21090 research outputs found
Sort by
Deep Learning Frameworks for Anomaly Detection in Time Series and Graphs with Limited Labels
No full textAnomaly detection involves identifying patterns or behaviors that substantially differ from normal instances in a dataset. It has a wide range of applications in diverse fields such as cybersecurity, manufacturing, finance, and e-commerce. However, real-world anomaly detection often grapples with two main challenges: label scarcity, as anomalies are rare and hard to label, and the complexity of data structures, which can involve intricate dependencies that require careful analysis.
In this thesis, we develop deep learning frameworks designed to work effectively for label-free or extremely limited labeling scenarios, with a focus on time series anomaly detection (TSAD) and graph anomaly detection (GAD). To overcome the issue of label scarcity, our initial work investigates unsupervised TSAD methods that extract meaningful patterns from abundant unlabeled data. Building on recent advances in contrastive learning from NLP and computer vision, we introduce the Contrastive Neural Transformation (CNT) framework. This approach integrates temporal contrastive learning with neural transformations to capture context-aware and discriminative patterns effectively.
Moreover, the dual-loss formulation prevents representation collapse by avoiding reliance on negative samples - a common challenge in anomaly detection, where the majority of instances represent normal behavior. While capturing temporal context is essential, understanding inter-series relationships is equally important in multivariate TSAD. Anomalies may seem normal in isolation but reveal abnormal patterns when compared to other series. To address this, we introduce DyGraphAD, a dynamic graph-driven dual forecasting framework that models both intra- and inter-series dependencies through a combination of graph and time series forecasting tasks. This allows anomalies to be detected via significant forecasting errors in both the channel-wise and time-wise dimensions. To further enhance computational efficiency, we propose an alternative framework, termed Prospective Multi-Graph Cohesion (PMGC). PMGC leverages graph structure learning to model inter-series relationships in a task-specific manner, reducing computational load compared to manual sequential graph construction in DyGraphAD. Furthermore, it introduces a multi-graph cohesion mechanism to adaptively learn both long-term dependencies and diverse short-term relationships. A prospective graphing strategy is also introduced to encourage the model to capture concurrent inter-series relationships, reducing reliance solely on historical data.
Beyond TSAD, GAD is also critical due to its prevalence in numerous applications. Graphs provide structural information alongside node and edge attributes, and understanding the interplay between graph structure and attributes is essential for uncovering subtle anomalies not apparent when examining nodes or edges alone. Given that obtaining labeled data is relatively more feasible in graphs than in time series data for experimental purposes, we focus on GAD settings with limited labeling, more reflecting practical real-world scenarios. Specifically, we make the first attempt to address GAD in cross-domain few-shot settings, aiming to detect anomalies in a sparsely labeled target graph by leveraging a related but distinct source graph. To handle domain shifts, our CDFS-GAD framework incorporates a domain-adaptive graph contrastive learning and a domain-specific prompt tuning, aiming to align the features across two domains while preserving unique characteristics tailored to each domain. A domain-adaptive hypersphere classification loss and a self-training phase are introduced to further refine predictions in the target domain exploiting the limited labeling information. In addition to static graphs, many real-world applications involve dynamic graph data, where both the structure and attributes evolve over time. This adds complexity to anomaly detection, as both temporal and structural variations must be accounted for. Moreover, obtaining sufficient labeled data remains challenging, and related-domain labeled data may not be available in certain scenarios. To tackle the two more practical issues, we propose the EL-DGAD framework, specifically designed for detecting anomalies in dynamic graphs in extremely labeled conditions. This framework enhances model robustness through a transformer-based temporal graph encoder that captures evolving patterns from local and global perspectives. An ego-context hypersphere classification loss is further introduced to adjust the anomaly detection boundary contextually under limited supervision, supplemented by an ego-context contrasting module to improve generalization with unlabeled data.
