Speed-up Strategies Research Articles

Graph-Based Blind Image Deblurring From a Single Photograph.

Blind image deblurring, i.e., deblurring without knowledge of the blur kernel, is a highly ill-posed problem. The problem can be solved in two parts: i) estimate a blur kernel from the blurry image, and ii) given an estimated blur kernel, de-convolve the blurry input to restore the target image. In this paper, we propose a graph-based blind image deblurring algorithm by interpreting an image patch as a signal on a weighted graph. Specifically, we first argue that a skeleton image-a proxy that retains the strong gradients of the target but smooths out the details-can be used to accurately estimate the blur kernel and has a unique bi-modal edge weight distribution. Then, we design a reweighted graph total variation (RGTV) prior that can efficiently promote a bi-modal edge weight distribution given a blurry patch. Further, to analyze RGTV in the graph frequency domain, we introduce a new weight function to represent RGTV as a graph l1-Laplacian regularizer. This leads to a graph spectral filtering interpretation of the prior with desirable properties, including robustness to noise and blur, strong piecewise smooth (PWS) filtering and sharpness promotion. Minimizing a blind image deblurring objective with RGTV results in a non-convex non-differentiable optimization problem. Leveraging the new graph spectral interpretation for RGTV, we design an efficient algorithm that solves for the skeleton image and the blur kernel alternately. Specifically for Gaussian blur, we propose a further speedup strategy for blind Gaussian deblurring using accelerated graph spectral filtering. Finally, with the computed blur kernel, recent non-blind image deblurring algorithms can be applied to restore the target image. Experimental results demonstrate that our algorithm successfully restores latent sharp images and outperforms state-of-the-art methods quantitatively and qualitatively.

Solving Dynamic Programming Problem by Pipeline Implementation on GPU

In this paper, we show the effectiveness of a pipeline implementation of Dynamic Programming (DP) on GPU. As an example, we explain how to solve a matrix-chain multiplication (MCM) problem by DP on GPU. This problem can be sequentially solved in $O(n^3)$ steps by DP where $n$ is the number of matrices, because its solution table is of size $n \times n$ and each element of the table can be computed in $O(n)$ steps. A typical speedup strategy for this is to parallelize the $O(n)$ step computation of each element, which can be easily achieved by parallel prefix computation, i.e., an $O(\log n)$ step computation with $n$ threads in a tournament fashion. By such a standard parallelizing method, we can solve the MCM problem in $O(n^2 \log n)$ steps with $n$ threads. In our approach, we solve the MCM problem on GPU in a pipeline fashion, i.e., we use GPU cores for supporting pipeline-stages so that many elements of the solution table are partially computed in parallel at one time. Our implementation determines one output value per one computational step with $n$ threads in a pipeline fashion and constructs the solution table totally in $O(n^2)$ steps with $n$ threads.

ParSymG: a parallel clustering approach for unsupervised classification of remotely sensed imagery

ABSTRACTSymmetry is a common feature in the real world. It may be used to improve a classification by using the point symmetry-based distance as a measure of clustering. However, it is time consuming to calculate the point symmetry-based distance. Although an efficient parallel point symmetry-based K-means algorithm (ParSym) has been propsed to overcome this limitation, ParSym may get stuck in sub-optimal solutions due to the K-means technique it used. In this study, we proposed a novel parallel point symmetry-based genetic clustering (ParSymG) algorithm for unsupervised classification. The genetic algorithm was introduced to overcome the sub-optimization problem caused by inappropriate selection of initial centroids in ParSym. A message passing interface (MPI) was used to implement the distributed master–slave paradigm. To make the algorithm more time-efficient, a three-phase speedup strategy was adopted for population initialization, image partition, and kd-tree structure-based nearest neighbor searching. The advantages of ParSymG over existing ParSym and parallel K-means (PKM) alogithms were demonstrated through case studies using three different types of remotely sensed images. Results in speedup and time gain proved the excellent scalability of the ParSymG algorithm.

