延迟深度回声状态网络及其在时间序列预测中的应用

为提高回声状态网络对于时间序列预测问题的处理能力, 本文提出了一种延迟深度回声状态网络构造方法.该方法将多个子神经元池顺序连接, 每两个相邻的子神经元池之间嵌入了一个滞后环节.由于滞后环节的存在，该网络可将长时记忆任务转化为一系列短时记忆任务, 从而简化长时依赖问题的求解, 同时降低神经元池的构建难度.实验表明, 该网络具有强大的短时记忆容量, 对初始参数有较好的鲁棒性, 对时间序列预测问题的处理能力也比常规回声状态网络有显著提高.     

基于层次化可塑性回声状态网络的混沌时间序列预测

为了提高回声状态网络对于混沌时间序列特征提取与预测的能力,提出一种层次化可塑性回声状态网络模型.该模型将多个储备池顺序连接,通过逐层特征变换的方式增强对非线性多尺度动态特征的提取能力.同时,引入神经科学中的内在可塑性机制模拟真实生物神经元的放电率分布,以最大化神经元的信息传递为目标对储备池进行预训练.层次化可塑性回声状态网络不仅能够增加模型的容量,降低随机投影所带来的不稳定性,而且也为理解储备池的表示、处理、记忆及储存操作提供一种新的思路.仿真实验结果表明,相比于其他7种改进的回声状态网络模型,所提出的模型在人造数据和真实数据所构成的混沌时间序列预测任务中均能取得最优的预测精度.     

基于马氏距离的任务图分割方法

对实时性要求较高的嵌入式电子系统,常利用多核处理器来提高其计算速度,故对复杂程度较大的嵌入式电子系统,需分解其功能任务,使其能够映射到不同大小的处理器上,从而可靠地完成复杂系统功能。结合图论相关理论基础,以马氏距离为度量标准,提出了一种任务图分割方法。以4?4二值乘法器为例,对其进行任务图分割,并与重分割方法进行比较。结果分析表明分割后任务子图的节点数目减少,且任务子图间的通信时间较短,验证了分割方法的有效性,有利于实现复杂嵌入式电子系统的应用。     

多元时间序列的子空间回声状态网络预测模型

针对采用回声状态网络预测多元混沌时间序列时储备池学习算法可能存在的病态解问题,该文提出了一种基于快速子空间分解方法的回声状态网络预测模型.所提模型利用Krylov子空间分解方法提取储备池状态矩阵的子空间,子空间代替原状态矩阵进行输出权值求解,可以消除储备池状态矩阵的冗余信息,有效地解决伪逆算法存在的病态解问题,并且降低计算复杂度,提高泛化性能和预测精度.基于两组多元混沌时间序列的仿真结果验证了该文所提模型的有效性和实用性.     

线程池技术在B/S 网络管理软件架构中的应用

针对网络管理软件后台存在应用服务器的数据处理量多和资源消耗过大的问题,提出了改进算法,研究了线程池技术,包括线程池的工作原理、线程池使用方式、线程池配置方法、线程池监控方法和线程池的关闭方法。线程池根据基本线程池、工作队列和整个线程池的饱和情况进行工作,依据任务性质、任务优先级、任务执行时间和任务依赖性进行线程配置,以达到高效执行和最优资源的利用。     

一种解决道路空间中连续k最近邻居查询的方法

提出了一种道路空间中计算连续k最近邻居查询(CkNN)的方法,它采用分治思想,将待查询路径分为不含目标点的子路径,利用子路径端点的kNN集与分割点的关系,计算出该子路径上的目标分割点和内部分割点的位置,最后合并各子路径的分割点集得到待查询路径的连续k最近邻居.并对算法进行了时间复杂度分析.     

