Auto-tuned Krylov methods on cluster of graphics processing unit

Exascale computers are expected to have highly hierarchical architectures with nodes composed by multiple core processors (CPU; central processing unit) and accelerators (GPU; graphics processing unit). The different programming levels generate new difficult algorithm issues. In particular when solving extremely large linear systems, new programming paradigms of Krylov methods should be defined and evaluated with respect to modern state of the art of scientific methods. Iterative Krylov methods involve linear algebra operations such as dot product, norm, addition of vectors and sparse matrix–vector multiplication. These operations are computationally expensive for large size matrices. In this paper, we aim to focus on the best way to perform effectively these operations, in double precision, on GPU in order to make iterative Krylov methods more robust and therefore reduce the computing time. The performance of our algorithms is evaluated on several matrices arising from engineering problems. Numerical experiments illustrate the robustness and accuracy of our implementation compared to the existing libraries. We deal with different preconditioned Krylov methods: Conjugate Gradient for symmetric positive-definite matrices, and Generalized Conjugate Residual, Bi-Conjugate Gradient Conjugate Residual, transpose-free Quasi Minimal Residual, Stabilized BiConjugate Gradient and Stabilized BiConjugate Gradient (L) for the solution of sparse linear systems with non symmetric matrices. We consider and compare several sparse compressed formats, and propose a way to implement effectively Krylov methods on GPU and on multicore CPU. Finally, we give strategies to faster algorithms by auto-tuning the threading design, upon the problem characteristics and the hardware changes. As a conclusion, we propose and analyse hybrid sub-structuring methods that should pave the way to exascale hybrid methods.     

一种改进的适合并行计算的共轭剩余算法

通过改变CR算法的计算次序。提出了一种改进的共轭剩余（ICR）算法．对比CR算法。ICR算法的数值稳定性和CR算法相同,几乎没有增加计算量。但考虑了在MIMD并行机上实现时并行算法的性能,其同步开销减少为CR算法的一半,并且所有内积计算以及矩阵向量乘是独立的,没有数据相关性。可以进行计算与通信的重叠．从理论和实验两个角度来讨论ICR算法的性能,当处理机台数较多时ICR算法的计算速度快于CR算法．在64台处理机机群上进行的数值实验表明,并行ICR算法的计算速度大约比CR算法快30％．     

Intrinsic Symmetry Detection on 3D Models with Skeleton‐guided Combination of Extrinsic Symmetries

The existing methods for intrinsic symmetry detection on 3D models always need complex measures such as geodesic distances for describing intrinsic geometry and statistical computation for finding non‐rigid transformations to associate symmetrical shapes. They are expensive, may miss symmetries, and cannot guarantee their obtained symmetrical parts in high quality. We observe that only extrinsic symmetries exist between convex shapes, and two intrinsically symmetric shapes can be determined if their belonged convex sub‐shapes are symmetrical to each other correspondingly and connected in a similar topological structure. Thus, we propose to decompose the model into convex parts, and use the similar structures of the skeleton of the model to guide combination of extrinsic symmetries between convex parts for intrinsic symmetry detection. In this way, we give up statistical computation for intrinsic symmetry detection, and avoid complex measures for describing intrinsic geometry. With the similar structures being from small to large gradually, we can quickly detect multi‐scale partial intrinsic symmetries in a bottom up manner. Benefited from the well segmented convex parts, our obtained symmetrical parts are in high quality. Experimental results show that our method can find many more symmetries and runs much faster than the existing methods, even by several orders of magnitude.     

Excluding Symmetries in Constraint-Based Search

We introduce a new method, called symmetry excluding search (SES), for excluding symmetries in constraint based search. To our knowledge, it is the first declarative method that can be applied to arbitrary symmetries. The SES-method is based on the notion of symmetric constraints, which are used in our modification of a general constraint based search algorithm. The method does not influence the search strategy. Furthermore, it can be used with either the full set of symmetries, or a subset of all symmetries.We proof correctness, completeness and symmetry exclusion properties of our method. Then, we show how to apply the SES-method in the special case of geometric symmetries (rotations and reflections) and permutation symmetries. Furthermore, we give results from practical applications.     