Overall, this thesis tackles anomaly detection for two commonly used data types, addressing unsupervised, semi-supervised, and cross-domain few-shot scenarios to meet the demands of real-world applications. Our extensive experiments show that the proposed frameworks perform well against various benchmark datasets and competitive anomaly detection baselines
Reinforcement Learning for Scheduling Processes Under Uncertainty in Chemical Engineering Facilities
No full textOptimal scheduling of chemical systems has gained interest as it provides economic advantages. The development of methodologies for approaching this problem have expanded considerably bringing multiple options to the process optimization field. Real world applications deal with uncertainty in multiple stages of the process, from price fluctuations to processing times and material quality. Preventive and reactive scheduling techniques, which adapt to uncertainty realizations were developed to approach the scheduling optimization problem under uncertainty. Recently, the use of Deep Neural Networks (DNN) combined with Reinforcement Learning (RL) algorithms became an option to generate policies for decision-making processes. From the context of scheduling, a policy can be beneficial as it produces the schedule online, responding to the needs of the process and the realization of uncertainties in real time. In this thesis, a set of methodologies to approach the scheduling problem under uncertainty for batch systems with Deep Reinforcement Learning (DRL) methods is presented. The state-of-the-art methods available in this area are improved in this work with the development of different techniques that study the translation of the scheduling problem into the framework of Reinforcement Learning.
Contrary to multiple approaches in the literature where the process environment is assumed to be a Markov Decision Process (MDP), the methods presented in this work assume a partial observability of the process. This is called Partially Observable Markov Decision Process (POMDP) and it is useful to handle the uncertainty in the process as well to perceive the evolution over time of the process. To hold this assumption for the scheduling process, the use of Recurrent Neural Networks (RNN) is implemented in order to analyze their performance in the scheduling optimization problem. To the author’s knowledge, these networks have not been used before with this purpose, i.e., their implementation in the literature is focused on other features of this type of network, for instance, their capacity to handle inputs of various lengths. This new perspective is compared with implementations framed as MDPs showing the advantage of approaching the scheduling problem in this way.
The decision space of the scheduling problem is usually discretized into a set of actions that can be taken in the online scheduling process with DRL. In this work we propose the use of hybrid agents that can make simultaneous decisions in the process at every decision-step. This allows to expand the applicability of the scheduling agent into more realistic scenarios where more than one decision is required. Moreover, this method extends the nature of the decisions into the continuous space by using DRL algorithms that can work on this domain. This method also allows the integration of the scheduling tasks with others levels in the hierarchical manufacturing systems, e.g., planning or control tasks. To the author’s knowledge, attempts to extend the number of decisions, from the agent to these other levels, are not reported in the literature.
The use of DNNs for modeling policies in the scheduling process brings the issue of the “black box” model in which the model does not provide any descriptions of its heuristics for humans to understand the logics behind its decisions. This becomes an obstacle when ensuring that the setting of hyperparameters aligns with the current decisions of the agent. In other words, the agent does not provide any insights about its decisions and the only way to check this is through the final results of the agent. To provide the agent with interpretability, the use of attention mechanisms is implemented in this work. They allow to develop an attention matrix at every decision-step in the process which allows to gather information on the logics behind the decisions. Attention mechanisms have been used in the literature due to their outstanding capacity for building correlations between the elements of the process. To the author’s knowledge, the use of this interpretability for inspection and correction of hyperparameters in the scheduling problem is not reported in the literature.
The algorithms from DRL that are used in the methods presented are: a) a variation of Deep Q-Learning and b) Proximal Policy Optimization (PPO). Deep Q-Learning is a well-known DRL method that specializes in problems with a discrete action space; on the other hand, PPO is applicable to discrete and continuous action spaces. These methods are of relatively simple implementation and showed good performance on the works presented in this thesis. The implementations of the methods developed were tested on job-shops, flow-shops and State Task Networks, following objective functions related to reduction of makespan and product maximization. Results showed that the trained agents with the presented methods can generate schedules considering the uncertain parameters of the system, thus making this online scheduling agent attractive for industrial-scale applications. The capacity to react of the agent is demonstrated with the experiments performed on each study. Moreover, the results were compared with different benchmarks including alternative DNNs and optimization solvers. The implementation of DRL methods to approach the scheduling problem under uncertainty demonstrated that this alternative has potential as a reactive online scheduler that can provide reliable responses in short turnaround times. The set of methods presented in this thesis illustrate the advantages and limitations of the incorporation of machine learning into the decision-making process in the context of chemical engineering
The Story of Tiles and Ropes In-Between
No full textThe Story of Tiles and Ropes In-Between explores migration, in-betweenness, and resilience of a female body through installation, sculpture, video, poetry and performance. Using clay and jute—materials tied to memory and place—the work navigates themes of fragmentation and reconstruction. Walking is an integral focus to this body of artwork, serving as both method and metaphor in navigating identity in the diaspora. Influenced by theorists like Homi Bhabha and Gloria Anzaldúa, the work engages with the third space, where hybridity and transformation unfold. Through movement, sound, and material, the work redefines home as a fluid, evolving state rather than a fixed location
High-Performance Coordination with Weaker Protocols: From Shared Registers to Data Feed Processing in Blockchains
No full textBlockchains are decentralized ledger technologies that provide secure, transparent, and tamper-resistant transaction records.