基于深度学习的遥感影像语义分割研究

高分辨率遥感图像的语义分割是遥感应用领域中的重要任务之一。针对经典语义分割网络在高分辨率遥感图像语义分割中存在边缘目标分割不准确、多尺度目标分割困难等问题，提出了一种基于改进空洞空间金字塔池的编码器-解码器结构网络(SMANet)。编码部分使用带有注意力机制的残差网络，使得网络充分提取图像的特征信息，其次通过多并行空洞空间金字塔模块（MASPP）获得特征图有关类别和空间上下文的更详细.信息；解码部分以自底向上方式将深层次语义信息逐步融入到低层次高分辨率图像中。使用WHDLD公开数据集对该算法进行实验，获得了6418%的平均交并比，实验结果表明SMANet优于目前主流的语义分割网络。     

分类回归树多吸引子细胞自动机分类方法及过拟合研究

基于多吸引子细胞自动机的分类方法多是二分类算法,难以克服过度拟合问题,在生成多吸引子细胞自动机时如何有效地处理多分类及过度拟合问题还缺乏可行的方法.从细胞空间角度对模式空间进行分割是一种均匀分割,难以适应空间非均匀分割的需要.将CART算法同多吸引子细胞自动机相结合构造树型结构的分类器,以解决空间的非均匀分割及过度拟合问题,并基于粒子群优化方法提出树节点的最优多吸引子细胞自动机特征矩阵的构造方法.基于该方法构造的多吸引子细胞自动机分类器能够以较少的伪穷举域比特数获得好的分类性能,减少了分类器中的空盆数量,在保证分类正确率的同时改善了过拟合问题,缩短了分类时间.实验分析证明了所提出方法的可行性和有效性.     

显著性引导及不确定性监督的深度编解码网络

基于U-Net的编码-解码网络及其变体网络在医学图像语义分割任务中取得了卓越的分割性能.然而,网络在特征提取过程中丢失了部分空间细节信息,影响了分割精度.另一方面,在多模态的医学图像语义分割任务中,这些模型的泛化能力和鲁棒性不理想.针对以上问题,本文提出一种显著性引导及不确定性监督的深度卷积编解码网络,以解决多模态医学图像语义分割问题.该算法将初始生成的显著图和不确定概率图作为监督信息来优化语义分割网络的参数.首先,通过显著性检测网络生成显著图,初步定位图像中的目标区域;然后,根据显著图计算不确定分类的像素点集合,生成不确定概率图;最后,将显著图和不确定概率图与原图像一同送入多尺度特征融合网络,引导网络关注目标区域特征的学习,同时增强网络对不确定分类区域和复杂边界的表征能力,以提升网络的分割性能.实验结果表明,本文算法能够捕获更多的语义信息,在多模态医学图像语义分割任务中优于其他的语义分割算法,并具有较好的泛化能力和鲁棒性.     

神经元混沌放电局部可视网络模型分析

为研究神经元的放电时间序列随时间的演化特性,提出了一种将放电时间序列的时间域映射到网络域进行处理的方法,即研究基于神经元的复杂网络随时间的演化特征来刻画神经元放电时间序列的时变特性.通过构建滑动时间窗内复杂网络拓扑,并计算其局部可视图的统计特性来实现时间序列时变特征的描述.对神经元map模型三种簇放电时间序列进行复杂网络构建并实现网络拓扑可视化,同时分析网络的统计特性来验证方法的有效性.结果表明,网络的拓扑、平均路径长度和聚类系数均能反映原时间序列的时变形态特征,并对神经元簇放电具有参数敏感性;簇放电稀疏程度与社团大小存在相关性.神经元放电时间序列网络域的时变演化特征能刻画其时间域特性,为神经电信号的处理提供了新的思路.     

Random Pattern and Frequency Generation Using a Photonic Reservoir Computer with Output Feedback

Reservoir computing is a bio-inspired computing paradigm for processing time dependent signals. The performance of its analogue implementations matches other digital algorithms on a series of benchmark tasks. Their potential can be further increased by feeding the output signal back into the reservoir, which would allow to apply the algorithm to time series generation. This requires, in principle, implementing a sufficiently fast readout layer for real-time output computation. Here we achieve this with a digital output layer driven by a FPGA chip. We demonstrate the first opto-electronic reservoir computer with output feedback and test it on two examples of time series generation tasks: frequency and random pattern generation. We obtain very good results on the first task, similar to idealised numerical simulations. The performance on the second one, however, suffers from the experimental noise. We illustrate this point with a detailed investigation of the consequences of noise on the performance of a physical reservoir computer with output feedback. Our work thus opens new possible applications for analogue reservoir computing and brings new insights on the impact of noise on the output feedback.     

