双分支迭代的深度增量图像分类方法

针对深度增量学习可能产生灾难性遗忘的问题,提出双分支迭代的深度增量图像分类方法,使用主网络存储旧类知识,分支网络学习增量数据中的新类知识,并在增量过程中使用主网络的权重优化分支网络的参数.使用基于密度峰值聚类的方法从迭代数据集中筛选典型样本并构建保留集,并加入增量迭代训练中,减轻灾难性遗忘.实验表明,文中方法的性能较优.     

一种增量式数据流分析方法

数据流分析是编译器中重要部分，而增量式分析在程序开发环境和过程间优化编译器中有着相关实用的价值，当程序发生变化时，它可以增量式地维护数据流信息，而不致因程序的任何小改动都重新进行数据流分析，给出了一种增量式的消去数据流算法，它基于路径简化算法，具有和路径简化算法同样的复杂度，同样的通用性（适用于不可归约流图和流函数不完备的情况），而且能方便地在程序发生变化时维护现有的数据流信息。     

面向裕量优化的高效时钟偏差规划和延迟提取

为了减少时钟偏差规划所需的时间,提出一种准线性时间复杂度的时钟偏差规划方法.该方法以整数来描述延迟大小的时钟偏差规划算法,限制每次对时钟延迟调整的步进至少为1,降低了算法的时间复杂度;改变了传统的预先生成完整的时序图作为算法输入的流程,采用一种新的增量式延迟提取策略为时钟偏差规划算法提取关键边的权重,减少了生成时序图所需要的时间.实验结果表明,采用文中方法进行时钟偏差规划的效率很高,对包含数千触发器的基准测试电路,其运行时间仅为数十秒.     

Prediction of performance and processor requirements in real-timedata flow architectures

Presents a new data flow graph model for describing the real-time execution of iterative control and signal processing algorithms on multiprocessor data flow architectures. Identified by the acronym ATAMM, for Algorithm to Architecture Mapping Model, the model is important because it specifies criteria for a multiprocessor operating system to achieve predictable and reliable performance. Algorithm performance is characterized by execution time and iteration period. For a given data flow graph representation, the model facilitates calculation of greatest lower bounds for these performance measures. When sufficient processors are available, the system executes algorithms with minimum execution time and minimum iteration period, and the number of processors required is calculated. When only limited processors are available or when processors fail, performance is made to degrade gracefully and predictably. The user off-line is able to specify tradeoffs between increasing execution time or increasing iteration period. The approach to achieving predictable performance is to control the injection rate of input data and to modify the data flow graph precedence relations so that a processor is always available to execute an enabled graph node. An implementation of the ATAMM model in a four-processor architecture based on Westinghouse's VHSIC 1750A Instruction Set Processor is described     

Simulating cyclic artery compression using a 3D unsteady model with fluid–structure interactions

High pulsating blood pressure and severe stenosis make fluid–structure interaction (FSI) an important role in simulating blood flow in stenotic arteries. A three-dimensional nonlinear model with FSI and a numerical method using GFD are introduced to study unsteady viscous flow in stenotic tubes with cyclic wall collapse simulating blood flow in stenotic carotid arteries. The Navier–Stokes equations are used as the governing equations for the fluid. A thin-shell model is used for the tube wall. Interaction between fluid and tube wall is treated by an incremental boundary iteration method. Elastic properties of the tube wall are determined experimentally using a polyvinyl alcohol hydrogel artery stenosis model. Cyclic tube compression and collapse, negative pressure and high shear stress at the throat of the stenosis, flow recirculation and low shear stress just distal to the stenoses were observed under physiological conditions. These critical flow and mechanical conditions may be related to platelet aggregation, thrombus formation, excessive artery fatigue and possible plaque cap rupture. Computational and experimental results are compared and reasonable agreement is found.     

基于对象变化的邻域决策粗糙集动态更新算法

针对现实环境下数据集不断动态变化的特性,提出一种邻域决策粗糙集模型的增量式更新算法。采用由简单到复杂的研究思路,分析了邻域型信息系统论域增加和减少单个对象时,目标近似集与邻域类之间概率的变化规律,进一步地利用这种规律来构造单个对象变化时邻域决策粗糙集模型上下近似集的增量式更新,在单个对象变化的基础上,通过逐步迭代的方式设计了对象批量变化时的增量式更新算法。实验分析表明,所提出的算法具有较高的增量式更新性能,适用于动态数据环境下邻域决策粗糙集模型的动态更新。     

A study of variable step-length incremental/iterative methods for nonlinear finite element equations

A deep study is carried out on the variable step-length incremental/iterative methods of nonlinear finite element equations in static structural analysis. Simple interpretations are presented for the conventional incremental/iterative method, single characteristic displacement controlling method and arc-length controlling method from their geometric characteristics in the load-deformation space. Based on the concept of structural equilibrium, three constraint equations (i.e. zero incremental displacement norm, zero residual force norm and zero incremental work) are established for controlling the iteration path. The automatic selection method of initial load increment in each increment is discussed. The suitability and efficiency of the aforementioned methods are studied and compared through tracing the snap through and snap back responses of two cylindrical shells.     