Some speed-up strategies for solving inverse radiative transfer problems

Inversion based on the radiative transfer equation (RTE) is generally highly CPU time consuming because the forward model itself is complicated to solve when the space dimension is greater than one, and because the inversion is based on a large number of forward model runs until convergence is reached. The goal of this paper is to set up some speed-up strategies specific of inversion when radiative transfer problems are dealt with. More specifically, the accurate identification of the volumetric radiative properties i.e. both the absorption and scattering coefficients is the objective of this study.

Directed self-assembly cut mask assignment for unidirectional design

Recently, directed self-assembly (DSA) has emerged as a promising lithography solution for cut manufacturing. We perform a comprehensive study on the DSA aware mask optimization problem to provide a DSA friendly design on cut layers. We first formulate the problem as an integer linear programming (ILP) to assign cuts to different guiding templates, targeting both conflict minimization and line-end extension minimization. As ILP may not be scalable for very large size problems, we then propose two speed-up strategies. The first one is to decompose the initial problem into smaller ones and solve them separately, followed by solution merging without much loss of quality. The second one is using the set cover algorithm to decide the DSA guiding pattern assignment, and then legalize the template placement. Our approaches can be naturally extended to handle arbitrary DSA guiding template patterns with complicated shapes. Experimental results show that our methodologies can significantly improve the DSA friendly, i.e., both the unresolved pattern number and the line-end extensions can be reduced.

Combine color and shape in real-time detection of texture-less objects

Object instance detection is a fundamental problem in computer vision and has many applications. Compared with the problem of detecting a texture-rich object, the detection of a texture-less object is more involved because it is usually based on matching the shape of the object with the shape primitives extracted from an image, which is not as discriminative as matching appearance-based local features, such as the SIFT features. The Dominant Orientation Templates (DOT) method proposed by Hinterstoisser et al. is a state-of-the-art method for the detection of texture-less objects and can work in real time. However, it may well generate false detections in a cluttered background. In this paper, we propose a new method which has three contributions. Firstly, it augments the DOT method with a type of illumination insensitive color information. Since color is complementary to shape, the proposed method significantly outperforms the original DOT method in the detection of texture-less object in cluttered scenes. Secondly, we come up with a systematic way based on logistic regression to combine the color and shape matching scores in the proposed method. Finally, we propose a speed-up strategy to work with the proposed method so that it runs even faster than the original DOT method. Extensive experimental results are presented in this paper to compare the proposed method directly with the original DOT method and the LINE-2D method, and indirectly with another two state-of-the-art methods.