Permissioned blockchains restrict participation to authorized entities and employ Byzantine Fault Tolerance (BFT) protocols for consensus.
As blockchain systems evolve, the need for exchanging data and assets across heterogeneous networks has grown.
To support such functionality, platforms increasingly integrate with external systems, enabling information flow between decentralized and conventional infrastructures.
Blockchain oracles serve a key role in this integration by injecting real-world data—such as asset prices or event outcomes—into smart contracts through structured oracle data feeds.
Conventional blockchain systems typically employ BFT consensus for oracle data integration, which introduces substantial latency and computational overhead.
Although consensus protocols offer robust consistency guarantees, they also incur significant computational and communication overhead, thereby constraining system responsiveness and scalability in blockchain environments.
To address these challenges, this thesis investigates weaker coordination protocols as lightweight alternatives that relax strict consistency requirements while preserving sufficient reliability for practical deployment.
This thesis evaluates the effectiveness of these protocols in enhancing the performance of read-write operations—a common access pattern in oracle data feed processing—across three distinct projects.
First, this thesis assesses the overhead of consensus in key-value processing by comparing consensus protocols with shared register protocols.
The comparison reveals two previously overlooked limitations that arise under workload-driven access patterns.
Second, it introduces a shared register protocol tailored for wide-area networks, which employs loosely synchronized clocks and asymmetric quorum techniques to reduce coordination latency across heterogeneous client loads.
Third, it applies weaker coordination techniques to oracle data feed processing—a fundamental component of many smart contract applications.
By decoupling data feed processing from BFT consensus using a probabilistic quorum protocol and a censorship-resistant broadcast primitive, the proposed design improves both throughput and data freshness for on-chain transactions dependent on timely off-chain inputs.
Collectively, these contributions demonstrate how workload-specific weaker protocols can enable more efficient distributed systems.
By reducing coordination overhead without sacrificing practical correctness, the proposed approaches support the development of scalable, responsive, and application-aware decentralized infrastructures
Embedded System Anomaly Detection via Boot Power Trace Analysis
No full textEmbedded systems play a crucial role in safety-critical domains, and it is essential to maintain their integrity. This thesis presents a robust framework for detecting hardware and firmware anomalies in embedded systems through boot-phase power consumption analysis. The proposed Sliding Window Anomaly Detection (SWAD) method establishes a nominal boot power profile and compares new boot traces against this baseline using sliding windows. By analyzing localized power dynamics, SWAD detects deviations caused by hardware or firmware modifications while accommodating natural variations in power behaviour. Experimental validation on single-board computers and flight controllers demonstrates the method’s effectiveness in identifying diverse hardware and firmware attacks, achieving overall F1 scores of 98\%, 96\%, and 85\% across three systems used in the case studies. These results highlight the promising role of power side-channel analysis in enhancing security in complex embedded systems
Breast Cancer Detection Using Microwave Low-Frequency Metasurface-Based Techniques
No full textBreast cancer has become an important global public health issue and is the leading cause of death among women. Early detection plays a pivotal role in the effective treatment of breast cancer. Microwave modalities offer a promising alternative because they allow deep penetration into the breast, resulting in an adequate contrast between the electrical properties of the malignant and benign breast tissue. Microwaves are also a mature technique that is relatively affordable and non-ionizing, thus addressing many of the deficiencies of traditional methods. In the first part of this study, we use computer biomechanics finite element simulation software (FEBio) to simulate the physical compression of anatomically realistic breast phantoms. The phantoms are 3D anatomically realistic and made up of MRI images. The compressed breast models derived from this part of this study will create a realistic scenario for the electromagnetic simulation in CST software. We successfully compressed a set of realistic phantoms, and these accurate compressed models created several natural scenarios to test the microwave technology developed in this investigation. The second part of this work introduces