教学课程资源库 更多>>

关键词： Computer Science   Communication Middlewares   Optimized Design   Process Mapping   HPC   Multi-core   Many-core   Parallel Computing   High-End Systems   Cloud Computing   Zero-copy Communication   Shared Address Space Communication   Emerging Architectures   MPI  

摘要： Modern High-Performance Computing (HPC) systems are enabling scientists from different research domains such as astrophysics, climate simulations, computational fluid dynamics, drugs discovery, and others, to model and simulate computation-heavy problems at different scales. In recent years, the resurgence of Artificial Intelligence (AI), particularly Deep Learning (DL) algorithms, has been made possible by the evolution of these HPC systems. The diversity of applications ranging from traditional scientific computing to the training and inference of neural-networks are driving the evolution of processor and interconnect technologies as well as communication middlewares. Today’s multi-petaflop HPC systems are powered by dense multi-/many-core architectures and this trend is expected to grow for next-generation systems. This rapid adoption of these high core-density architectures by the current- and next-generation HPC systems, driven by emerging application trends, are putting more emphasis on the middleware designers to optimize various communication primitives to meet the diverse needs of the applications. While these novelties in the processor architectures have led to increased on-chip parallelism, they come at the cost of rendering traditional designs, employed by the communication middlewares, to suffer from higher intra-node communication costs. Tackling the computation and communication challenges that accompany these dense multi-/manycores garner special design considerations. Scientific and AI applications that rely on such large-scale HPC systems to achieve higher performance and scalability often use Message Passing Interface (MPI), Partition Global Address Space (PGAS), or a hybrid of both as underlying communication substrate. These applications use various communication primitives (e.g., point-to-point, collectives, RMA) and often use custom data layouts (e.g., derived datatypes), spending a fair bit of time in communication and synchronization. The pe

关键词： Computer science  

摘要： Along with other new concepts, cloud computing is an emergent paradigm embraced by some organizational leaders. Although cloud computing is advantageous for an organization, some business leaders are reluctant to incorporate this new technology. This qualitative Delphi study explored why some information technology business leaders incorporated cloud computing architecture platforms as well as provided some possible reasons why others have not. In this study, the researcher addressed the surge in demand for business leaders to incorporate cloud computing architecture platforms for technological advancement. The modernization and accelerated rates of technological innovation require organization leaders who are highly flexible, adaptable, scalable, and on-demand with their computing skills. The chosen conceptual framework used to guide this study was Tornatzky and Fleischer’s technology, organization, and environment framework. The researcher used this framework to explore why certain information technology experts incorporated cloud computing architecture platforms at the research site. Prior to the start of the study, three senior executives from the research site participated in a pilot study and approved both questionnaires. The panelists chosen for this study were business leaders who had the authority to make purchases, including cloud computing, for their departments. These panelists reached a consensus in the second round of questions and agreed on eight themes: autonomy, competitive pressure, cost advantage, management support, organizational alignment, security, speed, and training. Before adopting cloud computing for a department or an entire organization, the panelists recommended the involved employees be provided with enough time for thorough training by a skilled information technologist.

关键词： Electrical engineering   Computer science  

摘要： In computing systems, memories are storage devices that keep instructions and data. To balance the performance and cost of modern computing systems, a hierarchy of memories---registers, caches, main memory, and storage---with different speeds and densities (costs) are utilized. In the memory hierarchy, a major performance gap between main memory and storage has become the bottleneck for many data-centric applications. Bridging this performance gap in the hierarchy has been the motivation of the development of many new memories. Among these new memories, resistive switching random access memory (RRAM) is a promising candidate for the next-generation non-volatile memory. Due to its simple structure, direct over-write, bit-alterability, fast speed, and low energy consumption, RRAM shows great potential for being used as both off-chip and on-chip binary digital memories. Additionally, the analog conductance modulation of RRAM allows it to be used as analog weights for machine learning specialized and neuromorphic computing hardware. This thesis presents the analysis, modeling, and characterization of RRAM that enable it to be used in both future digital memory systems and machine learning specialized/neuromorphic computing hardware.

