教学课程资源库 更多>>

关键词： Computer science   Acoustics   Electrical engineering   Artificial Reverberation   Augmented Reality   Feedback Delay Network   Machine Learning   Signal Processing   Virtual Reality  

摘要： For virtual and augmented reality applications, it is desirable to render audio sources in the space the user is in, in real-time without sacrificing the perceptual quality of the sound. One aspect of the rendering that is perceptually important for a listener is the late-reverberation, or "echo", of the sound within a room environment. A popular method of generating a plausible late reverberation in real-time is the use of Feedback Delay Network (FDN). However, its use has the drawback that it first has to be tuned (usually manually) for a particular room before the late-reverberation generated becomes perceptually accurate. In this thesis, we propose a data-driven approach to automatically generate a pre-tuned FDN for any given room described by a set of room parameters. When combined with existing method for rendering the direct path and early reflections of a sound source, we demonstrate the feasibility of being able to render audio source in real-time for interactive applications.

关键词： Arithmetic circuit   Multiplication   Proof Complexity   SAT solving   Verification   Computer science   Mathematics   Logic   Computer science and engineering  

摘要： Automated theorem provers have long struggled to efficiently reason about bit-precise properties of integer multiplication. Despite major advances in the efficiency of automated reasoning, from the Binary Decision Diagrams of the 1980s to the SAT solvers of today, integer multiplication has persisted as a major bottleneck in hardware and software verification. In this thesis, we use proof complexity to pave a new path towards verifying nonlinear integer arithmetic. We propose that pseudo-Boolean solvers equipped with cutting planes reasoning have the potential to combine the complementary strengths of the existing SAT and algebraic approaches while avoiding their weaknesses. We present several results on the proof complexity of fundamental multiplier identities. In the resolution proof system, we construct polynomial size proofs for degree two ring identities, refuting a widely believed conjecture that such proofs must be exponentially large. In the polynomial calculus proof system, we give optimal length proofs for word-level ring identities. But we also show that extracting simple bit-level consequences from a word-level property can require an exponentially large polynomial calculus derivation. In the cutting planes proof system, we give optimal length proofs for a large class of degree two identities, at both the word-level and the bit-level. We then present experiments testing the CDCL SAT solving approach (corresponding to resolution) and the pseudo-Boolean approach (corresponding to cutting planes) that uncover the potential of using pseudo-Boolean solvers to efficiently reason with mixtures of arithmetic and bit-level constraints. We demonstrate that pseudo-Boolean solvers can verify, at both the word-level and bit-level, the commutativity of a multiplier as well as the equivalence of different multiplier architectures. We also find examples of simple nonlinear bit-vector inequalities that are intractable for current bit-vector and SAT solvers but easy for p

关键词： Computer engineering   Computer science  

摘要： The breakthroughs in multi-layer convolutional neural networks (CNNs) have caused significant progress in the applications of image classification and recognition. The size of CNNs has continuously increased to improve their prediction capabilities on various applications, and it has become increasingly costly to perform the required computations. In particular, CNNs involve a large number of multiply-accumulate (MAC) operations, and it is important to minimize the cost of multiplication as it requires most computational resources. This dissertation proposes cost-efficient approximate log multipliers, optimized for performing CNN inferences. Approximate multipliers have reduced hardware costs compared to the conventional multipliers but produce products that are not exact. The proposed multipliers are based on Mitchell's Log Multiplication that converts multiplications to additions by taking approximate logarithm. Various design techniques are applied to Mitchell Log Multiplier, including fully-parallel LOD, efficient shift amount calculation, and exact zero computation. Additionally, the truncation of the operands is studied to create the customizable log multiplier that further reduces energy consumption. This dissertation also proposes using the one's complements to handle negative numbers to significantly reduce the associated costs while having minimal impact on CNN performances. The viability of the proposed designs is supported by the detailed formal analysis as well as the experimental results on CNNs. The proposed customizable design at w = 8 saves up to 88% energy compared to the exact fixed-point multiplier at 32 bits with just a performance degradation of 0.2% on AlexNet for the ImageNet ILSVRC2012 dataset. The effects of approximate multiplication are analyzed when performing inferences on deep CNNs, to provide a deeper understanding of why CNN inferences are resilient against the errors in multiplication. The analysis identifies the critical factors in

