRAN (e.g., due to lateral movement) and launch MiTM attacks, making existing solutions not sufficient to cover this new attack vector. In this thesis, we therefore aim at proposing a verification solution, named Orion, for detecting the presence of a stealthy and smart MiTM attacker between UE and the core. For this purpose, our main ideas are (a) to leverage generative ML to continuously generate indistinguishable (to evade an MiTM attacker) and synchronous (to enable verification at both UE and core with minimal communication overhead) probing messages and (b) to engage both UE and the core in the verification (unlike existing works that only focus on one of them) of the probing messages to detect MiTM attacks. We implement Orion and integrate it into our testbed based on the OpenAirInterface (a popular open-source project used to test 4G and 5G), deployed on a Kubernetes cluster on the Amazon Elastic Kubernetes Service. To show case the feasibility, efficiency, and effectiveness of Orion, we use a MiTM attack scenario targeting specific unprotected control plane messages carrying radio and security capabilities. Our experiments show that Orion can detect such MiTM attacks using 14 probing (virtual) UEs in only 40 seconds (during which only two regular messages are compromised) and requiring about 60 probes on average while achieving an indistinguishability rate of 95% under different adversarial strategies. Additionally, Orion requires less than 350MB of memory, up to 100% of CPU for a little less than 17 minutes of time to generate probing messages required for each day of probing
Early Detection of Emerging Technologies Using Machine Learning and Burst Detection
Get PDFCertainly, the impact of emerging technologies is changing our world and how we live, shaping our future significantly. In the constantly evolving landscape of these technologies, which attracts substantial yearly investments, spotting these trends early on is both challenging and expensive. However, applying an emerging technology detection method in an effective and efficient way is considered a challenging task for many stakeholders. In this thesis, we address these problems through applying a method to predict potential emerging technologies in the case study field of Artificial Intelligence (AI). Using this method may help policymakers to identify potential emerging technologies early in a more systematic way with little manual intervention. In the proposed method, using burst detection, machine learning, and deep learning, we attempt to predict the future sustaining emerging technologies. We applied the methodology by four methods, namely Random Forest, Gradient Boosting, XGBoost, and Multi-Layer Perceptron (MLP). Results showed that the method was successful in its tasks. The method had the Area under the Curve (AUC) rate of more than 75% to accurately predict the sustainability of the potential emerging technologies. More specifically, applying the MLP method showed the ability to increase the AUC rate and recall metric as the most important metrics of our work. In summary, this approach carries both theoretical and practical significance. Theoretically, the exploration of novel combinations, such as integrating deep learning and burst detection methods or employing transformers, offers researchers fresh insights into the challenge of detecting emergence. On the practical front, the application of methods providing high accuracy rates in machine learning methods empowers stakeholders to implement these methods effectively in practical scenarios
Focused ultrasound-guided delivery of microRNA-126 to endothelial cells in in vitro and ex vivo models
Get PDFThe prevalence of cardiovascular diseases, such as ischemia, underscores the need for innovative therapeutic strategies. Recent advances in the field of focused ultrasound offer a non-surgical, targeted, and promising technology to treat various life-threatening diseases, including brain disorders, inoperable cancers and some vascular diseases. This approach harnesses the therapeutic potential of ultrasound-stimulated microbubbles for the modulation of cellular and vascular permeability to guide the delivery of therapeutics. More recently, gene therapy has shown great potential as a less invasive approach in contrast to surgical interventions. In the context of cardiovascular diseases, microRNA-126 is a key target for therapeutic interventions, as it is abundant in endothelial cells that line blood vessels, and plays a pivotal role in promoting angiogenesis. This thesis explores the use of ultrasound-stimulated microbubbles as a non-viral and targeted approach for microRNA-126 delivery to endothelial cells in both in vitro and ex vivo models. First, I designed and characterized a cationic microbubble formulation that can carry a microRNA-126 payload on its surface. I then developed an ultrasound regimen to safely deliver microRNA-126 to endothelial cell suspensions and demonstrated its effect on blood vessel formation in vitro. My results indicate the increase of microRNA-126 in endothelial cells result in the modulation of key downstream proteins, notably PIK3R2 and SPRED1, and improves angiogenesis. Building on these findings, I then developed a more complex vascular model to study ultrasound-guided microRNA-126 delivery by isolating rat mesenteric arteries. This model allowed me to replicate a more physiologically relevant vascular environment ex vivo. By incorporating factors such as intralumenal pressure and fluid flow, this study investigated the effect of focused ultrasound on the vasoconstriction and vasodilation of an artery. My findings revealed a positive relationship between ultrasound-mediated cell permeabilization and increased flow velocity, as well as an inverse relationship between the levels of microRNA-126 delivered with increased flow velocity. These results suggest that understanding hemodynamic conditions in specific anatomic regions could enhance the effectiveness of gene delivery. Overall, this thesis highlights the potential to deliver microRNA-126 using focused ultrasound and microbubbles while minimizing cellular and vascular viability, with anticipation for its application in gene therapy for cardiovascular diseases
