YorkSpace has migrated to a new version of its software. Access our Help Resources to learn how to use the refreshed site. Contact diginit@yorku.ca if you have any questions about the migration.
 

Computer Science and Engineering

Permanent URI for this collection

Browse

Recent Submissions

Now showing 1 - 20 of 110
  • ItemOpen Access
    Noise Modelling for Smartphone Cameras
    (2021-03-08) Abdelhamed, Abdelrahman Kamel Siddek; Brown, Michael S.
    In our everyday life we capture photographs. We create these images by measuring the amount of light, radiated from scenes in our physical world, on camera sensors embedded in our smartphones. Image noise is variation in the measurement of intensities or colours in digital images and it has the undesirable effect of obscuring information in images. Image noise is produced from two main sources: (1) the unavoidable, random nature of light and (2) the imaging sensor and associated circuitry. Unlike professional cameras, smartphone cameras have much smaller imaging sensors which makes them more susceptible to higher and more complex noise. To model, and ultimately remove, image noise, many mathematical models have been proposed. These models either represent synthetic noise or rely on assumptions that makes them unable to model real noise distributions observed from empirical data. One major reason for that is the lack of sufficient real noisy image datasets with ground truth images that can enable the study of real camera noise. The purpose of this dissertation is to provide a study on image noise modelling based on data-driven approaches specific to smartphone cameras. To this end, we first propose a systematic method for estimating ground truth noise-free images from noisy images captured by smartphone cameras. Using the proposed method, we collect a large-scale dataset, termed the Smartphone Image Denoising Dataset (SIDD), of high-quality images that can be used for noise modelling. Next, we utilize the SIDD dataset to devise a generative noise model, termed Noise Flow, that can be used to synthesise realistic noisy images to be utilized in many computer vision tasks. We also use our datatset to provide a benchmark for image denoising algorithms on real noisy images. As part of this benchmarking effort, we have developed an online image denoising challenge with the necessary software tools to facilitate the evaluation of image denoising methods applied to realistic noisy images. We believe the work in this dissertation helps to advance the state of the art in image noise modelling and image denoising.
  • ItemOpen Access
    Distracted Agents in Crowd Simulations
    (2021-03-08) Kremer, Melissa Anne; Faloutsos, Petros
    Crowd simulations can provide meaningful insights about crowd movement and flow in arbitrary environments, which is useful for predictive planning and analysis. High fidelity agents and simulations are important factors in the quality of the analysis and their validity. Historically, agents in crowd simulations have been largely homogeneous in terms of physical and behavioural properties. However, this homogenous structure does not accurately reflect crowds seen in the real world, which are largely heterogeneous in a wide variety of aspects including behaviours. In particular, distracted behaviour in crowds has been largely ignored in crowd simulations to date. The use of cellular phones has become one of the most commonly observed distracted behaviour in real world urban crowds. This thesis presents a first step towards modelling this important phenomenon in crowd simulations.
  • ItemOpen Access
    Neural Question Generation with Transfer Learning and Utilization of External Knowledge
    (2020-11-13) Delpisheh, Marjan; An, Aijun
    Neural question generation (NQG) applies deep neural networks to solve the problem of automatically generating questions from text documents. The performance of deep neural networks relies heavily on the availability of a large amount of labelled training data. For domains where labelled training data are very limited, NQG models suffers from poor performance. Another problem that NQG encounters is the problem of rare and unknown words that occur during training and inference but do not exist in the vocabulary list. We first investigate the impact of transfer learning on NQG, and explore the effects of transferring knowledge learned from data in a general domain into different layers of the NQG network. To deal with the rare and unseen word problem, we integrate semantic relationships in the WordNet lexical database, which is a type of general knowledge external to the training data, into the input representation of the NQG system.
