教学课程资源库 更多>>

关键词： Aerospace engineering   Mechanical engineering   Computer science  

摘要： This thesis addresses the development of low-cost remote-sensing and navigational techniques for commercial and private small, unmanned aircraft systems (sUAS) to support a multimodal surveying platform capable of navigating in little to no light. The motivation for this research arises from the growing need to expand commercial sUAS’s ability to safely and efficiently navigate in complex environments and collect useful data in an affordable manner. The research in this thesis particularly focuses upon systems that would be beneficial for an sUAS equipped with a multimodal sensor suite for visible and thermal remote sensing. The research seeks to develop a low-cost data collection workflow for gathering visible light and thermal orthomosaic maps and digital elevation models (DEM’s) without relying on expensive software packages. A series of commercial and open-source image reconstruction software packages are compared and augmented with custom tools to support a low-cost aerial mapping and data analysis technique. Additionally, this research explores alternative sensors to enable enhanced obstacle avoidance in complex flight environments in little to no light. Two novel sensor prototypes are discussed including a concept for an ultrasonic thermometer as well as a laser-projection system for detecting thin features. Sensors such as these offer low-cost alternatives to LIDAR and radar systems while continuing to expand sUAS operation into low-light environments. This project seeks to present the building blocks for a hypothetical UAV system capable of rapidly collecting a large amount of aerial imagery and thermal data in a dark environment. Such a platform can have uses in numerous applications such as nighttime delivery services in urban environments, search and rescue missions in forested areas, as well as thermal and structural inspection of buildings.

关键词： Computer science  

摘要： Biometric authentication is significantly growing to strengthen a secured verification system. The latest exponential growth in sophisticated technology and electronic media drastically enhances the quality of synthetically altered biometric substances. The further progression of the spoofing techniques entails more attention on trusted user authentication in diverse industries. Researcher have been working towards mitigation of different biometric attacks. Recently deep learning-based method have been applied successfully to detect biometric based presentation attack. In this dissertation, we apply the deep learning approaches to mitigate the iris and face based presentation attacks. Our research is divided into three phases. In the first phase, we explored several traditional machine learning techniques like support vector machines, logistic regression, random forest, and decision trees. We identified that the Convolutional Neural Networks (CNNs) with their deep internal layer structure, works well in mitigating the presentation attack. We implemented ‘modified AlexNet,’ ‘modified SpoofNet,’ and ‘modified VGGNet’ to classify real and fake face and iris images in our first phase of implementation. We obtained classification accuracy of 95% for the ‘modified-AlexNet,’ 97% for the ‘modified-SpoofNet,’ and 97% for the ‘modified-VGGNet’ on the iris spoofing dataset ATVS. In the second phase, we focus on the feature extraction techniques and implemented the discrete wavelet transform (DWT) as a feature extraction tool. We also developed a dynamic ‘modified-ResNet’ for classification on real and spoofed images. We experimented on video dataset and developed multiclass ‘modified-AlexNet’ for the video-based PAD. We obtained 94% classification accuracy using DWT enabled ‘modified-ResNet’ on ATVS dataset. Also, using the multiclass ‘modifed-AlexNet’ we obtained 99.98% classification accuracy on SiW dataset. In the third phase, we developed a multichannel deep learning model

关键词： Computer science  

摘要： With the ever-increasing amount of sensitive information being sent and received over the internet, the importance of fast secure cryptographic schemes will only continue to grow. In this paper we examine an existing but not widely known signcryption scheme, and propose modifications to it which increase its efficiency and protect it against side-channel cryptographic attacks.

