教学课程资源库 更多>>

关键词： Computer science  

摘要： Data analytics stands to benefit from the increased availability of datasets that are held without their conceptual relationships being explicitly known. When collected, these datasets form a data lake from which, by processes like data preparation, also known as data wrangling, specific target datasets can be constructed that enable value-adding analytics. Given the potential vastness and heterogeneity of data lakes, obtaining value from such targets often requires significant prior effort in preparing the data for analysis. For example, data wrangling is reported to take as much as 80% of the time of data scientists. The issue then arises of how to decrease this cost. This thesis investigates what makes data preparation costly and how data preparation can become more cost-effective through automation. Specifically, this thesis inquires into two challenges that have been insufficiently covered by the state-of-the-art, viz., how to automatically pull out of the data lake those datasets that might contribute to wrangling out a given target, and how to automatically homogenise the representation of their instance value. We refer to the former as the problem of dataset discovery and to the latter as the problem of format transformation. This thesis contributes effective and efficient solutions to both problems. The work described in this thesis should be of interest to researchers and professionals in the areas of data analysis and data wrangling, who, in the process of preparing the data for analysis, confront themselves with heterogeneously represented data originating from many autonomous sources.

关键词： Electrical engineering   Information science   Computer science  

摘要： Signature searching is the process of finding a signature “pattern” of interest in a data file. Signature searching can be found in many applications such as radar, sensor and data processing. The objective of this research is to provide approaches for optimal signature searching. The plan is to study the signature searching optimization problem from multiple aspects such as mathematical programming, divisible load scheduling and topology. The focus of this research is sensor applications. An active sensor emits signals for detection and information acquisition purposes. Optimal sensing means the best sensing quality with the least time and energy cost, which allow processing more data. Intuitively, optimal sensing leads to optimal signature searching. This dissertation presents novel work by using an integer linear programming “algorithm” to achieve the optimal sensing by selecting the best possible number of signals of a type or a combination of multiple types to ensure the best sensing quality possible considering all given constraints. Then, a solution based on a heuristic algorithm is implemented to improve the performance. Finally, an optimal processing method is presented. The considered processing methods are local computing, cloud computing or a combination of both methods.

关键词： Computer Science   machine learning   neural networks   computer music   supervised learning   interactive machine learning   audio synthesizers  

摘要： Musicians who perform with electronic synthesizers often adjust synthesis parameters during live performance to achieve more expressive sounds. Enabling the performer to teach a computer to make these adjustments automatically during performance frees the performer from this responsibility, while maintaining an expressive sound in line with the performer’s musical vision. We have created a machine learning system called Larasynth that can be trained by a musician to make these parameter adjustments in real-time during live performances. Larasynth is trained using examples in the form of MIDI files created by the user. Learning is achieved using Long Short-Term Memory (LSTM) recurrent neural networks. To accomplish this, we have devised a set of features which capture the state of the synthesizer controller at regular intervals and are used to make regular predictions of parameter values using an LSTM network. To achieve sufficient generalization during training, transformations are applied to the training data set before each training epoch to simulate variations that may occur during performance. We have also created a new lightweight LSTM library suitable for small networks under real-time constraints. In this thesis we present details behind Larasynth’s implementation and use, and experiments that were performed to demonstrate Larasynth’s ability to learn behaviors based on different musical situations.

关键词： Computer science  

关键词： Computer science  

摘要： Hardware solutions for isolated execution aim to protect vulnerable applications in environments of untrusted software, including potentially compromised system software. One prominent industry implementation of hardware-supported isolated execution is Intel Software Guard Extensions (SGX). Security-critical applications can be run inside isolated enclaves, preventing outside software from tampering with enclave data and code. For protecting against physical attacks on memory, SGX encrypts and integrity checks the enclave data when the data is moved between the processor and memory. Due to the encryption overhead, enclave applications experience a higher memory overhead than non-enclave applications. However, current cache replacement decisions do not consider the cost of bringing enclave data into the cache. Since industry implementations of hardware isolated execution are relatively new, current processor simulators do not provide a way to evaluate designs that surround hardware isolated execution. In this thesis, we first develop a tool to evaluate new cache designs for secure enclave execution. Then we design a simple cache replacement algorithm that considers the encryption overhead of enclave data, and use our tool to evaluate the performance of our replacement algorithm.

关键词： Computer Science  

摘要： Many scientific applications rely on the evaluation of elementary transcendental functions (e.g. log(x), 1/x, √x, ex). Software math libraries are a popular approach for realizing such functions, and frequently use series expansion and/or lookup-table-based (LUT-based) methods. However, software approaches necessarily suffer from the traditional overheads of fetching/decoding instructions, limited cache sizes, and so on. To this end, we present hardware accelerators for such functions that deliver high computational throughput, high accuracy and a small circuit. We implement the reciprocal (1/x) and square root (√x) functions into pipelined hardware accelerators on FPGA. The proposed accelerators are designed with iterative, and LUT-based algorithms. Here, the LUT-based algorithm uses approximately 1 KB LUT along with a degree-2 polynomial interpolation. All algorithms are specified using C language, and synthesized into RTL with the LegUp HLS. In an experimental study, we compare our LUT-based accelerators with IP cores from the Intel/Altera FPGA vendor. Results show that our LUT-based accelerators offer considerably better area usage, while Intel/Altera IP cores operate at a modestly higher throughput. Both ours and Intel/Altera IP cores achieve 1 ULP error. The LUT-based algorithm is generic in the sense that it could be used to implement an entire library of single-precision elementary functions into high-performance hardware accelerators. We also implement a 32-bit integer RISC-V multi-cycle processor on FPGA, which consists of 39 user instructions. The processor is specified using C language, and synthesized into RTL with the LegUp HLS. Custom testing programs are deveped to verify if each instruction adheres to the RISC-V specification. We demonstrate that through changes to the C specification and HLS constraints, RISC-V processors with different performance/area trade-offs can be explored rapidly. One implementation of the multi-cycle processor uses 795 ALM