  • ItemOpen Access
    Analyzing Human-Building Interactions in Virtual Environments Using Crowd Simulations
    (2020-11-13) Usman, Muhammad; Faloutsos, Petros
    This research explores the relationship between human-occupancy and environment designs by means of human behavior simulations. Predicting and analyzing user-related factors during environment designing is of vital importance. Traditional Computer-Aided Design (CAD) and Building Information Modeling (BIM) tools mostly represent geometric and semantic aspects of environment components (e.g., walls, pillars, doors, ramps, and floors). They often ignore the impact that an environment layout produces on its occupants and their movements. In recent efforts to analyze human social and spatial behaviors in buildings, researchers have started using crowd simulation techniques for dynamic analysis of urban and indoor environments. These analyses assist the designers in analyzing crowd-related factors in their designs and generating human-aware environments. This dissertation focuses on developing interactive solutions to perform spatial analytics that can quantify the dynamics of human-building interactions using crowd simulations in the virtual and built-environments. Partially, this dissertation aims to make these dynamic crowd analytics solutions available to designers either directly within mainstream environment design pipelines or as cross-platform simulation services, enabling users to seamlessly simulate, analyze, and incorporate human-centric dynamics into their design workflows.
  • ItemOpen Access
    Elastic Synchronization for Efficient and Effective Distributed Deep Learning
    (2020-11-13) Zhao, Xing; An, Aijun
    Training deep neural networks (DNNs) using a large-scale cluster with an efficient distributed paradigm significantly reduces the training time. However, a distributed paradigm developed only from system engineering perspective is most likely to hindering the model from learning due to the intrinsic optimization properties of machine learning. In this thesis, we present two efficient and effective models in the parameter server setting based on the limitations of the state-of-the-art distributed models such as staleness synchronous parallel (SSP) and bulk synchronous parallel (BSP). We introduce DynamicSSP model that adds smart dynamic communication to SSP, improves its communication efficiency and replaces its fixed staleness threshold with a dynamic threshold. DynamicSSP converges faster and to a higher accuracy than SSP in the heterogeneous environment. Having recognized the importance of bulk synchronization in training, we propose the ElasticBSP model which shares the proprieties of bulk synchronization and elastic synchronization. We develop fast online optimization algorithms with look-ahead mechanisms to materialise ElasticBSP. Empirically, ElasticBSP achieves the convergence speed 1.77 times faster and an overall accuracy 12.6% higher than BSP.
  • ItemOpen Access
    Design and Operation Mechanisms of Battery and Hydrogen Based Transport Systems Integrated with Power Grids
    (2020-11-13) El-Taweel, Nader Ashraf Abd El-Fattah; Farag, Hany E. Z.
    As a result of the low Greenhouse Gases (GHG) emissions in the electricity generation profiles, electrification of transit networks represents a promising approach to reduce transportation-related GHG emissions. Two fundamental concepts have been adopted to electrify transport systems: utilization of (i) battery storage for Battery Electric Vehicle (BEV), and (ii) hydrogen for Fuel-Cell Electric Vehicle (FCEV). Each of the two concepts has its own design and operation challenges in order to be widely and efficiently deployed. Accordingly, this thesis focuses on developing new models to address the imminent challenges of design and operation practices that are associated with the adoption of both concepts. First, novel analytical methodologies are developed to be applied to the size estimation of BEV and FCEV fueling stations, as a critical step to set the stage for the transportation electrification. The ratings of various components are expressed in terms of the system operation percentage using the proposed formulation, and the desired ratings are selected at which the net profit reaches the maximum point. Second, both Public Bus Transit (PBT) and power utility operators retain various challenges in facilitating the seamless integration of Battery Electric Bus (BEB) fleet systems. The most salient challenges are: (i) the lack/unavailability of real-world and high-resolution speed data of BEB to accurately calculate the Electric Bus Energy Consumption (EBEC), and (ii) the lack of appropriate simulation tools to model and optimize BEB fleet systems. Therefore, a novel model to generate a set of synthetic speed profiles is proposed using the basic information of the bus trip: duration, distance, and bus stops. A new mathematical formulation is also proposed to model and optimize the design of BEB fleet systems. The model considers the operational requirements of PBT systems, utility grid model and the EBEC characteristics. Third, the proliferation of hydrogen fueling stations throughout the transportation network and justifying their economic viability are key factors to the success of the FCEVs. Accordingly, a new model for optimal scheduling of distributed hydrogen storage stations is proposed to serve the transport sector and the electricity market Demand Response (DR) program, besides optimizing the hydrogen sale price. Further, the Liquid Organic Hydrogen Carrier (LOHC) technology now offers a promising solution for the reliable and safe storage of hydrogen. Hence, this thesis also demonstrates how such plants should be optimally sized and operated for joint applications for concurrent services to both the transportation sector and utility grid ancillary services. The findings of this thesis highlight the feasibility of current BEBs technology to replace diesel-based transit buses, shall appropriate technical design and measures be considered to alleviate the negative interactions between power utilities and transit networks. In addition, ancillary services provision to the grid is concluded to be a win-win situation to the utility grid and the hydrogen facility that can reduce the hydrogen sale price.