An Incremental Algorithm for Parallel Training of the Size and the Weights in a Feedforward Neural Network

An algorithm of incremental approximation of functions in a normed linearspace by feedforward neural networks is presented. The concept of variationof a function with respect to a set is used to estimate the approximationerror together with the weight decay method, for optimizing the size andweights of a network in each iteration step of the algorithm. Two alternatives, recursively incremental and generally incremental, are proposed. In the generally incremental case, the algorithm optimizes parameters of all units in the hidden layer at each step. In the recursively incremental case, the algorithm optimizes the parameterscorresponding to only one unit in the hidden layer at each step. In thiscase, an optimization problem with a smaller number of parameters is beingsolved at each step.     

Optimal Pose Estimation in Two and Three Dimensions

This paper concerns the problem of pose estimation, in particular the application of linear and nonlinear methods to the case of anisotropic error distributions. Incremental and batch techniques are discussed, and a linear solution to the two dimensional problem is developed by reference to the noise model. In this case, manipulation of the normal equations is found to offer a flexible and transparent framework for both techniques. This framework is used to support batch iterative solutions to the 3D imaging problem, and methods are developed that improve the range of stability, the number of iterations, and the computation per iteration each by a factor of about two. Experiments with the batch methods on synthetic data in three dimensions are reported and compared with incremental techniques based on the Kalman filter. The incremental techniques are found to produce significantly suboptimal results and limited stability. For practical applications the computational efficiency of the new batch techniques is found to be better than incremental ones. The techniques are extended to 2D images of 3D data, in the form of single and stereo projections. Experiments with synthetic and real image data show that a suitable choice of anisotropic weighting offers greater accuracy and robustness than perpendicular error measures and examine stereo estimates with and without reconstruction.     

基于RUP的MIS需求分析过程模型

本文归纳了RUP的主要核心思想为：用例驱动,以构架为中心的迭代和增量开发。以RUP的主要核心思想为基础,提出了MIS需求分析的过程模型,并通过一个具体的MIS应用实例验证了这个模型的实用性。     

粗决策树动态规则提取算法研究及应用

针对静态算法对大数据和增量数据处理不足的问题,构造了基于粗决策树的动态规则提取算法,并将其应用于旋转机械故障诊断中.将粗集与决策树结合,用增量方式实现样本抽取;经过动态约简、决策树构造、规则提取与选择、匹配4个步骤的循环迭代过程,实现了数据的动态规则提取,使得提取的规则具有更高的可信度;同时,将算法应用于旋转机械故障诊断这一动态问题中,验证了算法的有效性;最后,将所提算法分别与静态算法和增量式动态算法进行了效率对比分析,实验结果表明,所提算法能够以最精简的规则获得更多数据隐含信息.     

一种改进的增量数据挖掘算法

针对现行增量挖掘算法都不能解决最小支持度和最小置信度发生变化的同时数据库也发生变化的增量更新问题,分析现有算法的不足,提出一种改进算法。该算法充分利用以前挖掘的结果,将链表插入、删除以及hash定位高效率的特性结合起来,摆脱了传统算法多次迭代的不足,可实现最小支持度与最小置信度发生变化的同时,数据库也相应的发生变化的增量更新。并通过实验对该算法进行验证,同时将测试结果与Aprior算法的测试结果进行比较,证明其高效性。     

基于Fisher判别分析的增量式非负矩阵分解算法

增量式非负矩阵分解算法是基于子空间降维技术的无监督增量学习方法.文中将Fisher判别分析思想引入增量式非负矩阵分解中,提出基于Fisher判别分析的增量式非负矩阵分解算法.首先,利用初始样本训练的先验信息,通过索引矩阵对新增系数矩阵进行初始化赋值.然后,将增量式非负矩阵分解算法的目标函数改进为批量式的增量学习算法,在此基础上施加类间散度最大和类内散度最小的约束.最后,采用乘性迭代的方法计算分解后的因子矩阵.在ORL、Yale B和PIE等3个不同规模人脸数据库上的实验验证文中算法的有效性.     