Automated Planning and Scheduling Method for Modular Construction Manufacturing
Get PDFModular construction is a promising alternative to conventional construction; offering improved productivity, high-quality end products, and reduced labour requirements. To realize these benefits, sequencing module components during prefabrication process in a manner that ensures efficient allocation and utilization of labor resources at workstations is essential. However, one of the significant challenges in modular construction manufacturing (MCM) is that it follows a make-to-order process, resulting in customized module components. This customization leads to variations in the design specifications of module components, causing different processing times at each workstation. These imbalances in production lines result in increasing the waiting time for module components between the workstations, ultimately extending the makespan. This poses a challenge for production line managers, requiring frequent adjustments to plans and schedules related to the sequencing of module components at workstations using conventional methods.
To address these challenges, this thesis introduces a framework composed of three modules: (i) a simulation-based statistical method for planning in modular construction; (ii) a deep neural network (DNN)-based method for predicting production process times; and (iii) a hybrid optimization technique for scheduling in modular construction. In the first module, a simulation based statistical method is developed to plan the sequencing of module fabrication and the allocation of workers at workstations. The method encompasses data collection process to obtain historical/near real-time data and identification of significant impact factors affecting process times at workstations along the production line. In the second module, a newly developed method for predicting processing time at each workstation is introduced utilizing Deep Neural Network (DNN), Artificial Neural Network (ANN), and Multiple Linear Regression (MLR) for predicting production process time spent at each workstation in a manufacturing plant. The third module focuses on planning and scheduling method that ensures optimal sequencing of module components at workstations using Genetic Algorithm (GA), Simulated Annealing (SA), and Hybrid Genetic Algorithm Simulated Annealing (HGASA).
Two case studies were analyzed to demonstrate the use of the developed methods and test their performance. The first case is of a light gauge steel (LGS) wall panel production line operated by a modular fabricator in Edmonton, Canada, and the second is of a wood-based semi-automated wall panel production line also in Edmonton, Canada. These cases involve the production of 200 wall panels in the first case and 39703 wall panels in the second at various workstations along the production line. The simulation-based statistical method developed in the first module yielded 89.39% accuracy in prediction of process time and indicate a 44.42 hr duration to produce 309 wall panels with regards to first case. The results of the second case showing process time predictive method developed in the second module for most workstations had a mean absolute error (MAE) of under 2.50 minutes, with symmetric mean absolute percentage error (SMAPE) ranging between 22 % - 28%, respectively. The developed scheduling method of the third module provided an optimal sequence of wall panels for prefabrication, minimizing makespan. As a result, the hybrid optimization reduces makespan to 105.63 hr from those generated by GA (138.08 hr) and SA (108.06 hr)
Advanced Anomaly Detection and Quality Control in PCB Manufacturing
Get PDFPrinted Circuit Boards (PCBs) are essential in electronic devices, where even minor defects
can signifcantly impact products and the environment. Thus, rigorous quality control is imperative
in PCB manufacturing. This thesis tackles critical challenges by developing robust strategies for
defect detection and accurately predicting repair needs. It begins with an extensive background on
current fault detection and repair strategies. Central to this study is the use of advanced machine
learning (ML) and deep learning (DL) techniques to enhance the accuracy of the PCB labeling process, integrating data from Solder Paste Inspection (SPI) and Automatic Optical Inspection (AOI)
datasets. The research is structured into distinct phases, each addressing different aspects of the
PCB manufacturing process. The initial phase focuses on improving the prediction of human inspection labels using advanced ML and DL techniques, particularly addressing the challenges of
imbalanced datasets with synthetic data augmentation techniques like Synthetic Minority Oversampling Technique (SMOTE) and Conditional Tabular Generative Adversarial Network (CTGAN).