  • ItemOpen Access
    Ultra-Low Power Wireless Sensor Circuits for IOT Applications
    (2020-11-13) Shang, Zhongxia; Lian, Yong Peter
    Wireless sensors, which are responsible for local data acquisition, processing and communication, play an important role in Internet of Things (IoT) applications. This research focuses on two basic components in wireless sensors, i.e., the low-power frequency tunable wireless receiver and the power management unit (PMU) for autonomous operation. In IoT applications, different sensors may need to operate in different frequency bands in order to meet environment constraints and industrial/medical standards. Thus, it is highly desirable to design a frequency configurable wireless receiver that provides flexibility in operation frequency. A 4-path filter based frequency shift keying (FSK) receiver is proposed to meet such a need, where the carrier frequency can be adjusted without changing the circuit. In addition, the proposed receiver requires no low-noise amplifier (LNA), which boosts the power efficiency. Frequency synthesizer is critical in FSK transceiver as it provides an accurate reference frequency. Based on 4-path mixer, a novel two-step calibration frequency synthesizer structure is proposed for low power consumption and wide locking range. Measurement results show that the proposed receiver achieves an energy per bit as 74pJ/bit with 2.5Mbps data rate and 184W power consumption. Post-layout simulation results show that the proposed frequency synthesizer has a figure of merit (FOM) value as 1.4W/MHz with 220MHz tuning bandwidth and 305W power consumption. Autonomous operation is another requirement for the sensors in many IoT applications, such as wearable sensors. Energy harvester is commonly used for autonomous sensors, where a PMU with low start-up voltage is necessary. To meet such a requirement, a novel controller for PMU is proposed to boost the power efficiency under very low load current. The proposed PMU can be started up with input voltage as low as several tens of millivolts. The novel controller costs little power and the overall efficiency is increased. Also, a hardware efficient maximum power point tracking (MPPT) algorithm which is suitable for energy source with fixed internal resistance is proposed. Measurement results show that the proposed system has a low controller power as 3.6W and the overall conversion efficiency is 83.9%.
  • ItemOpen Access
    Wavelet Flow: Fast Training of High Resolution Normalizing Flows
    (2020-11-13) Yu, Jason Jiasheng; Brubaker, Marcus
    Normalizing flows are a class of probabilistic generative models which allow for fast density computation, efficient sampling, and are effective at modelling complex distributions like images. A drawback among current methods is their significant training cost, sometimes requiring months of GPU training time to achieve state-of-the-art results. This thesis introduces Wavelet Flow, a multi-scale, normalizing flow architecture based on wavelets. A Wavelet Flow has an explicit representation of signal scale that inherently includes models of lower resolution signals and conditional generation of higher resolution signals, i.e., super resolution. A major advantage of Wavelet Flow is the ability to construct generative models for high resolution data (e.g., 1024 1024 images) that are impractical with previous models. Furthermore, Wavelet Flow is competitive with previous normalizing flows in terms of bits per dimension on standard (low resolution) benchmarks while being up to 15 faster to train.