How to calculate symmetries of Petri nets

Symmetric net structure yields symmetric net behaviour. Thus, knowing the symmetries of a net, redundant calculations can be skipped. We present a framework for the calculation of symmetries for several net classes including place/transition nets, timed nets, stochastic nets, self–modifying nets, nets with inhibitor arcs, and many others. Our approach allows the specification of different symmetry groups. Additionally it provides facilities either to calculate symmetries on demand while running the actual analysis algorithm, or to calculate them in advance. For the latter case we define and calculate a ground set of symmetries. Such a set has polynomial size and is sufficient for an efficient implementation of the for all symmetries loop and the partition of net elements into equivalence classes. These two constructions are the usual way to integrate symmetries into an analysis algorithm. Received 7 July 1997 / 10 August 1999     

Möbius Transformations For Global Intrinsic Symmetry Analysis

The goal of our work is to develop an algorithm for automatic and robust detection of global intrinsic symmetries in 3D surface meshes. Our approach is based on two core observations. First, symmetry invariant point sets can be detected robustly using critical points of the Average Geodesic Distance (AGD) function. Second, intrinsic symmetries are self‐isometries of surfaces and as such are contained in the low dimensional group of Möbius transformations. Based on these observations, we propose an algorithm that: 1) generates a set of symmetric points by detecting critical points of the AGD function, 2) enumerates small subsets of those feature points to generate candidate Möbius transformations, and 3) selects among those candidate Möbius transformations the one(s) that best map the surface onto itself. The main advantages of this algorithm stem from the stability of the AGD in predicting potential symmetric point features and the low dimensionality of the Möbius group for enumerating potential self‐mappings. During experiments with a benchmark set of meshes augmented with human‐specified symmetric correspondences, we find that the algorithm is able to find intrinsic symmetries for a wide variety of object types with moderate deviations from perfect symmetry.     

Kronecker product approximations for image restoration with whole-sample symmetric boundary conditions

Reflexive boundary conditions (BCs) assume that the array values outside the viewable region are given by a symmetry of the array values inside. The reflection guarantees the continuity of the image. In fact, there are usually two choices for the symmetry: symmetry around the meshpoint and symmetry around the midpoint. The first is called whole-sample symmetry in signal and image processing, the second is half-sample. Many researchers have developed some fast algorithms for the problems of image restoration with the half-sample symmetric BCs over the years. However, little attention has been given to the whole-sample symmetric BCs. In this paper, we consider the use of the whole-sample symmetric boundary conditions in image restoration. The blurring matrices constructed from the point spread functions (PSFs) for the BCs have block Toeplitz-plus-PseudoHankel with Toeplitz-plus-PseudoHankel blocks structures. Recently, regardless of symmetric properties of the PSFs, a technique of Kronecker product approximations was successfully applied to restore images with the zero BCs, half-sample symmetric BCs and anti-reflexive BCs, respectively. All these results extend quite naturally to the whole-sample symmetric BCs, since the resulting matrices have similar structures. It is interesting to note that when the size of the true PSF is small, the computational complexity of the algorithm obtained for the Kronecker product approximation of the resulting matrix in this paper is very small. It is clear that in this case all calculations in the algorithm are implemented only at the upper left corner submatrices of the big matrices. Finally, detailed experimental results reporting the performance of the proposed algorithm are presented.     

基于共轭梯度的极速学习机

极速学习机(ELM)由于具有较快的训练速度和较好的泛化能力而被广泛的应用到很多的领域,然而在计算数据样例个数较大的情况下,它的训练速度就会下降,甚至会出现程序报错,因此提出在ELM模型中用改进的共轭梯度算法代替广义逆的计算方法。实验结果表明,与求逆矩阵的ELM算法相比,在同等泛化精度的条件下,共轭梯度ELM有着更快的训练速度。通过研究发现:基于共轭梯度的极速学习机算法不需要计算一个大型矩阵的广义逆,而大部分广义逆的计算依赖于矩阵的奇异值分解(SVD),但这种奇异值分解对于阶数很高的矩阵具有很低的效率;因为已经证明共轭梯度算法可通过有限步迭代找到其解,所以基于共轭剃度的极速学习机有着较高的训练速度,而且也比较适用于处理大数据。     

Analyzing skewed symmetries

Symmetry is pervasive in both man-made objects and nature. Since symmetries project to skew symmetries, finding axes of skew symmetry is an important vision task. This paper presents a linear time algorithm for finding the axes of skew symmetry, where the degree of symmetry is known. First, we present a review and critique of current methods for finding the axes of skew symmetry. Next, we decompose the problem of finding skew symmetry into the subproblems of solving for the rotational parameter of a shear symmetry and recovering the shear parameter of a reflexive symmetry. Using this approach, the authors derive a direct, non-heuristic moment-based technique for finding the axes of skew symmetry. For skew symmetric figures with degree of symmetry less than five we obtain a closed-form solution. The method does not rely on continuous contours but assumes there is no occlusion and requires knowing the contour's degree of symmetry. It is the first algorithm to find the axes of skew symmetry inO(n) time, where n is the number of contour points. The method is especially suited to industrial applications where the degree of symmetry is often knowna priori. Examples of the method are presented for both real and synthetic images, and an error analysis of the method is given.     