An Efficient Speedup Strategy for Constant Sum Game Computations

Large classes of game theoretic problems seem to defy attempts of finding polynomial-time algorithms while analyzing large amounts of data. This premise leads naturally to the possibility of using efficient parallel computing implementations when seeking exact solutions to some of these problems. Although alpha beta algorithms for more than one-player game-tree searches show moderate parallel performance, this paper sets forth an alpha beta strategy enhanced with transposition tables in order to offer satisfactory speedups on high performance servers. When the access to the transposition tables is done in low constant delay time, the achieved speedups should approach the theoretical upper bounds of the code parallelism. We tested the strategy on a well-known combinatorial game.Keywords: Constant Sum Games, Game Computation Parallelization Strategies1IntroductionHigh-performance parallel computing is being increasingly used in computational science to solve large-scale algorithmically intensive problems. Recent technological advances in processor development gave rise to a large range of unexpansive chips with tens or even hundreds of cores. These multi-core chips enhanced with simultaneous multithreading, dedicated fast access memory and manifold specific cores, promise additional performance and efficiency gains [1] [2], especially in processing computational science applications.Large classes of game theoretic problems seem to defy attempts of finding polynomialtime algorithms [3-7], requiring super-polynomial time algorithmic schemes and, thereby, being computationally intensive. Furthermore, large quantities of data can potentially be generated, analyzed and stored while exploring exponentially growing game trees.These premises lead naturally to the possibility of applying highly parallel computational implementations when aiming to find exact solutions for many game theory problems. We propose a speedup strategy to allow constant sum games computations to take full advantage of the parallelism and the efficient scheduling strategies implemented on these architectures [8] [9], This way, we obtain fast and accurate sub-game utility estimations by scaling up the amount of required resources and accelerating the dissemination of intermediate results.This paper is organized as follows: section 2 is devoted to presenting the underlying game theoretic concepts used in this research and related algorithmic complexity aspects. The speedup strategy for constant sum games is expounded in the third section. A case study together with its experimental plan is detailed in section 4. Results and conclusions are presented in sections 5 and 6.2 Overview of relevant Game Theory ConceptsThe current section starts off by briefly restating the definitions of the main concepts used throughout the paper, and proceeds by expounding the underlying considerations on the algorithmic complexity of game equilibria computations.Brief Overview of underlying game theoretic ConceptsA finite perfect-information game in extensive form is defined as a tuple G = (JV, A, H, Z, χ, p, a, u) where:* N is the finite set of rational players;* A is the set of possible actions of the players;* H is the set of choice nodes;* χ: H -> 2a is the choice function;* p: H -> N is the player function which indicates player to perform the move at some given choice node;* σ: H x A -> H U Z is the successor function. This function maps pairs of choice nodes and possible actions to other successor terminal or choice nodes;* u = (u1; ...,U|N|), whereu?:Z -> R is a utility function for the player i on the terminal nodes Z,* Z is the set of terminal nodes in the game three. Each terminal node of the game tree has an N-tuple of payoffs, one for each player at the end of every possible play.Each player has a utility function defined for every game outcome. …

Conditional distance based matching for one-shot gesture recognition

A problem of matching gestures, where there are one or few samples per class, is considered in this paper. The proposed approach shows that much better results are achieved if the distance between the pattern of frame-wise distances of two gesture sequences with a third (anchor) sequence from the modelbase is considered. Such a measure is called as conditional distance and these distance pattern are referred to as “warp vectors”. If these warp vectors are similar, then so are the sequences; if not, they are dissimilar. At the algorithmic core, there are two dynamic time warping processes, one to compute the warp vectors with the anchor sequences and the other to compare these warp vectors. In order to reduce the complexity a speedup strategy is proposed by pre-selecting “good” anchor sequences. Conditional distance is used for individual and sentence level gesture matching. Both single and multiple subject datasets are used. Experiments show improved performance above 82% spanning 179 classes.

Optimizing the hierarchical prediction and coding in HEVC for surveillance and conference videos with background modeling.

For the real-time and low-delay video surveillance and teleconferencing applications, the newly video coding standard HEVC can achieve much higher coding efficiency over H.264/AVC. However, we still argue that the hierarchical prediction structure in the HEVC low-delay encoder still does not fully utilize the special characteristics of surveillance and conference videos that are usually captured by stationary cameras. In this case, the background picture (G-picture), which is modeled from the original input frames, can be used to further improve the HEVC low-delay coding efficiency meanwhile reducing the complexity. Therefore, we propose an optimization method for the hierarchical prediction and coding in HEVC for these videos with background modeling. First, several experimental and theoretical analyses are conducted on how to utilize the G-picture to optimize the hierarchical prediction structure and hierarchical quantization. Following these results, we propose to encode the G-picture as the long-term reference frame to improve the background prediction, and then present a G-picture-based bit-allocation algorithm to increase the coding efficiency. Meanwhile, according to the proportions of background and foreground pixels in coding units (CUs), an adaptive speed-up algorithm is developed to classify each CU into different categories and then adopt different speed-up strategies to reduce the encoding complexity. To evaluate the performance, extensive experiments are performed on the HEVC test model. Results show our method can averagely save 39.09% bits and reduce the encoding complexity by 43.63% on surveillance videos, whereas those are 5.27% and 43.68% on conference videos.