an electromagnetic energy scanning technique with a specific application to detect breast cancer. The technique is based on a field detector probe inspired by the concept of a metasurface composed of an ensemble of electrically small elements where a single element represents the field detector. This novel probe provides a resolution that cannot be matched by either electrically small probes or resonance-based probes that have dimensions comparable to the wavelength. The field emanating from a specific structure, such as a human female breast, can be scanned by the proposed probe to achieve a resolution in the millimeter range while operating in the low-microwave frequency spectrum. The probe was tested numerically, and a prototype was tested experimentally, demonstrating its effectiveness in providing a field resolution of approximately 5 mm. The third part of this thesis introduces a novel metasurface sensor designed to detect breast cancer by operating at sub-microwave frequency. Each cell of the metasurface gives rise to a voltage difference that can be measured at the back of the metasurface. Thus, the metasurface has the ability to generate an impression that relates to the constituents of the breast. Simulations were performed to capture and analyze a collection of images using artificial intelligence methodologies. The findings of this testing stage demonstrated the device's capability to identify anomalies in the breast that may or may not be cancerous. This concept was also extended and validated in an experimental trial; the fabricated device was able to detect a small PEC sphere of 10 mm diameter. Both results, simulation and experimental test, show the potential of this device in successfully detecting breast cancer. This study also explores, in part four, the classification between healthy and unhealthy breast tissue by integrating a microwave probe and a self-organized algorithm. This system collects the reflection coefficients from multiple breast samples where there may or may not be the presence of cancerous tissue. The probe showed remarkable capability to detect the changes in dielectric properties in the breast tissue, and the self-organized algorithm addressed the challenge of classifying this complex data overpassing traditional machine learning methods. We tested the capabilities of this detection system under simulation and experimental validation
Functionalized Polymer Composites for Electrochemical Sensing of Glucose and Smart Insulin Delivery on a Microneedle- Based Platform
No full textAdvancements in continuous glucose monitoring (CGM) devices combined with readily available insulin pumps have recently enabled development of artificial pancreas device systems that can automatically deliver insulin in response to real-time glucose measurements. Although effective, CGM devices still face challenges in accuracy and stability due to enzymatic detection. In addition, their long solid and metallic needles are not compatible with soft skin and often cause user discomfort. To address these issues, we developed skin-compatible hydrogel microneedle (HMN) electrodes that utilize state-of-the-art conductive and flexible yet mechanically strong polymers to tackle the problems associated with rigid long needles. Via linking dopamine into the backbone of HMN electrodes, we developed an HMN-CGM device based on dopamine's redox properties to generate platinum nanoparticles for non-enzymatic glucose detection. To overcome swelling-related limitations, we developed an enhanced HMN sensor, HMN-P4Au, by incorporating PEDOT:PSS, gold nanoparticles, and polyvinylpyrrolidone. This integration improves sensing capability, mechanical strength, and overall performance reliability. In vivo validation of HMN based CGMs in diabetic rats demonstrated a mean absolute relative difference (MARD) of 12% compared to glucometers, with stable readings maintained over multiple days.
Current insulin pumps face challenges due to long needles that cause discomfort and adhesive issues that can lead to detachment, disrupting insulin delivery. To overcome these problems, we developed an HMN embedded with insulin-loaded lipid nanoparticles (LNPs). This design leverages the combined benefits of PEDOT:PSS and dopamine as iontophoresis enhancers. This combination reduces the required iontophoresis voltage to levels compatible with human skin, enabling efficient and safe insulin delivery in form of LNPs. The LNP-insulin formulations, incorporated into microneedle arrays, achieved effective glucose regulation in diabetic rats through a dual-release mechanism. A basal passive release provided sustained insulin delivery, while an active, iontophoresis-triggered release enabled precise glucose control when needed. This combination successfully lowered blood glucose levels to a normoglycemic range without inducing hypoglycemia, demonstrating the potential of this approach for safe and effective diabetes management.