关键词： Environmental studies   Artificial intelligence   Computer science  

摘要： Wind power is a renewable and abundant source of energy. Wind farm locations maximize wind power through placement in high-altitude areas and continual improvements in wind turbine blade design. Blades are vital to wind turbine performance, but are susceptible to damages from weather, irregular loading, malfunction, or other catastrophes. Blade damage is associated with a reduction in both turbine lifespan as efficiency as well as increased safety concerns, maintenance costs, and sensor errors. Due to the prohibitive cost of installing physical detectors on blades, data-driven approaches are increasingly popular for detecting blade failures. This study aimed to identify blade failures through analysis of multivariate time series sensor data monitored by a Supervisory Control and Data Acquisition (SCADA) system. Several machine learning and statistical methods were attempted to forecast sensor trends and classify turbines into groups of those with and without probable blade failures. The results suggested that supervised learning techniques such as support vector machines (SVM), long short-term memory (LSTM) and other neural networks with autoencoders may outperform other blade failure classification methods. The study compared linear kernel and radial basis function (RBF) kernels for SVM. A simple LSTM architecture correctly classified approximately two-thirds of turbines with previous failures and forecasted failures up to two days in advance. A dense autoencoder model provided promising results on one studied wind turbine and requires further study into generalization potential. Finally, vector autoregression (VAR) predicted sensor trends with fair accuracy and may be combined with LSTM or other neural networks to improve performance.

关键词： Computer science  

摘要： A method for estimating the Camera Response Function (CRF) of an electronic motion picture camera is presented in this work. The accurate estimation of the CRF allows for proper encoding of camera exposures into motion picture post-production workflows, like the Academy Color Encoding Specification (ACES), this being a necessary step to correctly combine images from different capture sources into one cohesive final production and minimize non-creative manual adjustments. Although there are well known standard CRFs implemented in typical video camera workflows, motion picture workflows and newer High Dynamic Range (HDR) imaging workflows have introduced new standard CRFs as well as custom and proprietary CRFs that need to be known for proper post-production encoding of the camera footage. Current methods to estimate this function rely on the use of measurement charts, using multiple static images taken under different exposures or lighting conditions, or assume a simplistic model of the function’s shape. All these methods become problematic and tough to fit into motion picture production and post-production workflows where the use of test charts and varying camera or scene setups becomes impractical and where a method based solely on camera footage, comprised of a single image or a series of images, would be advantageous. This work presents a methodology initially based on the work of Lin, Gu, Yamazaki and Shum that takes into account edge color mixtures in an image or image sequence, that are affected by the non-linearity introduced by a CRF. In addition, a novel feature based on image noise is introduced to overcome some of the limitations of edge color mixtures. These features provide information that is included in the likelihood probability distribution in a Bayesian framework to estimate the CRF as the expected value of a posterior probability distribution, which is itself approximated by a Markov Chain Monte Carlo (MCMC) sampling algorithm. This allows for a m

关键词： Computer science  

摘要： In this doctoral dissertation we focus on the most common method of authentication, the username-password combination. The reason for the frequent use of this authentication mechanism is its simplicity and low cost of implementation. Although passwords are so useful, they have many problems. Morris and Thompson, for the first time almost four decades ago, found that textual passwords were a weak security point of information systems. They have come to the conclusion that users are one of the biggest threats to information system’s security. Since then, we face these problems on a daily basis. Users do not perform the behaviours they need to be done in order to stay safe and secure, although they are aware of the security issues. Because this is a research area that security experts have been dealing with for a long time, in this dissertation we wanted to identify problems related to textual passwords and possible suggested solutions. For this purpose, we first performed a systematic literature review on textual passwords and their security. In doing so, we wanted to evaluate the current status of passwords in terms of their strength, ways of managing passwords, and whether users are still the “weakest link”. We found that one of the less researched solutions is proactive password checking. A proactive password checker could filter out the passwords that are easy-to-guess and only let through the passwords that are harder to guess. In order for a proactive password checking to be more effective, it is necessary for the checker to be able to check the probability that a certain password will be selected by the user. For this purpose, the better password checkers usually use certain tools to calculate password probability i.e., password strength. To find out which method is most suitable for calculating password strength, we have looked at similar solutions throughout history. We have found that Markov models are one of the most common methods used for password strengt