关键词： Computer science  

摘要： An electrocardiogram (ECG) is used to diagnose the functionality of the heart since the ECG is the easiest and least expensive method to examine heart disease. The ECG is a composite analysis of the P wave, the QRS complex, and the T wave, which indicate the heart's electrical activity over time. In which the QRS complex is a combination of the Q wave, the R wave and the S wave. In this study, a method for ECG classification is proposed based on a combination of Support Vector Machines (SVM). The method is focused on the time intervals and the amplitude of waves for the classification of ECG beats. Various features have been used to extract the information from the ECG waves including, the wavelet descriptor, the higher-order statistics, and the local binary pattern. In this work, we suggested to develop separate models for each of these features rather than combining all these features and providing them into a single SVM model. Then the results of all the SVM models are combined with multiplication, summation, and majority rules. The MIT-BIH arrhythmia dataset, which consists of irregular and regular beats, is used to perform the training and testing of models. In this work, we proposed the use of the combination of several SVMs, which showed better results in comparisons with approaches based on a single SVM.

关键词： Geographic information science and geodesy   Computer science  

摘要： In spatio-temporal systems, precise location data is desirable but often not available due to obfuscation, privacy, hardware inaccuracies, and other factors. Progress has been made in research which deals with the uncertainty of moving objects’ location data. However, much of the existing work does not always consider factors such as constraints imposed by the topology of road networks, and harmonic integration between past movements, current, and prospective imprecise positions. In this thesis, we propose an approach that utilizes time, distance, and connectivity constraints of a road network to infer a moving object’s past, present, and future locations more precisely when its exact location data is not available. The experimental results using real GPS trajectories confirm the efficiency of our proposed solution for reducing uncertainty and inferring historical, and future locations.

关键词： Computer science  

摘要： Distributed systems are pervasively demanded and adopted in nowadays for processing data-intensive workloads since they greatly accelerate large-scale data processing with scalable parallelism and improved data locality. Traditional distributed systems initially targeted computing clusters but have since evolved to data centers with multiple clusters. These systems are mostly built on top of homogeneous, tightly integrated resources connected in high-speed local-area networks (LANs), and typically require data to be ingested to a central data center for processing. Today, with enormous volumes of data continuously generated from geographically distributed locations, direct adoption of such systems is prohibitively inefficient due to the limited system scalability and high cost for centralizing the geo-distributed data over the wide-area networks (WANs). More commonly, it becomes a trend to build geo-distributed systems wherein data processing jobs are performed on top of geo-distributed, heterogeneous resources in proximity to the data at vastly distributed geo-locations. However, critical challenges and mechanisms for efficient execution of data-intensive applications in such geo-distributed environments are unclear by far. The goal of this dissertation is to identify such challenges and mechanisms, by extensively using the research principles and methodology of conventional distributed systems to investigate the geo-distributed environment, and by developing new techniques to tackle these challenges and run data-intensive applications with efficiency at scale. The contributions of this dissertation are threefold. Firstly, the dissertation shows that the high level of resource heterogeneity exhibited in the geo-distributed environment undermines the scalability of geo-distributed systems. Virtualization-based resource abstraction mechanisms have been introduced to abstract the hardware, network, and OS resources throughout the system, to mitigate the underlying res

关键词： Applied mathematics   Computer science   Acoustics  

摘要： Reducing aircraft noise is a major objective in the field of computational aeroacoustics. When designing next generation quiet and environmentally friendly aircraft, it is important to be able to accurately and efficiently predict the acoustic scattering by an aircraft body from a given noise source. Acoustic liners are an effective tool for aircraft noise reduction and are characterized by a frequency-dependent impedance. Converted into the time-domain using Fourier transforms, an impedance boundary condition can be used to simulate the acoustic wave scattering by geometric bodies treated with acoustic liners. This work considers using either an impedance or an admittance (inverse of impedance) boundary condition to allow for acoustic scattering problems to be modeled with geometries consisting of both un-lined and lined surfaces. Three acoustic liner models are discussed: the Extended Helmholtz Resonator Model, the Three-Parameter Impedance Model, and the Broadband Impedance Model. In both the Helmholtz and Three-Parameter models, liner impedance is specified at a given frequency, whereas the Broadband model allows for the investigation of multiple frequencies simultaneously. The impedance and admittance boundary conditions for acoustic liners are derived for each model and coupled with a time-domain boundary integral equation. The scattering solution is obtained iteratively using a boundary element method with constant spatial and third-order temporal basis functions. Time-domain boundary integral equations are unfortunately prone to numerical instabilities due to resonant frequencies resulting from non-trivial solutions in the interior domain. When reformulated with the Burton-Miller method, the instabilities are eliminated. Using a Burton-Miller reformulation, the stability of the boundary element method assuming a liner boundary condition is assessed using eigenvalue analysis. The stability of each liner model is discussed, and it is shown that the Three-Param