  • ItemOpen Access
    Soft-Switched Multi-Input Converter for Renewable Energy Systems
    (2020-08-11) Moury, Sanjida; Lam, John Chi Wo
    In a renewable energy system, multiple energy sources (such as combining renewable energy and energy storage) are utilized together to enhance the overall energy conversion reliability since renewable energy, such as wind and solar, are intermittent in nature. In conventional power architecture, each energy source requires a dedicated individual power converter to perform specific control or power management function. To reduce the overall number of circuit components, multi-input converter (MIC) configuration provides a cost effective power architecture when multiple energy sources are utilized, as they can use fewer filter circuit components and utilize smaller system space. To maximize the features offered by MICs, a truly power efficient and compact MIC should utilize minimal number of active switching components with soft-switching features, while at the same time, assist each energy power interface to achieve all the required control functions. In the first part of this thesis, a class of several soft-switched DC-DC MICs is proposed, where each input module of the devised MICs utilizes only a single switch. Each of the presented MIC circuits can also be integrated with a front-end (either single phase or three phase) AC-DC stage without adding additional switches to interface with renewable energy generation unit that outputs an AC voltage. In addition, input power factor correction is also provided. The second part of this thesis investigates the use of the devised DC-DC MIC circuit to improve the double power conversion steps typically seen in a solar-battery energy conversion system with a common DC grid that utilizes a bi-directional energy storage power converter. This chapter focuses on the development of a soft-switched MIC circuit that consists of integrated unidirectional energy storage (i.e. battery) power interface, for use in module-connected solar energy power optimizer system with distributed energy storage. In the proposed circuit, the storage charging (storage absorbs power) circuit is integrated with the input stage of the main converter via high frequency AC link, whereas the storage discharging (storage delivers power) circuit is connected to the output stage of MIC. As a result, energy is directed from the input to the battery and from the battery to the load through a single power conversion step. The third part of this thesis utilizes the circuit concepts devised previously to develop a new soft-switched MIC configuration that consists of an integrated unidirectional energy storage power interface, as well as to provide output voltage regulation. Hence, the proposed MIC is applicable for both regulated and unregulated grids. The control mechanism of the proposed system is presented. The operating principles and characteristics of each proposed converter topology are provided in detail. Simulation and experimental results on proof-of-concept prototypes are provided to demonstrate the functionalities of each devised MIC topology.
  • ItemOpen Access
    The Role of Context in Understanding and Predicting Pedestrian Behavior in Urban Traffic Scenes
    (2020-08-11) Rasouli, Amir; Tsotsos, John K.
    Today, one of the major challenges faced by autonomous vehicles (AVs) is the ability to drive in urban environments. Such a task requires interactions between AVs and other road users, in particular pedestrians, to resolve various traffic ambiguities. To interact with pedestrians, AVs must be able to understand the objectives of pedestrians and predict their forthcoming actions. In this dissertation, we investigate the role of context on understanding and predicting pedestrian behavior in urban traffic scenes. Towards this goal, we begin by explaining why behavior prediction is necessary for social interactions. Next, we conduct a meta-analysis of a large body of behavioral literature and identify the factors that potentially impact pedestrian behavior and how these factors are interconnected. We extend the past findings by conducting two behavioral studies of pedestrians. The first study shows that pedestrians often engage in different forms of communication, mainly implicit, with changes in their movement patterns and the frequency of communication varying depending on road structure, social factors, and scene dynamics. The second study identifies the diversity of pedestrian behavioral patterns at the time of crossing and how it is influenced by factors such as the road width, demographics, crosswalk delineation, and driver behavior. As part of the behavioral studies, we collected two novel large-scale datasets of pedestrian crossing behaviors. Using the data, we empirically evaluate various state-of-the-art and classical pedestrian detection algorithms and show how diversifying training data in terms of visual properties, such as lighting conditions and pedestrian attributes, enhance the generalizability of such algorithms. Furthermore, we propose a novel pedestrian trajectory prediction algorithm that achieves state-of-the-art performance. We show that incorporating pedestrian intention to cross helps improve reasoning about future motion trajectories. In addition, we propose a novel pedestrian crossing action prediction algorithm and illustrate that by including contextual information, such as pedestrian appearance, pedestrian pose, and velocity, we can enhance the accuracy of crossing action prediction. We also show that by combining different modalities of contextual data in a hierarchical fashion better performance can be achieved compared to alternative approaches.