关键词： 詹姆斯·W·约翰逊   自我表征   黑人美学   黑肤色   黑人文化   James W.Johnson   self-representation   black aesthetics   black skin color   black culture  

摘要： 詹姆斯·W·约翰逊是二十世纪最重要的美国非裔作家和批评家，他在文学作品和文学、文化批评中阐释和肯定黑肤色和黑人文化的美丽和力量，形成独特的黑人美学。他对黑肤色和黑人文化之美的阐释和肯定超越了自我表征，重构美国非裔族群的社会文化形象和种族身份，鼓励非裔同胞提升自己的种族和文化自豪感，倡导黑人艺术家汲取黑人民间文化、创作具有普遍价值和永恒意义的作品，创作真正的“有意识的黑人艺术”。约翰逊认为，美国非裔文化在积极融入美国文化、世界文化的同时，应彰显其自身的独特魅力和价值。

关键词： Spatiotemporal phenomena   Recurrent neural networks   Licenses   Semisupervised learning   Convolution   Computer science   Behavior features   feature   graph features   recommendation systems   text features  

摘要： Recommendation systems have become an important field of research in computer science and physics. In recent years, breakthroughs have been achieved in social, biological, and research cooperation networks. With the popularization of big data and deep learning technology development, graph structures are increasingly being used to represent large-scale and complex data in the real world. In this paper, we reviewed the progress made in recommendation systems research in the past 20 years and comprehensively classified recommender systems based on the heterogeneous input features. We introduced layering in the classification of recommendation systems. Furthermore, we proposed a new hierarchical classification model of recommendation systems divided into three layers: feature input, feature learning, and output layers. In the feature learning layer, existing recommendation systems were divided into graph-based, text-based, behavior-based, spatiotemporal-based, and hybrid recommendation systems. Additionally, we provided evaluation index, open-source implementation, experimental comparison and the relative merits for each recommendation method. Subsequently, future development directions of recommendation systems are discussed.

关键词： Computer science  

摘要： Object detection has gained much attention in recent years because of its ability to localize and classify the objects in videos and images that can be incorporated into many applications. Traditional object detection algorithms need substantial computational resources to build a model. There is an increasing demand for a practical approach to constructing object detection models adapted to the local context, limited computing resources, and application logics while supporting real-time inferencing without degrading accuracy and performance. In this thesis, we proposed a distributed-collaborative framework to build practical object detection models. The framework is based on a novel approach for collaborative group inferencing that is designed with a single-class-single-model mechanism for multiple objects in a distributed manner. For useful grouping, we made use of the intraclass correlation from existing models during inferencing. Results from the case studies with Pascal VOC 2007 and our data collected through Google Street View showed that the proposed model significantly improved performance while making it potentially suitable for customized application building with limited computing resources. A prototype for the proposed model has been built, and a neighborhood application has been demonstrated to validate the proposed work.

关键词： Computer science  

摘要： The ability of deep neural networks (DNNs) to achieve state-of-the-art performance on complicated tasks has increased their adoption in safety-critical cyber physical systems (CPS). However, DNNs are susceptible to a variety of attacks, including 1) adversarial attacks where imperceptible perturbations of the input can induce mispredictions, 2) data poisoning attacks where the training data is manipulated to encode a malicious backdoor key, and 3) physically realizable point anomalies that exploit corner cases of the model. Simple two-layer radial basis function (RBF) networks are known to exhibit low confidence on point anomalies and adversarial images; however, until recently deeper RBFs have been difficult to train on complicated data. This paper extends recent advancements in deep RBFs and presents novel methods to address major security threats of DNNs. First, we show that in a self-driving task a deep RBF version of NVIDIA's DAVE-II architecture can reliably detect physically realizable point anomalies and assess its sensitivity. Second, we show that deep RBFs are less susceptible to data poisoning attacks and describe a novel algorithm to clean sparsely poisoned data without relying on a verified, clean data set. Finally, we train a deep RBF based on the InceptionV3 architecture and evaluate its robustness on a range of adversarial attacks, including FGSM, I-FGSM, Deepfool, Carlini & Wagner, and Projected Gradient Descent. Most importantly, we demonstrate that deep RBFs should replace traditional classifiers in CPS tasks due to their robustness against a variety of security threats and their ability to simultaneously achieve production-grade performance.