关键词： Cameras   Dynamic range   Image fusion   Fuses   Sensors   Image edge detection   Computer science   Deghosting   ghost effect   guided filter   high dynamic range imaging   multi-exposure image fusion   multi-focus image   weight map  

摘要： In this paper, a novel de-ghosting image fusion technique is presented, which enhances the quality of low dynamic range images using multi-level exposures taken from the ordinary camera and also removes the ghosting artifact. In the proposed algorithm, first, the source images, taken under different exposure settings, are decomposed into base and detail layers using two-scale decomposition. The base and detail layers contain small and large-scale variation details of the source images, respectively. The Laplacian-of-Gaussian filter is applied to the source images to get the edge information. Afterward, the saliency map of the edges is computed. To remove the ghosting artifacts, a weight matrix is calculated by applying the median filter on the histogram equalized source images. The weight matrix is combined with the saliency map to generate more accurate weights. The separate weights for the base and detail layers are calculated using guided image filters. Finally, the base and detail layers' weights are fused with the source images to generate a vivid and enhanced image without any artifacts. The proposed technique is evaluated both qualitatively and quantitatively. The comparison of our technique in terms of Yang's Metric (Q(Y)), Quality Mutual Information (Q(MI)), Gradient-based Fusion Metric (Q(G)) and Chen Blum's Metric (Q(CB)) with other state-of-the-art techniques proves that the proposed technique outperforms existing techniques.

关键词： Computer science   Operations research   Graph databases   Artificial intelligence   Graph theory--Data processing  

摘要： This thesis describes the development of the SmartGraph, an AI enabled graph database. The need for such a system has been independently recognized in the isolated fields of graph databases, graph computing, and computational graph deep learning systems, such as TensorFlow. Though prior works have investigated some relationships between these fields, we believe that the SmartGraph is the first system designed from conception to incorporate the most significant and useful characteristics of each. Examples include the ability to store graph structured data, run analytics natively on this data, and run gradient descent algorithms. It is the synergistic aspects of combining these fields that provide the most novel results presented in this dissertation. Key among them is how the notion of “graph querying” as used in graph databases can be used to solve a problem that has plagued deep learning systems since their inception; rather than attempting to embed graph structured datasets into restrictive vector spaces, we instead allow the deep learning functionality of the system to natively perform graph querying in memory during optimization as a way of interpreting (and learning) the graph. This results in a concept of natural and interpretable processing of graph structured data. Graph computing systems have traditionally used distributed computing across multiple compute nodes (e.g. separate machines connected via Ethernet or internet) to deal with large-scale datasets whilst working sequentially on problems over entire datasets. In this dissertation, we outline a distributed graph computing methodology that facilitates all the above capabilities (even in an environment consisting of a single physical machine) while allowing for a workflow more typical of a graph database than a graph computing system; massive concurrent access allowing for arbitrarily asynchronous execution of queries and analytics across the entire system. Further, we demonstrate how this methodology is

关键词： Linguistics   Computer science  

摘要： Much work in Natural Language Processing (NLP) is broadly concerned with extracting useful information from unstructured text passages. In recent years there has been an increased focus on informal writing as is found in online venues such as Twitter and Yelp. Processing this text introduces additional difficulties for NLP techniques, for example, many of the terms may be unknown due to rapidly changing vocabulary usage. A straightforward NLP approach will not have any capability of using the information these terms provide. In such information poor environments of missing and incomplete information, it is necessary to develop novel, clever methods for leveraging the information we have explicitly available to unlock key nuggets of implicitly available information. In this work we explore several such methods and how they can collectively help to improve NLP techniques in general, with a focus on Sentiment Analysis (SA) and Entity Matching (EM). The problem of SA is that of identifying the polarity (positive, negative, neutral) of a speaker or author towards the topic of a given piece of text. SA can focus on various levels of granularity. These include finding the overall sentiment of a long text document, finding the sentiment of individual sentences or phrases, or finding the sentiment directed toward specific entities and their aspects (attributes). The problem of EM, also known as Record Linkage, is the problem of determining records from independent and uncooperative data sources that refer to the same real-world entities. Traditional approaches to EM have used the record representation of entities to accomplish this task. With the nascence of social media, entities on the Web are now accompanied by user generated content, which allows us to apply NLP solutions to the problem. We investigate specifically the following aspects of NLP for missing and incomplete information: (1) Inferring a sentiment polarity (i.e., the positive, negative, and neutral composition