  • ItemOpen Access
    Learning Semantic Relationships of Geographical Areas Based on Trajectories
    (2020-08-11) Mehmood, Saim; Papangelis, Emmanouil
    Mining trajectory data to find interesting patterns is of increasing research interest due to a broad range of useful applications, including analysis of transportation systems, location-based social networks, and crowd behavior. The primary focus of this research is to leverage the abundance of trajectory data to automatically and accurately learn latent semantic relationships between different geographical areas (e.g., semantically correlated neighborhoods of a city) as revealed by patterns of moving objects over time. While previous studies have utilized trajectories for this type of analysis at the level of a single geographical area, the results cannot be easily generalized to inform comparative analysis of different geographical areas. In this work, we study this problem systematically. First, we present a method that utilizes trajectories to learn low-dimensional representations of geographical areas in an embedded space. Then, we develop a statistical method that allows to quantify the degree to which real trajectories deviate from a theoretical null model. The method allows to (a) distinguish geographical proximity to semantic proximity, and (b) inform a comparative analysis of two (or more) models obtained by trajectories defined on different geographical areas. This deep analysis can improve readers understanding of how space is perceived by individuals and inform better decisions of urban planning. Our experimental evaluation aims to demonstrate the effectiveness and usefulness of the proposed statistical method in two large-scale real-world data sets coming from the New York City and the city of Porto, Portugal, respectively. The methods we present are generic and can be utilized to inform a number of useful applications, ranging from location-based services, such as point-of-interest recommendations, to finding semantic relationships between different cities.
  • ItemOpen Access
    Configuration of Microgrids Considering State Estimation, Service Restoration, and Integration with Natural Gas Systems
    (2020-08-11) Mohamed, Mohamed Zaki Abdelhamed; Farag, Hany E. Z.
    The interest in the adoption of smart grid technologies as a means for digitalization and automation of power distribution systems has increased rapidly in the last few years. This interest can be explained by the common belief that smart grid technologies greatly enhance the system reliability, power quality, overall efficiency, and most importantly the accommodation of distributed generations (DGs). As DG penetration levels increase, distribution networks are divided into a new set of management layers based on a microgrid structure. A typical microgrid is formed of a cluster of DG units feeding a group of loads that operates in parallel to or isolated from the main grid. Microgrids are the building blocks of smart distribution grids (SDG). The concept of microgrid brings numerous benefits; among which, the improvement of system reliability is the most salient. However, the realization of such benefit is strongly dependent on the implementation of appropriate design and operation methodologies that take into account the special philosophy and operational characteristics of microgrids. Accordingly, this thesis introduces new methodologies to enhance the operation and reliability of SDGs clustered into microgrids. In particular, three main functions are dealt with in this research work: optimum configuration, self-healing restoration, and the integration between power and natural gas microgrids. First, an optimal zone clustering (i.e. configuration) algorithm is proposed for dynamic state estimation in islanded microgrids (IMG) considering the supply adequacy of each zone. Second, a centralized-based optimization model with multi-objective functions is formulated to perform the service restoration process for microgrids operating in both grid-connected and islanded modes of operation. Further, to obviate the need for a central unit and reduce the problem complexity, the optimization problem is reformulated using distributed automated agents. Third, a new model is proposed for optimal scheduling of power-to-gas (PtG), gas-fired generation (GfG), and gas storage units in a multi-carrier energy system (MCES)-based microgrid. The model aims to facilitate the integration of renewable DGs, utilize gas and power price arbitrage, provide regulation services to the real-time market, and contribute to the restoration of power and gas loads during unplanned outages.
