The latest development of neural word segmentation is governed by bi-directional Long Short-Term Memory Networks (Bi-LSTMs) that utilize Recurrent Neural Networks (RNNs) as standard sequence tagging models, resulting ...
详细信息
The latest development of neural word segmentation is governed by bi-directional Long Short-Term Memory Networks (Bi-LSTMs) that utilize Recurrent Neural Networks (RNNs) as standard sequence tagging models, resulting in expressive and accurate performance on large-scale dataset. However, RNNs are not adapted to fully exploit the parallelism capability of Graphics Processing Unit (GPU), limiting their computational efficiency in both learning and inferring phases. This paper proposes a novel approach adopting Iterated Dilated Convolutional Neural Networks (ID-CNNs) to supersede Bi-LSTMs for faster computation while retaining accuracy. Our implementation has achieved state-of-the-art result on SIGHAN Bakeoff 2005 datasets. Extensive experiments showed that our approach with ID-CNNs enables 3X training time speedups with no accuracy loss, achieving better accuracy compared to the prevailing Bi-LSTMs. Source code and corpora of this paper have been made publicly available on GitHub(1).
If learning methods are to scale to the massive sizes of modern data sets, it is essential for the field of machine learning to embrace parallel and distributed computing. Inspired by the recent development of matrix ...
详细信息
If learning methods are to scale to the massive sizes of modern data sets, it is essential for the field of machine learning to embrace parallel and distributed computing. Inspired by the recent development of matrix factorization methods with rich theory but poor computational complexity and by the relative ease of mapping matrices onto distributed architectures, we introduce a scalable divide-and-conquer framework for noisy matrix factorization. We present a thorough theoretical analysis of this framework in which we characterize the statistical errors introduced by the "divide" step and control their magnitude in the "conquer" step, so that the overall algorithm enjoys high-probability estimation guarantees comparable to those of its base algorithm. We also present experiments in collaborative filtering and video background modeling that demonstrate the near-linear to superlinear speed-ups attainable with this approach.
The mining task of outlier detection is essential in many expert and intelligent systems exploited in a wide range of applications, from intrusion detection to molecular biology. In some of such applications the abili...
详细信息
The mining task of outlier detection is essential in many expert and intelligent systems exploited in a wide range of applications, from intrusion detection to molecular biology. In some of such applications the ability to process large amounts of data in a very short time can be critical, for instance in intrusion and fraud detection. This paper explores a solution for the optimisation of an exact, unsupervised outlier detection method by avoiding unnecessary computations, and therefore reducing the running time and making the method usable also in settings where response times are crucial. In particular, we enhance the SolvingSet-based approach by using a mechanism that exploits the knowledge learned during the algorithm execution and avoids a large amount of distance computations. We demonstrate the strength of the proposed solution, named FastSolvingSet, through both theoretical and experimental analysis. (C) 2020 Elsevier Ltd. All rights reserved.
The performance demands of modem control and signal processing systems is increasing beyond the capacity of conventional sequential processors, requiring parallel processing solutions to satisfy the real-time requirem...
详细信息
The performance demands of modem control and signal processing systems is increasing beyond the capacity of conventional sequential processors, requiring parallel processing solutions to satisfy the real-time requirements. In this paper a new version of Matlab toolbox for simulating homogeneous systems built with Inmos transputers or digital signal processors is presented. This toolbox extended the capabilities of a previous approach. Its development aims to help the designer to compare, effortlessly, the performance of alternative parallel solutions, and also to monitor the program execution, within each processing node. This simulator is under further development to extend its applicability to parallel heterogeneous systems.
A randomized algorithm is one that uses random numbers or bits during the runtime of the algorithm. Such algorithms, when properly designed, can ensure a correct solution on every input with high probability. For many...
详细信息
A randomized algorithm is one that uses random numbers or bits during the runtime of the algorithm. Such algorithms, when properly designed, can ensure a correct solution on every input with high probability. For many problems, randomized algorithms have been designed that are simpler or more efficient than the best deterministic algorithms known for the problems. In this article, we define a natural randomized parallel complexity class, RNC, and give a survey of randomized algorithms for problems in this class.
This paper presents an efficient parallel-distributed methodology for solving multi-physic problems. This methodology is based on functional and geometric decompositions. Solution algorithms for coupled problems are d...
详细信息
This paper presents an efficient parallel-distributed methodology for solving multi-physic problems. This methodology is based on functional and geometric decompositions. Solution algorithms for coupled problems are discussed. All these techniques are illustrated for the case of CFD-based aeroelasticity. A numerical study is performed for the Agrad 445.6 aeroelastic test case. (C) 2005 Elsevier Ltd. All rights reserved.
If learning methods are to scale to the massive sizes of modern data sets, it is essential for the field of machine learning to embrace parallel and distributed computing. Inspired by the recent development of matrix ...
详细信息
If learning methods are to scale to the massive sizes of modern data sets, it is essential for the field of machine learning to embrace parallel and distributed computing. Inspired by the recent development of matrix factorization methods with rich theory but poor computational complexity and by the relative ease of mapping matrices onto distributed architectures, we introduce a scalable divide-and-conquer framework for noisy matrix factorization. We present a thorough theoretical analysis of this framework in which we characterize the statistical errors introduced by the "divide" step and control their magnitude in the "conquer" step, so that the overall algorithm enjoys high-probability estimation guarantees comparable to those of its base algorithm. We also present experiments in collaborative filtering and video background modeling that demonstrate the near-linear to superlinear speed-ups attainable with this approach.
The scale of modern datasets necessitates the development of efficient distributed optimization methods for machine learning. We present a general-purpose framework for distributed computing environments, CoCoA, that ...
详细信息
The scale of modern datasets necessitates the development of efficient distributed optimization methods for machine learning. We present a general-purpose framework for distributed computing environments, CoCoA, that has an efficient communication scheme and is applicable to a wide variety of problems in machine learning and signal processing. We extend the framework to cover general non-strongly-convex regularizers, including L1-regularized problems like lasso, sparse logistic regression, and elastic net regularization, and show how earlier work can be derived as a special case. We provide convergence guarantees for the class of convex regularized loss minimization objectives, leveraging a novel approach in handling non-strongly-convex regularizers and non-smooth loss functions. The resulting framework has markedly improved performance over state-of-the-art methods, as we illustrate with an extensive set of experiments on real distributed datasets.
暂无评论