关键词： Artificial intelligence   Computer science  

摘要： The recent advances in deep learning and the availability of vast volume of online personal images and videos have drastically improved the reality of synthesized faces in images and videos. While there are interesting and creative applications of the AI face synthesis systems, they can also be weaponized, as it can create the illusions of a person's presence and activities that do not occur in reality, which results in serious political, social, financial, and legal consequences. Therefore, it is of great importance to develop effective method to expose the AI-synthesized faces. In this thesis, a set of our recent efforts on detecting and protecting against AI-synthesized faces is described. For detection, the signals such as facial landmark difference, eye blinking and head pose inconsistency, and face warping artifacts are utilized to spot the synthesized faces respectively. Moreover, a new large-scale challenging synthesized face dataset with highly realistic quality is constructed, to further improve the development of detection. For protecting against the AI-synthesized faces, a proactive method is proposed to prevent people from being the victim of AI systems by disturbing the training process. Several experiments are conducted to demonstrate the effectiveness of our works.

关键词： Bioinformatics   Computer science  

摘要： Acute Myeloid Leukemia (AML) is a heterogeneous cancer of the blood that progresses quickly, with approximately 10,000 AML related deaths reported annually in the United States. Patients with AML tend to have genetic variations, which can significantly affect drug sensitivity and treatment outcomes. While massive amounts of big biomedical data have been generated to characterize AML, these genetic variations are not yet well-understood. Thus, the development of individualized approaches to AML therapy using these big data has great potential. The promise of precision medicine is that knowledge of the genetic characteristics present within a cancer will enable better choices for therapy. In this thesis, we applied data science techniques to analyze AML biomedical data in collaboration with Dr. Pamela Becker, Hematology, UW-Seattle, with the goal of identifying genetic variants that drive drug responses in AML. We identified 30 novel gene-drug pairs with statistical significant responses. Additionally, both univariate and multivariate machine learning models were created, with multivariate feature selection via Bayesian Model Averaging. We found multivariate models to outperform univariate models in most cases. Additionally, an automated workflow for the analysis was created to allow for incremental additions of patient data over the course of Dr. Becker's study.

关键词： Molecular biology   Bioinformatics   Computer science  

摘要： The work in this dissertation focuses on summarization, visualization, and recognition problems that arise in high-dimensional spatial data. Specifically, the data of interest to us are those that arise in computational biology problems aiming to relate molecular three-dimensional structure to molecular function. Molecular structure has long been recognized as a carrier and regulator of biological function. Technological advances have revealed that intrinsically-dynamic molecules, such as proteins, populate a vast structure space, assuming different structures to modulate interactions with different partners in the cell. In particular, it is now commonplace to obtain hundreds of thousands of three-dimensional structures of a biological molecule, leading to needle-in-a-haystack problems that have natural formulations under a machine learning framework. The work in this dissertation aims to advance research on visualization, summarization, and recognition problems on molecular structure spaces by exploiting the molecular energy landscape associated with a structure space. Specifically, graph-based representations and graph mining of such landscapes have been leveraged to summarize the molecular energy landscape. The work presented in this thesis shows that mining of the energy landscape corresponding to molecular structure space promises to advance our ability to recognize biologically-active structures/needles in a vast space/haystack via supervised and unsupervised learning. While the focus of this work is on molecular structure data, the approaches we presented here are of general utility beyond computational biology to any domain where the goal is to reveal informative organizations of high-dimensional, spatial data that support prediction.

关键词： Artificial intelligence   Computer science  

摘要： The internet and its various social media platforms allow for the rapid spread of information. While this has a number of benefits for society, it also facilitates proliferation of incorrect or misleading information. Especially in recent years, concerns have been raised over ``fake news'', factually incorrect claims, and biased or heavily slanted news articles, regarding the potential impact on society and resulting polarization. The fields of machine learning and natural language processing contain building blocks and tools with the potential to help address some of these issues. This thesis explores the application of weak supervision of machine learning models to predict the bias and reliability of news articles based on their content. This weak supervision takes place using a dataset of source-level labels of mainstream news outlets and a large collection of news article texts from those outlets. We compare different methods of embedding the text and report on model generalizability to a dataset containing individual article-level bias and reliability labels.