关键词： African American Studies   African Americans   African History   African Studies   African Literature   American Studies   Black Studies   Black History   Cultural Anthropology   Gender Studies   Gender   Womens Studies   Ethnic Studies   Natural Hair   Hair Straightening   Politics   African   Blackness   Ghana   Ghanaian Women   Hair   Aesthetics   Beauty   Afro-textured Hair  

摘要： Hair straightening and the usage of hair extensions, particularly chemical relaxing and `Brazilian’ weaves are common and, in many cases, preferable over natural Afro-textured hair among African women in both the continent and the diaspora. The practice, which has been extensively studied from the African American perspective, is largely explained as (1) a result of internalized racism and self-hate and (2) simply stylization, versatility, and choice. However, even though diasporic and continental Africans share similar hairstyling practices, their historical and cultural situations are not the same. This study looks at Ghanaian women’s hair culture from a whiteness-decentered approach. The goal of the study was to research African women’s hair norms from continental African women’s perspective and find out whether their hair culture could be explained as a result of internalized racism and self-hate or stylization. The study interviewed thirty women. It asks why do they relax their hair and wear hair extensions: whether they are pursuing `white beauty’, or they have other reasons why they alter their hair. It found out that Ghanaian women do not relax their hair and wear hair extensions as a result of `mental slavery,’ and they do not hate their hair, or perceive white women’s appearance iconic. It also found that hairstyling among Ghanaian women is not simply for stylization; there are specific reasons why they relax their hair and wear weaves. These reasons include the attachment of privileges and benefits to straight hair and stereotypes and consequences to coily hair in Ghanaian society. Contrary to earlier studies that explain African women’s negative relationship with their hair as a result of colonialism and Euro-American influence on the continent, the study found out that negative attitudes displayed toward hairstyles like (dread)locks in many African societies stem from pre-colonial traditional cultural norms.

摘要： In education, many assessments boil down to getting the correct solution or necessary result to receive credit. This end goal mentality, in turn, influences how educators transfer knowledge to students. For example, some educators may present or walk through completed solutions. However, continually displaying and using worked-out solutions to teach can quickly become an obstruction to learning. In computer science, a significant amount of learning occurs while fixing the errors that litter the pathway from a blank page to a working solution. This dissertation establishes a methodology for teaching from student errors in computer science. The first part of the dissertation establishes how we developed over fifty lightweight exercises to integrate into a ten-week course without content replacement. Using past research in computer science, education theory, and cognitive load theory, we developed and refined a standard exercise structure that incorporates student submissions containing erroneous code, past student solutions presented during student-instructor interactions, and instructor feedback. Collectively, our core exercises are known as "What's Wrong With My Code" exercises. Next, we evaluated the "What's Wrong With My Code" exercises in three distinct ways. First, we performed a study to assess student improvement when using our exercises in place of current course activities (e.g., CodeLab). Second, we analyzed the differences in student error encounters for an entire term by comparing error counts in prior course offerings to offerings with our exercises integrated into the weekly course workload. Lastly, we evaluated student self-efficacy improvements over an entire term by comparing offerings with and without our exercises. In each of the studies, our exercises proved beneficial with increased student performance (with effect sizes of 0.56 and 0.42), increased self-efficacy (p-value < 0.05), and diminished student error encounter rates. Finally, we used our

关键词： Mathematics   Engineering   Computer science  

摘要： The design of any engineering component requires robust analysis using numerical methods like the Finite Element Method. Of paramount importance is to develop convergent formulations that can achieve accurate estimates for the solution at cheaper computational costs. We investigate a method for improving the accuracy of the stress predicted from models using the mean-strain finite elements recently proposed by Krysl and collaborators [IJNME 2016, 2017]. In state-of-the-art finite element programs, the stress values at the integration points are commonly post-processed to obtain nodal stresses. The mean stresses are element-wise constant, and hence the nodal values obtained from the mean stresses tend to be less accurate. The proposed method post-processes the uniform stress in each element in combination with a linearly-varying stabilization stress field to compute more accurate nodal stresses. Selected examples are presented to demonstrate improvements achievable with the proposed methodology for hexahedral and quadratic tetrahedral mean-strain finite elements. The nodally integrated formulations exhibit spuriousness in dynamic analyses (such as in modal analysis). Previously proposed methods involved a heuristic stabilization factor, which may not work for a large range of problems, and a uniform stabilization was used over all the finite elements in the mesh. The method proposed herein makes use of energy-sampling stabilization. The stabilization factor depends on the shape of the element and appears in the definition of the properties of a stabilization material. The stabilization factor is non-uniform over the mesh, and can be computed to alleviate shear locking, which directly depends on the aspect ratios of the finite elements. The nodal stabilization factor is then computed by volumetric averaging of the element-based stabilization factors. Energy-sampling stabilized nodally integrated elements (ESNICE) tetrahedral and hexahedral are proposed. We demonstrate