Optimized OpenCL implementation of the Elastodynamic Finite Integration Technique for viscoelastic media

Development of parallel codes that are both scalable and portable for different processor architectures is a challenging task. To overcome this limitation we investigate the acceleration of the Elastodynamic Finite Integration Technique (EFIT) to model 2-D wave propagation in viscoelastic media by using modern parallel computing devices (PCDs), such as multi-core CPUs (central processing units) and GPUs (graphics processing units). For that purpose we choose the industry open standard Open Computing Language (OpenCL) and an open-source toolkit called PyOpenCL. The implementation is platform independent and can be used on AMD or NVIDIA GPUs as well as classical multi-core CPUs. The code is based on the Kelvin–Voigt mechanical model which has the gain of not requiring additional field variables. OpenCL performance can be in principle, improved once one can eliminate global memory access latency by using local memory. Our main contribution is the implementation of local memory and an analysis of performance of the local versus the global memory using eight different computing devices (including Kepler, one of the fastest and most efficient high performance computing technology) with various operating systems. The full implementation of the code is included.     

基于动态能量感知的云计算任务调度模型

针对传统云计算任务调度模型出现的计算量大、能耗高、效率低、调配精度差等问题,基于动态能量感知设计了一种新的云计算任务调度模型;以动态能量感知为基础,选取资源分配服务器的中央处理器的使用率、存储器的占用率、控制器的负载率等3个参数,构建三维云计算任务节点投影空间,将上述参数向量投影到空间中;引入动态能量感知建立云计算任务调度模型,采用虚拟技术将多个服务器合并成一台服务器,对调度任务进行需求分析和分类,采用能量感知算法将待调度任务分配给满足调度需求的虚拟资源,将任务调度到服务器资源上,实现任务调度;实验结果表明,基于动态能量感知的云计算任务调度模型在从小任务集和大任务集两个角度都能给有效缩短调度时间,降低调度能耗。     

Simple deterministically constructed cycle reservoirs with regular jumps

A new class of state-space models, reservoir models, with a fixed state transition structure (the "reservoir") and an adaptable readout from the state space, has recently emerged as a way for time series processing and modeling. Echo state network (ESN) is one of the simplest, yet powerful, reservoir models. ESN models are generally constructed in a randomized manner. In our previous study (Rodan & Tiňo, 2011), we showed that a very simple, cyclic, deterministically generated reservoir can yield performance competitive with standard ESN. In this contribution, we extend our previous study in three aspects. First, we introduce a novel simple deterministic reservoir model, cycle reservoir with jumps (CRJ), with highly constrained weight values, that has superior performance to standard ESN on a variety of temporal tasks of different origin and characteristics. Second, we elaborate on the possible link between reservoir characterizations, such as eigenvalue distribution of the reservoir matrix or pseudo-Lyapunov exponent of the input-driven reservoir dynamics, and the model performance. It has been suggested that a uniform coverage of the unit disk by such eigenvalues can lead to superior model performance. We show that despite highly constrained eigenvalue distribution, CRJ consistently outperforms ESN (which has much more uniform eigenvalue coverage of the unit disk). Also, unlike in the case of ESN, pseudo-Lyapunov exponents of the selected optimal CRJ models are consistently negative. Third, we present a new framework for determining the short-term memory capacity of linear reservoir models to a high degree of precision. Using the framework, we study the effect of shortcut connections in the CRJ reservoir topology on its memory capacity.     

Time Series Classification in Reservoir- and Model-Space

We evaluate two approaches for time series classification based on reservoir computing. In the first, classical approach, time series are represented by reservoir activations. In the second approach, on top of the reservoir activations, a predictive model in the form of a readout for one-step-ahead-prediction is trained for each time series. This learning step lifts the reservoir features to a more sophisticated model space. Classification is then based on the predictive model parameters describing each time series. We provide an in-depth analysis on time series classification in reservoir- and model-space. The approaches are evaluated on 43 univariate and 18 multivariate time series. The results show that representing multivariate time series in the model space leads to lower classification errors compared to using the reservoir activations directly as features. The classification accuracy on the univariate datasets can be improved by combining reservoir- and model-space.     