Building upon these advancements and via seamless integration of glucose sensing with insulin delivery, this work brings us closer to fully autonomous glucose regulation, offering a promising solution for next-generation diabetes management
Climate-denying rumor propagation in a coupled socio-climate model: Impact on average global temperature
No full text© 2025 Satheesh Kumar et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.Individual attitudes vastly affect the transformations we are experiencing and are vital in mitigating or intensifying climate change. A socio-climate model by coupling a model of rumor dynamics in heterogeneous networks to a simple Earth System model is developed, in order to analyze how rumors about climate change impact individuals' opinions when they may choose to either believe or reject the rumors they come across over time. Our model assumes that when individuals experience an increase in the global temperature, they tend to not believe the rumors they come across. The rumor rejectors limit their CO2 emissions to reduce global temperature. Our numerical analysis indicates that, over time, the temperature anomaly becomes less affected by the variations in rumor propagation parameters, and having larger groups (having more members) is more efficient in reducing temperature (by efficiently propagating rumors) than having numerous small groups. It is observed that decreasing the number of individual connections does not reduce the size of the rejector population when there are large numbers of messages sent through groups. Mitigation strategies considered by the rejectors are highly influential. The absence of mitigative behavior in rejectors can cause an increase in the global average temperature by 0.5oC. Our model indicates that rumor propagation in groups has the upper hand in controlling temperature change, compared to individual climate-denying propagation.Natural Sciences and Engineering Research Council of Canada (NSERC), Discovery Grants
A University-Landscape Architect Collaboration to Develop an Augmentative and Alternative Communication (AAC) Board for a City Park Playground: A Full Report of its Design
No full textIn this report, we detail the development of an AAC board for a community playground situated in a large park in the heart of Waterloo, Ontario, Canada. AAC boards for community locations have received much less attention and study than AAC tools intended for individual use at home and/or in schools, or by children on school playgrounds. This AAC board was developed through an ongoing collaboration between the landscape architect redeveloping the playground for the city and a professor of developmental psychology with expertise in children’s early social pragmatic communication, along with her students at the University of Waterloo. For this AAC board, we prioritized practical communication in a playground context by adopting an activity-focused approach. We describe the major steps taken in developing the AAC board and explain the reasoning and empirical support that guided our content and design decisions. These included choices about the vocabulary included, the symbols used and developed, and the positioning of all components on the board. We also considered symbol and border colours to support accessibility and ease of use. In this process, we sourced empirical evidence from many domains, including speech-language pathology, AAC, and language and cognitive development within developmental psychology. We also trialled a draft version of the AAC board in the park and made revisions based on feedback from children and their parents. We considered novel factors our expertise suggested would be relevant, such as the age and height of children and the vertical position of symbols on the board, and we introduced some novel elements such as a colour wheel. We conclude by outlining limitations and recent developments by manufacturers that may improve the design and use of playground AAC boards. We also highlight areas for further research that might be particularly helpful. We have prioritized sharing all aspects of our work openly so that it can be easily accessed, used, and shared by others without restrictions
Environment, Not Social or Governance, Oh My: Sustainability Priorities in Canadian University Sustainability Documents
No full textThis author accepted manuscript is deposited under a Creative Commons Attribution Non-commercial 4.0 International (CC BY-NC) licence. This means that anyone may distribute, adapt, and build upon the work for non-commercial purposes, subject to full attribution. If you wish to use this manuscript for commercial purposes, please contact [email protected]/methodology/approach
Comparing sustainability policies with the overarching goals of the United Nations Sustainable Development Goals through the approach of narrative policy analysis, this research explores the priorities present and what patterns emerge in these priorities.
Purpose
Universities are significant in their sustainability action not just for the education that they provide, but how they operate as mini cities, with massive environmental footprints, meaning that, their sustainability direction matters. As a result, this research investigates the priorities present in university sustainability policy documents from 33 Canadian universities.
Findings
Findings include that across Canada, planet related sustainability priorities, particularly those focused on cities, climate action and consumption are the most present, while people and prosperity elements of sustainability often fall outside the scope of these policies. Some regional variation emerges in key areas such as energy and climate action, and size also sees a correlation to prioritization of specific areas of sustainability such as waste.
Originality/value
This research will be of interest to researchers in the emerging field of sustainability in higher education, practitioners and administrators in university sustainability and policy makers looking to understand sustainability prioritization shape the future of nuanced sustainability directions.Social Science and Humanities Research Council of Canada's Canada Graduate Scholarship Doctoral Program, 767-2022-1443 || Balsillie School of International Affairs', Balsillie Fellowship || University of Waterloo's President's Graduate Scholarship