增量Hessian LLE算法研究

利用基于Ritz加速的逆幂迭代算法,在经典的Hessian LLE算法基础上提出一种增量LLE算法,能够高效地处理新增的一个或多个样本。该算法的核心思想是将增量流形学习问题转化为一个增量特征值问题,利用数值线性代数的工具进行求解,并分析算法的收敛性。在合成数据集和图像数据集上,验证该增量算法的效率和精确度。     

Efficient handling of data structures in definitional languages

Implementations of operations on general data structures in definitional languages often lead to excessive copying and storage requirements. To partially overcome this problem, users are given facilities to select efficient storage structures or to guide storage allocation. This contradicts the spirit of definitional languages, requiring the user to get involved with implementation details.

This paper presents a method for automatically recognizing excessive copying and optimizing the storage for data structures. Based on analysis of data dependencies, the storage may be reduced from an entire structure to individual elements of the structure. The benefits are especially significant in incremental structures, where only a constant number of elements of a large data structure is modified in each operation. For incremental structures, copying of unchanged parts of the structure is avoided, and unnecessary iterations are eliminated, without involving the user in this consideration. The user is thus relieved of considering the inefficiencies inherent in specifications in definitional languages. The method is applicable to a variety of language processors and computer architectures. The proposed optimization method produces better results than those obtained by explicit storage references.

The paper describes the implementation of the method in the compiler of the MODEL definitional language. First, criteria are presented for recognizing structures that may be optimized. Then, a transformation that removes iterations implied by the operations on incremental structures is described. The method is exemplified by its application to the non-recursive iteration computation of the Ackermann's function.     

基于云模型的最接近支持向量机增量学习方法*

针对经典支持向量机在增量学习中的不足,提出一种基于云模型的最接近支持向量机增量学习算法。该方法利用最接近支持向量机的快速学习能力生成初始分类超平面,并与k近邻法对全部训练集进行约简,在得到的较小规模的精简集上构建云模型分类器直接进行分类判断。该算法模型简单,不需迭代求解,时间复杂度较小,有较好的抗噪性,能较好地体现新增样本的分布规律。仿真实验表明,本算法能够保持较好的分类精度和推广能力,运算速度较快。     

软件体系结构的迭代增量构建方法

软件体系结构是有效实现大粒度软件复用的重要方法，但如何具体实现软件体系结构一直没有得到有效的解决。在分析用例和软件体系结构之间关系的基础上，提出了一种用例驱动、迭代增量方式构建软件体系结构的方法及其模型，分析了迭代增量的过程及相关问题。     

Improved solution methods for inelastic rate problems

The objective of the paper is an assessment of the incremental solution methods for the analysis of inelastic rate problems. In particular, the possibilities of the initial load method are explored to improve the accuracy and stability of the traditional explicit operators by higher-order time expansions and implicit weighting schemes.The convergence limitations are examined for different classes of inelastic growth laws (viscous flow, viscoelasticity, viscoplasticity) which restrict the time step because of the iterative solution of the implicit algorithm. The range and rate of convergence of the initial load method (constant stiffness predictor-corrector iteration) is enlarged by tangential gradient techniques which account for the inelastic response in the structural stiffness matrix. In this way the time step restriction disappears although at a considerable increase of computational expense because of the costly computation and decomposition of structural gradients within each iteration cycle (Newton-Raphson methods).As compared to the linear single-step methods, the cubic Hermitian time expansions furnish far better accuracy than the traditional linear expansions for very little increase of computational cost. Stability and convergence limits correspond to those of the lower-order operators, whereby the implicit midstep of backward weighting schemes are most advantageous. In this context it is worth noting that aging or strain-hardening effects in the inelastic growth law reduce dramatically the time step restrictions of the iterative initial load solution methods (predictor-corrector schemes), as compared to the simplest creep model in which the inelastic growth law depends only on stress, e.g. for viscous flow and viscoplasticity.     

An implicit marching method for the two-dimensional reduced Navier-Stokes equations at arbitrary mach number

An implicit marching procedure is developed from a parabolized form of the two-dimensional Navier-Stokes equations with allowance for chemical reaction in an effort to predict the flow in highly viscous chemical laser nozzles. The equations governing the fluid dynamics are linearized and solved as a coupled implicit system in a manner which requires no iteration and in which the pressure gradient for internal flows emerges directly as part of the solution without iteration. Accuracy and stability of the method are tested by computing the flow in the entrance region between parallel flat plates and comparing the results to analytical results as well as experimental data. Upon completion of these tests the procedure is used to predict the highly viscous chemically reacting flow in a laser nozzle and inlet duct.