Analysis of macro-dataflow dynamic scheduling on nonuniform memoryaccess architectures

The author studies dynamic scheduling of computational tasks with communication costs using nonuniform memory access architecture. The computing model assumes that data transfer can be partitioned into parallel and sequential parts with respect to the task execution. A scheduling heuristic, called least-communication (LC), together with a two-level scheduler is proposed in an attempt to minimize the finish time. The LC selects the task that removes the largest amount of remaining data transfer, if no such tasks are available the task that has been ready to run at the earliest is selected first. The time complexity of LC is O(n₂). Testing the finish time of LC and first-come first-served scheduling (FCFS) shows that LC is useful for tasks having moderate granularity and whose computation and communication requirements vary widely for different data sets     

An optimal scheduling procedure for matrix inversion on linear array at a processor level

This paper presents a parallel algorithm for computing the inversion of a dense matrix based on modified Jordan's elimination which requires fewer calculation steps than the standard one. The algorithm is proposed for the implementation on the linear array with a small to moderate number of processors which operate in a parallel-pipeline fashion. A communication between neighboring processors is achieved by a common memory module implemented as a FIFO memory module. For the proposed algorithm we define a task scheduling procedure and prove that it is time optimal. In order to compute the speedup and efficiency of the system, two definitions (Amdahl's and Gustafson's) were used. For the proposed architecture, involving two to 16 processors, estimated Gustafson's (Amdahl's) speedups are in the range 1.99 to 13.76 (1.99 to 9.69).     

QingLong:一种基于常变量异步拷贝的神经网络编程模型

近年来,人工神经网络的研究取得了巨大成就,在图像识别、自然语言处理等领域均有突破性的成果,同时产生了众多商业应用,方便了我们的生活,比如语音助手、辅助驾驶等.由于神经网络算法属于计算密集型和访存密集型的负载,传统CPU处理器已不能满足其大规模商业化应用的需求,因此学术界和产业界试图在GPU、FPGA和ASIC上寻求突破.其中,神经网络加速器作为一种ASIC,它提供了高性能、低功耗的硬件解决方案,相关研究也越来越多.神经网络加速器作为一种协处理器,在其计算前后需要将数据在主机与设备之间进行搬运.特别是对吞吐量要求较高的神经网络前向推理任务,需要将网络模型参数、硬件指令等常量数据和输入、输出等变量数据,分别从主机内存拷入设备内存.如果常量数据在每一份输入数据计算前都拷贝一次,就存在常量数据重复拷贝的问题,浪费了时间与存储资源.如何在神经网络开发工具软件中实现拷贝多次变量数据但只拷贝一次常量数据,如何保证指令在每次计算中都正确寻址常量和变量,如何简化用户编程,提供用户友好的接口,就成为一系列值得研究的问题.在本文中,我们提出了一种基于常变量异步拷贝的神经网络开发工具软件及其编程模型QingLong来解决上述问题.QingLong编程模型包含三个阶段:定义网络、编译网络和计算.在定义网络阶段,用户可以为神经网络的数据节点绑定常量数据;在编译网络阶段,通过REOFF数据包装法将常量数据封装为数据包;在计算网络阶段,用户拷贝一次数据包后即可多次拷入输入数据并计算输出结果.该编程模型具有编译、计算分离,常变量异步拷贝,计算和数据拷贝可切分为三级流水线等优势.实验表明,在连续计算100份输入样本时,QingLong比DLPlib有平均17.48倍的性能提升,且输入样本越多,性能提升的倍数越大.     

Implementation of a constant‐time dynamic storage allocator

This paper describes the design criteria and implementation details of a dynamic storage allocator for real‐time systems. The main requirements that have to be considered when designing a new allocator are concerned with temporal and spatial constraints. The proposed algorithm, called TLSF (two‐level segregated fit), has an asymptotic constant cost, O(1), maintaining a fast response time (less than 200 processor instructions on a x86 processor) and a low level of memory usage (low fragmentation). TLSF uses two levels of segregated lists to arrange free memory blocks and an incomplete search policy. This policy is implemented with word‐size bitmaps and logical processor instructions. Therefore, TLSF can be categorized as a good‐fit allocator. The incomplete search policy is shown also to be a good policy in terms of fragmentation. The fragmentation caused by TLSF is slightly smaller (better) than that caused by best fit (which is one of the best allocators regarding memory fragmentation). In order to evaluate the proposed allocator, three analyses are presented in this paper. The first one is based on worst‐case scenarios. The second one provides a detailed consideration of the execution cost of the internal operations of the allocator and its fragmentation. The third analysis is a comparison with other well‐known allocators from the temporal (number of cycles and processor instructions) and spatial (fragmentation) points of view. In order to compare them, a task model has been presented. Copyright © 2007 John Wiley & Sons, Ltd.