关键词： Computer science  

摘要： Peer review is a commonly used tool to manage large classes. It allows students to grade and provide feedback to each other based on rubrics provided by instructors. Peer review has been proved to be effective in improving students' learning outcomes by many research. During providing peer review, students are exposed to more solutions to the same question, which makes students more innovative. Also, peer review reduces instructors' or TAs' workload on providing feedback despite teaching big classes. However, peer review still has shortcomings. First, peer review increases students' workload on reviewing other submissions. To collect enough peer reviews for submissions, each peer usually needs to review several submissions per homework. Some peers may spend more time on peer reviewing than finishing the homework. Second, since peers typically lack the subject matter mastery of the instructor, peer grades exhibit both bias and variance, which makes consensus grade estimation a challenging task. This dissertation addresses these limitations of peer review using Natural Language Processing techniques. Specifically, this dissertation proposes novel neural models that predict important parts of submissions. By doing so, peers can save time by only focusing on the essential parts. Our models take advantage of textual reviews and review labels to improve prediction accuracy. This dissertation also proposes novel peer grading methods, which enhance peer grading accuracy by using historical instructor grades to estimate peer bias, and textual review comments to estimate review quality.

关键词： Computer science  

摘要： In our current generation, we are very much habituated to many mobile services like communication, e-commerce, etc. In mobile communication services, SMS’s (Short Message Service's) are very common and important services which we are using for personal purposes and profession. In these services, some messages may cause spam attacks which are traps to users to access their personal information or attracting them to purchase a product from unauthorized websites. It is very easy for companies to send any information or service or alert to their customers/users with these SMS API. Based on these services it is also possible for sending spam messages. Therefore, in this system, we are using advanced Machine Learning concepts for detection of the spam filtering in the SMS. In this system, we are importing the dataset from UCI repository and for spam SMS detection we are implementing machine learning classifiers like Support Vector Machine (SVM), K-Nearest Neighbor (KNN), and Naïve Bayes (NB) algorithms and with their metrics like accuracy, precision, recall, and f-score. We calculate performances between these algorithms and present the experiment results with visualization techniques and try to find the best algorithms for spam SMS detection.

关键词： Computer science   Computer science and engineering  

摘要： The explosion of data volume and ever-increasing speed of accelerators shift the bottleneck of large-scale distributed training tasks from computation to communication. We observe significant pressure on the communication backends of various mainstream learning systems in multiple environments when running such tasks. Achieving efficient large scale learning relies on more effective communication planes. We provide detailed analysis that root-causes the bottlenecks affecting the communication efficiency of these systems in the context of different environments. We pinpoint such bottlenecks from the software, hardware and network infrastructure stacks. We show how these obstacles can be overcome with a systematic codesign of a streamlined communication stack, a balanced hardware and cluster configuration with the distributed training workload, together with awareness of network topology and environment. We show this series of approaches, named Parameter Box, Parameter Hub, Parameter Link along with Cloud Collectives, accelerate distributed training from small clusters to datacenters and all the way to the commercial clouds while providing varying degrees of customization to suit different needs.

关键词： information science \u0026 technology   computer science   information science \u0026 technology   computing: general   business   management \u0026 finance   business \u0026 management   project management  

ISBN: (数字)9781636390390 ISBN: (纸本)9781636390383;9781636390406

摘要： Recent data shows that 87% of Artificial Intelligence/Big Data projects don’t make it into production (VB Staff, 2019), meaning that most projects are never deployed. This book addresses five common pitfalls that prevent projects from reaching deployment and provides tools and methods to avoid those pitfalls. Along the way, stories from actual experience in building and deploying data science projects are shared to illustrate the methods and tools. While the book is primarily for data science practitioners, information for managers of data science practitioners is included in the Tips for Managers sections.