  • ItemOpen Access
    Effective Density Visualization of Multiple Overlapping Axis-Aligned Objects
    (2020-08-11) Costa, Niloy Eric; Papangelis, Emmanouil
    Large-scale analytics of multiple overlapping axis-aligned objects is a challenging computational geometry problem that can inform several applications and services, in diverse domains. The primary focus of this research is, given many axis-aligned objects, to devise efficient and effective data visualization methods that inform whether, where and how much they overlap. Currently, such visualizations rely on inefficient implementations to determine the size of the overlap of objects. We address this problem by exploiting state-of-the-art computational geometry methods based on the sweep line paradigm. These methods are fast and can determine the exact size of the overlap of multiple axis-aligned objects, therefore can effectively inform the visualization method. Towards that end, we propose OL-HeatMap, a novel density-based visualization technique that can be used to represent complex information about overlapping objects. Our experimental evaluation demonstrates the effectiveness of the proposed method in several synthetic and real-world data sets.
  • ItemOpen Access
    Analytically Defined Spatiotemporal ConvNets for Spacetime Image Understanding
    (2020-08-11) Hadji, Isma; Wildes, Richard
    This dissertation introduces a novel hierarchical spatiotemporal orientation representation for spacetime image analysis. This representation is designed to combine the benefits of the multilayer architecture of Convolutional Networks (ConvNets) and a more controlled approach to spacetime analysis. A distinguishing aspect of the approach is that unlike most contemporary convolutional networks no learning is involved; rather, all design decisions are specified analytically with theoretical motivations. This approach makes it possible to understand what information is being extracted at each stage and layer of processing as well as to minimize heuristic choices in design. Another key aspect of the network is its recurrent nature, whereby the output of each layer of processing feeds back to the input. The multilayer architecture that results systematically reveals hierarchical image structure in terms of multiscale, multiorientation properties of visual spacetime. To illustrate the utility of the proposed research, the designed networks has been tested on two spacetime image understanding tasks, dynamic texture recognition and video object segmentation. Further, the role of learning in the context of the proposed analytic approach to network design is systematically explored, thereby yielding a promising hybrid architecture. Finally, a new, large scale dynamic texture dataset is introduced and used for evaluation.
  • ItemOpen Access
    Biolocomotion Detection in Videos
    (2020-05-11) Kang, Soo Min; Wildes, Richard
    Animals locomote for various reasons: to search for food, to find suitable habitat, to pursue prey, to escape from predators, or to seek a mate. The grand scale of biodiversity contributes to the great locomotory design and mode diversity. In this dissertation, the locomotion of general biological species is referred to as biolocomotion. The goal of this dissertation is to develop a computational approach to detect biolocomotion in any unprocessed video. The ways biological entities locomote through an environment are extremely diverse. Various creatures make use of legs, wings, fins, and other means to move through the world. Significantly, the motion exhibited by the body parts to navigate through an environment can be modelled by a combination of an overall positional advance with an overlaid asymmetric oscillatory pattern, a distinctive signature that tends to be absent in non-biological objects in locomotion. In this dissertation, this key trait of positional advance with asymmetric oscillation along with differences in an object's common motion (extrinsic motion) and localized motion of its parts (intrinsic motion) is exploited to detect biolocomotion. In particular, a computational algorithm is developed to measure the presence of these traits in tracked objects to determine if they correspond to a biological entity in locomotion. An alternative algorithm, based on generic handcrafted features combined with learning is assembled out of components from allied areas of investigation, also is presented as a basis of comparison to the main proposed algorithm. A novel biolocomotion dataset encompassing a wide range of moving biological and non-biological objects in natural settings is provided. Additionally, biolocomotion annotations to an extant camouflage animals dataset also is provided. Quantitative results indicate that the proposed algorithm considerably outperforms the alternative approach, supporting the hypothesis that biolocomotion can be detected reliably based on its distinct signature of positional advance with asymmetric oscillation and extrinsic/intrinsic motion dissimilarity.