Nonlinear systems possessing linear symmetry

This paper tackles linear symmetries of control systems. Precisely, the symmetry of affine nonlinear systems under the action of a sub‐group of general linear group GL(n,?). First of all, the structure of state space (briefly, ss) symmetry group and its Lie algebra for a given system is investigated. Secondly, the structure of systems, which are ss‐symmetric under rotations, is revealed. Thirdly, a complete classification of ss‐symmetric planar systems is presented. It is shown that for planar systems there are only four classes of systems which are ss‐symmetric with respect to four linear groups. Fourthly, a set of algebraic equations are presented, whose solutions provide the Lie algebra of the largest connected ss‐symmetry group. Finally, some controllability properties of systems with ss‐symmetry group are studied. As an auxiliary tool for computation, the concept and some properties of semi‐tensor product of matrices are included. Copyright © 2006 John Wiley & Sons, Ltd.     

Spring algorithms and symmetry

Spring algorithms are regarded as effective tools for visualizing undirected graphs. One major feature of applying spring algorithms is to display symmetric properties of graphs. This feature has been confirmed by numerous experiments. In this paper, firstly we formalize the concepts of graph symmetries in terms of “reflectional” and “rotational” automorphisms; and characterize the types of symmetries, which can be displayed simultaneously by a graph layout, in terms of “geometric” automorphism groups. We show that our formalization is complete. Secondly, we provide general theoretical evidence of why many spring algorithms can display graph symmetry. Finally, the strength of our general theorem is demonstrated from its application to several existing spring algorithms.     

Permutation equivalence of cubic rotation symmetric Boolean functions

Rotation symmetric Boolean functions have been extensively studied for about 15 years because of their applications in cryptography and coding theory. Until recently little was known about the basic question of when two such functions are affine equivalent. The simplest case of quadratic rotation symmetric functions which are generated by cyclic permutations of the variables in a single monomial was only settled in 2009. For the much more complicated case of cubic rotation symmetric functions generated by a single monomial, the affine equivalence classes under permutations which preserve rotation symmetry were determined in 2011. It was conjectured then that the cubic equivalence classes are the same if all nonsingular affine transformations, not just permutations, are allowed. This conjecture is probably difficult, but here we take a step towards it by proving that the cubic affine equivalence classes found in 2011 are the same if all permutations, not just those preserving rotation symmetry, are allowed. The needed new idea uses the theory of circulant matrices.     

Simulating Boolean Circuits on a DNA Computer

We demonstrate that DNA computers can simulate Boolean circuits with a small overhead. Boolean circuits embody the notion of massively parallel signal processing and are frequently encountered in many parallel algorithms. Many important problems such as sorting, integer arithmetic, and matrix multiplication are known to be computable by small size Boolean circuits much faster than by ordinary sequential digital computers. This paper shows that DNA chemistry allows one to simulate large semi-unbounded fan-in Boolean circuits with a logarithmic slowdown in computation time. Also, for the class NC ¹ , the slowdown can be reduced to a constant. In this algorithm we have encoded the inputs, the Boolean AND gates, and the OR gates to DNA oligonucleotide sequences. We operate on the gates and the inputs by standard molecular techniques of sequence-specific annealing, ligation, separation by size, amplification, sequence-specific cleavage, and detection by size. Additional steps of amplification are not necessary for NC ¹ circuits. The feasibility of the DNA algorithm has been successfully tested on a small circuit by actual biochemical experiments. Received May 29, 1997; revised February 15, 1998.     

Symmetry Definitions for Constraint Satisfaction Problems

We review the many different definitions of symmetry for constraint satisfaction problems (CSPs) that have appeared in the literature, and show that a symmetry can be defined in two fundamentally different ways: as an operation preserving the solutions of a CSP instance, or else as an operation preserving the constraints. We refer to these as solution symmetries and constraint symmetries. We define a constraint symmetry more precisely as an automorphism of a hypergraph associated with a CSP instance, the microstructure complement. We show that the solution symmetries of a CSP instance can also be obtained as the automorphisms of a related hypergraph, the k-ary nogood hypergraph and give examples to show that some instances have many more solution symmetries than constraint symmetries. Finally, we discuss the practical implications of these different notions of symmetry.     

Static and dynamic structural symmetry breaking

We reconsider the idea of structural symmetry breaking for constraint satisfaction problems (CSPs). We show that the dynamic dominance checks used in symmetry breaking by dominance-detection search for CSPs with piecewise variable and value symmetries have a static counterpart: there exists a set of constraints that can be posted at the root node and that breaks all the compositions of these (unconditional) symmetries. The amount of these symmetry-breaking constraints is linear in the size of the problem, and yet they are able to remove a super-exponential number of symmetries on both values and variables. Moreover, we compare the search trees under static and dynamic structural symmetry breaking when using fixed variable and value orderings. These results are then generalised to wreath-symmetric CSPs with both variable and value symmetries. We show that there also exists a polynomial-time dominance-detection algorithm for this class of CSPs, as well as a linear-sized set of constraints that breaks these symmetries statically.     