The subsequent phase expands ML algorithms to refne the process of assigning ”RepairLabel” to
PCBs, incorporating ensemble methods and sophisticated feature engineering to boost accuracy and
effciency. The proposed methods have shown promising results, demonstrating their substantial potential for real-world applications. The thesis concludes with a summary of fndings and discusses
the implications for PCB manufacturing. It also outlines potential directions for future research,
suggesting further enhancements in fault detection techniques and the development of more intelligent and effcient systems
Multi-Encoder Semantic Communication for Human Digital Twin Synchronization
Get PDFHuman Digital Twin (HDT) concept introduces an innovative framework for creating digital
counterparts of individuals, enabling real-time synchronization between the physical twin (PT) and
the virtual twin (VT). This PT-VT synchronization underpins various human-centered services but
demands significant data processing and efficient communication resource allocation, particularly
in resource-constrained environments. Semantic communication has emerged as a promising al-
ternative to traditional data-driven methods; however, single-encoder models struggle to meet the
diverse and dynamic requirements of HDT applications.
To address these challenges, this thesis presents a multi-encoder semantic communication frame-
work that adaptively allocates resources—such as bandwidth, computational power, and processing
frequency—based on application-specific needs. The short-term optimization problem is formulated
as a mixed-integer nonlinear programming (MINLP) problem and solved using a genetic algorithm
(GA). Simulation results demonstrate that the proposed multi-encoder model significantly improves
synchronization quality and power efficiency, outperforming traditional single-encoder models in
terms of accuracy, latency, and resource utilization.
To achieve a balance between immediate performance and long-term objectives (e.g., queue sta-
bility, sustainable energy usage, and high throughput), the thesis explores the long-term optimiza-
tion problem, formulated as a Markov Decision Process (MDP) and solved using Lyapunov-assisted
deep reinforcement learning. This adaptable and scalable approach positions the multi-encoder se-
mantic communication model as a highly efficient solution for future HDT applications
Machine Learning-Driven Strategies for Efficient Traffic Congestion Management
Get PDFUrban regions have a notable obstacle in the form of traffic congestion, which results in longer trip durations, higher fuel usage, and increased pollution levels. This study aims to tackle this issue by presenting a three step approach. The first approach uses Machine learning for Proactive Traffic Congestion Prediction. We explore multiple machine learning algorithms, such as Long Short-Term Memory (LSTM), Decision Tree (DT), Recurrent Neural Network (RNN), AutoRegressive Integrated Moving Average (ARIMA), and Seasonal ARIMA (SARIMA), to predict traffic congestion levels in different zones of the Montreal area. The results indicate that the Decision Tree approach surpasses other algorithms, attaining faster convergence, lower loss values, and a considerably higher R2 score. After predicting the congestion using one of the prediction algorithms mentioned above, metaheuristic optimization algorithms are used to find near optimal cycle time for each
traffic light. In step 2 Enhanced Bat Algorithm (EBAT) is proposed to adaptively modify traffic signal timings based on expected congestion levels. The EBAT algorithm utilizes adaptive parameter adjustment and guided exploration techniques that are dependent on the expected congestion. This results in enhanced performance when compared to the conventional Bat Algorithm. We conduct a comparative analysis of EBAT with various meta-heuristics, namely Particle Swarm Optimization
(PSO), Cuckoo Search (CS), JAYA, Sine Cosine Optimization (SCO), and Harris Haws Optimization (HHO). The evaluation considers three scenarios: fixed-time traffic lights (baseline), dynamic traffic lights without prediction, and dynamic traffic lights with predicted congestion. The results demonstrate that EBAT yields substantial enhancements in both the rate at which convergence is achieved and the quality of the solutions, as compared to fixed and non-predictive scenarios. The second approach is using Multilevel Learning for Enhanced Prediction Accuracy. The precision of predicting traffic congestion depends on the ability to recognize and manage abnormal traffic patterns, especially in highly populated regions. Traditional prediction methods are vulnerable to these anomalies, as they frequently do not handle or clean the data. This can result in inaccurate forecasts, as the data may encompass anomalous occurrences such as accidents or unforeseen road closures,