  • ItemOpen Access
    Adaptive Momentum for Neural Network Optimization
    (2020-05-11) Rashidi, Zana; An, Aijun
    In this thesis, we develop a novel and efficient algorithm for optimizing neural networks inspired by a recently proposed geodesic optimization algorithm. Our algorithm, which we call Stochastic Geodesic Optimization (SGeO), utilizes an adaptive coefficient on top of Polyaks Heavy Ball method effectively controlling the amount of weight put on the previous update to the parameters based on the change of direction in the optimization path. Experimental results on strongly convex functions with Lipschitz gradients and deep Autoencoder benchmarks show that SGeO reaches lower errors than established first-order methods and competes well with lower or similar errors to a recent second-order method called K-FAC (Kronecker-Factored Approximate Curvature). We also incorporate Nesterov style lookahead gradient into our algorithm (SGeO-N) and observe notable improvements. We believe that our research will open up new directions for high-dimensional neural network optimization where combining the efficiency of first-order methods and the effectiveness of second-order methods proves a promising avenue to explore.
  • ItemOpen Access
    Dual Fixed-Size Ordinally Forgetting Encoding (FOFE) For Natural Language Processing
    (2020-05-11) Watcharawittayakul, Sedtawut; Jiang, Hui
    In this thesis, we propose a new approach to employ fixed-size ordinally-forgetting encoding (FOFE) on Natural Language Processing (NLP) tasks, called dual-FOFE. The main idea behind dual-FOFE is that it allows the encoding to be done with two different forgetting factors; this would resolve the original FOFEs dilemma in choosing between the benefits offered by having either small or large values for its single forgetting factor. For this research, we have conducted our experiments on two prominent NLP tasks, namely, language modelling and machine reading comprehension. Our experiment results shown that the dual-FOFE provide a definite improvement over the original FOFE by approximately 11% in perplexity (PPL) for language modelling task and 8% in Exact Match (EM) score for machine reading comprehension task.
  • ItemOpen Access
    Design and Performance Analysis of Next Generation Heterogeneous Cellular Networks for the Internet of Things
    (2020-05-11) Ibrahim, Hazem Mohamed Saad; Nguyen, Uyen T.
    The Internet of Things (IoT) is a system of inter-connected computing devices, objects and mechanical and digital machines, and the communications between these devices/objects and other Internet-enabled systems. Scalable, reliable, and energy-efficient IoT connectivity will bring huge benefits to the society, especially in transportation, connected self-driving vehicles, healthcare, education, smart cities, and smart industries. The objective of this dissertation is to model and analyze the performance of large-scale heterogeneous two-tier IoT cellular networks, and offer design insights to maximize their performance. Using stochastic geometry, we develop realistic yet tractable models to study the performance of such networks. In particular, we propose solutions to the following research problems: -We propose a novel analytical model to estimate the mean uplink device data rate utility function under both spectrum allocation schemes, full spectrum reuse (FSR) and orthogonal spectrum partition (OSP), for uplink two-hop IoT networks. We develop constraint gradient ascent optimization algorithms to obtain the optimal aggregator association bias (for the FSR scheme) and the optimal joint spectrum partition ratio and optimal aggregator association bias (for the OSP scheme). -We study the performance of two-tier IoT cellular networks in which one tier operates in the traditional sub-6GHz spectrum and the other, in the millimeter wave (mm-wave) spectrum. In particular, we characterize the meta distributions of the downlink signal-to-interference ratio (sub-6GHz spectrum), the signal-to-noise ratio (mm-wave spectrum) and the data rate of a typical device in such a hybrid spectrum network. Finally, we characterize the meta distributions of the SIR/SNR and data rate of a typical device by substituting the cumulative moment of the CSP of a user device into the Gil-Pelaez inversion theorem. -We propose to split the control plane (C-plane) and user plane (U-plane) as a potential solution to harvest densification gain in heterogeneous two-tier networks while minimizing the handover rate and network control overhead. We develop a tractable mobility-aware model for a two-tier downlink cellular network with high density small cells and a C-plane/U-plane split architecture. The developed model is then used to quantify effect of mobility on the foreseen densification gain with and without C-plane/U-plane splitting.