Symmetry Breaking Constraints for Value Symmetries in Constraint Satisfaction

Constraint satisfaction problems (CSPs) sometimes contain both variable symmetries and value symmetries, causing adverse effects on CSP solvers based on tree search. As a remedy, symmetry breaking constraints are commonly used. While variable symmetry breaking constraints can be expressed easily and propagated efficiently using lexicographic ordering, value symmetry breaking constraints are often difficult to formulate. In this paper, we propose two methods of using symmetry breaking constraints to tackle value symmetries. First, we show theoretically when value symmetries in one CSP correspond to variable symmetries in another CSP of the same problem. We also show when variable symmetry breaking constraints in the two CSPs, combined using channeling constraints, are consistent. Such results allow us to tackle value symmetries efficiently using additional CSP variables and channeling constraints. Second, we introduce value precedence, a notion which can be used to break a common class of value symmetries, namely symmetries of indistinguishable values. While value precedence can be expressed using inefficient if-then constraints in existing CSP solvers, we propose efficient propagation algorithms for implementing global value precedence constraints. We also characterize several theoretical properties of the value precedence constraints. Extensive experiments are conducted to verify the feasibility and efficiency of the two proposals.     

基于频域共轭梯度算法的盲目图像恢复

在分析了目前各种算法存在的一些基本问题的基础上，提出了盲目图像恢复的频域共轭梯度算法，算法假设退化系统的点扩散函数具有圆对称性，并考虑图像的频域具有复共轭对称性，在频域里建立了新的优化判据；采用共轭梯度算法搜索判据，从而在有实际物理意义的区域中收敛于最优解；对光学成像系统常见的离焦、衍射模糊问题进行了仿真；用该算法实现了其退化图像的恢复，给出并分析了结果。     

Multipartite quantum and classical correlations in symmetric <Emphasis Type="Italic">n</Emphasis>-qubit mixed states

We discuss how to calculate genuine multipartite quantum and classical correlations in symmetric, spatially invariant, mixed n-qubit density matrices. We show that the existence of symmetries greatly reduces the amount of free parameters to be optimized in order to find the optimal measurement that minimizes the conditional entropy in the discord calculation. We apply this approach to the states exhibited dynamically during a thermodynamic protocol to extract maximum work. We also apply the symmetry criterion to a wide class of physically relevant cases of spatially homogeneous noise over multipartite entangled states. Exploiting symmetries we are able to calculate the non-local and genuine quantum features of these states and note some interesting properties.     

Skeleton-based intrinsic symmetry detection on point clouds

We present a skeleton-based algorithm for intrinsic symmetry detection on imperfect 3D point cloud data. The data imperfections such as noise and incompleteness make it difficult to reliably compute geodesic distances, which play essential roles in existing intrinsic symmetry detection algorithms. In this paper, we leverage recent advances in curve skeleton extraction from point clouds for symmetry detection. Our method exploits the properties of curve skeletons, such as homotopy to the input shape, approximate isometry-invariance, and skeleton-to-surface mapping, for the detection task. Starting from a curve skeleton extracted from an input point cloud, we first compute symmetry electors, each of which is composed of a set of skeleton node pairs pruned with a cascade of symmetry filters. The electors are used to vote for symmetric node pairs indicating the symmetry map on the skeleton. A symmetry correspondence matrix (SCM) is constructed for the input point cloud through transferring the symmetry map from skeleton to point cloud. The final symmetry regions on the point cloud are detected via spectral analysis over the SCM. Experiments on raw point clouds, captured by a 3D scanner or the Microsoft Kinect, demonstrate the robustness of our algorithm. We also apply our method to repair incomplete scans based on the detected intrinsic symmetries.     

Some properties and construction of multiwavelets related to different symmetric centers

In this paper we are interested in discuss the symmetry property and construction of an m-band compactly supported orthonormal multiwavelets related to the filters with different symmetric centers. With the development of the several equivalent conditions on this type of symmetry in terms of filter sequences and polyphase matrices, we derive several necessary constraints on the number of symmetric filters of the system, which is crucial for the construction of multiwavelets associated with given multiscaling functions with different symmetry centers. Then, we show how to construct multiwavelets with desired symmetric property by matrix extensions. Finally, to illustrate our proposed general scheme, we give two examples in this paper.