which can greatly distort the underlying trends. The study presents a novel and creative strategy to learning at several levels, which combines anomaly detection and ensemble learning approaches to
tackle this problem. Anomaly detection techniques are used to find abnormal patterns within the data, which is then followed by the process of data cleansing. First, a set of initial learner models are trained. The top-performing models are then selected for an ensemble procedure, which involves combining their predictions through stacking and voting. Evaluated using a real-world Montreal traffic dataset, this multilevel methodology demonstrates higher prediction accuracy when compared to traditional approaches. The dataset is subjected to preprocessing techniques, such as windowing, to transform time-series data into frequency patterns in order to create a more generalized model. To leverage the detected anomalies, we utilized clustering algorithms, specifically K-Means and Hierarchical Clustering, to segment these anomalies. Each clustering algorithm was used to determine the optimal number of clusters. Subsequently, we characterized these clusters through detailed visualization and mapped them according to their unique characteristics. This approach not only identifies traffic anomalies effectively but also provides a comprehensive understanding of their spatial and temporal distributions, which is crucial for traffic management and urban planning. In summary, this study showcases the efficacy of a synergistic method that combines machine learning
for proactive prediction of traffic congestion with metaheuristics for dynamic regulation of traffic lights. This method has the capacity to mitigate urban traffic congestion and enhance traffic flow efficiency. In addition, the use of a multilevel learning strategy to improve forecast accuracy is a noteworthy contribution to intelligent transportation systems. An application for city of Montreal is provided
Models and Algorithms for Concept Drift Detection, Adaptation, and Resolution in Streaming Data
Get PDFThe evolution of streaming data during long periods of time presents significant challenges for maintaining the accuracy and efficiency of predictive models due to concept drift — where changes in data distribution can lead to performance degradation. In this research, we study the problems of concept drift detection (CDD) and adaptation (CDA). Unlike traditional approaches that treat CDD and CDA independently and in isolation, often under non-streaming, static conditions, we propose a novel methodology based on multivariate vector error-correction analysis of feature importance measures (FIMs). The FIMs provided a solid foundation that allowed us to reformulate concept drift detection and adaptation in streaming data.
We additionally introduce, formalize, and develop the notion of concept drift resolution (CDR) as an innovative model preference technique. This solution further enhances the overall performance by effectively using multiple models undergoing concept drift, including the main learner and the proposed CDA model. The results of our numerous experiments and analyses indicate that the proposed CDD method significantly reduces computation time, particularly in applications experiencing abrupt drifts, while our CDA model delivers notable improvements in prediction accuracy and F1 score on both gradual drift and abrupt drift datasets, outperforming existing methods on varying drift rates and characteristics of concept drift.
By utilizing FIMs as a common basis, we develop a unified framework that integrates CDD, CDA, and CDR tasks, thus bridging the gap between detection and adaptation. Extensive experiments validate the effectiveness of our proposed methods, demonstrating their applicability in various real-world and synthetic benchmark datasets.
This work not only advances the understanding of concept drift in streaming data but also provides a general solution framework that balances performance with interpretability, thus paving the way for development of more reliable and explainable data-driven applications and systems