  • ItemOpen Access
    Biomechanical Locomotion Heterogeneity in Synthetic Crowds
    (2020-05-11) Haworth, Michael Brandon; Faloutsos, Petros
    Synthetic crowd simulation combines rule sets at different conceptual layers to represent the dynamic nature of crowds while adhering to basic principles of human steering, such as collision avoidance and goal completion. In this dissertation, I explore synthetic crowd simulation at the steering layer using a critical approach to define the central theme of the work, the impact of model representation and agent diversity in crowds. At the steering layer, simulated agents make regular decisions, or actions, related to steering which are often responsible for the emergent behaviours found in the macro-scale crowd. Because of this bottom-up impact of a steering model's defining rule-set, I postulate that biomechanics and diverse biomechanics may alter the outcomes of dynamic synthetic-crowds-based outcomes. This would mean that an assumption of normativity and/or homogeneity among simulated agents and their mobility would provide an inaccurate representation of a scenario. If these results are then used to make real world decisions, say via policy or design, then those populations not represented in the simulated scenario may experience a lack of representation in the actualization of those decisions. A focused literature review shows that applications of both biomechanics and diverse locomotion representation at this layer of modelling are very narrow and often not present. I respond to the narrowness of this representation by addressing both biomechanics and heterogeneity separately. To address the question of performance and importance of locomotion biomechanics in crowd simulation, I use a large scale comparative approach. The industry standard synthetic crowd models are tested under a battery of benchmarks derived from prior work in comparative analysis of synthetic crowds as well as new scenarios derived from built environments. To address the question of the importance of heterogeneity in locomotion biomechanics, I define tiers of impact in the multi-agent crowds model at the steering layer--from the action space, to the agent space, to the crowds space. To this end, additional models and layers are developed to address the modelling and application of heterogeneous locomotion biomechanics in synthetic crowds. The results of both studies form a research arc which shows that the biomechanics in steering models provides important fidelity in several applications and that heterogeneity in the model of locomotion biomechanics directly impacts both qualitative and quantitative synthetic crowds outcomes. As well, systems, approaches, and pitfalls regarding the analysis of steering model and human mobility diversity are described.
  • ItemOpen Access
    Top-Down Selection in Convolutional Neural Networks
    (2020-05-11) Biparva, Mahdi; Tsotsos, John K.
    Feedforward information processing fills the role of hierarchical feature encoding, transformation, reduction, and abstraction in a bottom-up manner. This paradigm of information processing is sufficient for task requirements that are satisfied in the one-shot rapid traversal of sensory information through the visual hierarchy. However, some tasks demand higher-order information processing using short-term recurrent, long-range feedback, or other processes. The predictive, corrective, and modulatory information processing in top-down fashion complement the feedforward pass to fulfill many complex task requirements. Convolutional neural networks have recently been successful in addressing some aspects of the feedforward processing. However, the role of top-down processing in such models has not yet been fully understood. We propose a top-down selection framework for convolutional neural networks to address the selective and modulatory nature of top-down processing in vision systems. We examine various aspects of the proposed model in different experimental settings such as object localization, object segmentation, task priming, compact neural representation, and contextual interference reduction. We test the hypothesis that the proposed approach is capable of accomplishing hierarchical feature localization according to task cuing. Additionally, feature modulation using the proposed approach is tested for demanding tasks such as segmentation and iterative parameter fine-tuning. Moreover, the top-down attentional traces are harnessed to enable a more compact neural representation. The experimental achievements support the practical complementary role of the top-down selection mechanisms to the bottom-up feature encoding routines.