Mixed-Crossing-Avoided Escape Routing of Mixed-Pattern Signals on Staggered-Pin-Array PCBs

Escape routing has become a critical issue in high-speed PCB routing. Most of the previous work paid attention to either differential-pair escape routing or single-signal escape routing, but few considered them together. In this paper, a significant three-stage algorithm is proposed to solve the problem of escape routing of both differential pairs and single signals (mixed-pattern signals). First, differential pairs are preconditioned to reduce the complication of the problem. Then, a unified ILP model is used to formulate the problem and a novel Boolean coding-driven algorithm is proposed to avoid mixed crossings. Finally, a slice-based method is presented to prune the variables and speed up the algorithm. Experimental results show that the proposed method is very effective. For single-pattern escape routing, it can solve all the test cases in short time and reduce wire length and chip area by 16.1% and 15.5%, respectively. For mixed-pattern escape routing, it can increase the routability by 17.5% and reduce the wire length by 14.1% compared to a two-stage method. At the same time, the proposed method can effectively avoid mixed crossings with only a little increase on wire length. Furthermore, with slice-based speedup strategy, the method can reduce the solving time by 76.7%.

A vessel active contour model for vascular segmentation.

This paper proposes a vessel active contour model based on local intensity weighting and a vessel vector field. Firstly, the energy function we define is evaluated along the evolving curve instead of all image points, and the function value at each point on the curve is based on the interior and exterior weighted means in a local neighborhood of the point, which is good for dealing with the intensity inhomogeneity. Secondly, a vascular vector field derived from a vesselness measure is employed to guide the contour to evolve along the vessel central skeleton into thin and weak vessels. Thirdly, an automatic initialization method that makes the model converge rapidly is developed, and it avoids repeated trails in conventional local region active contour models. Finally, a speed-up strategy is implemented by labeling the steadily evolved points, and it avoids the repeated computation of these points in the subsequent iterations. Experiments using synthetic and real vessel images validate the proposed model. Comparisons with the localized active contour model, local binary fitting model, and vascular active contour model show that the proposed model is more accurate, efficient, and suitable for extraction of the vessel tree from different medical images.

Dynamic simulation based on Jacobian-free Newton-GMRES methods with adaptive preconditioner for power systems

Dynamic simulation is one of the most complex and important computations for power systems researches. Traditional solutions based on normal Newton iterations almost all depend on evaluations of Jacobian matrixes, which increases the programming complexity of and limits the parallelizability of the whole simulation. In this paper, a new adaptive preconditioned Jacobian-free Newton-GMRES(m) method is proposed to be applied to dynamic simulations of power systems. This new method has totally Jacobian-free characteristics, which saves calculations and storages of Jacobian matrixes and features strong parallelizability. Moreover, several speedup strategies are introduced to enhance efficiency and parallelizability of overall computations. Numerical tests are carried out on IEEE standard test systems and results show that in series computing environment, simulations based on the proposed method have comparable speed to those based on classical Newton-Raphson methods.

A flexible 3D cerebrovascular extraction from TOF-MRA images

For accurately extracting 3D cerebral tree from time-of-flight magnetic resonance angiography (TOF-MRA) images, a novel active contour model is presented by combining the statistical information and the vessel shape information in a variational level set framework. Firstly, a finite mixture model of four Gaussian distributions is proposed to model the statistical distribution of the density of TOF-MRA data, which can excellently describe the distribution of the cerebral vessels and background. Secondly, a vascular vector field, derived from the eigenanalysis of the Hessian matrix, is employed to model vessel shape information, which can guide the contour evolving along the vessel center line towards the thin or weak vessels. Thirdly, an automatic method of initializing the contour, derived from enhancement of the vessel shape, is proposed, which is capable of setting the initial contour efficiently and automatically. Furthermore, the narrow band technique is adopted, which effectively reduces the computation. Finally, a speedup strategy in the implementation is developed by labeling the steadily-evolved points, which avoids the repeated computation of these points in the later iterations. Experiments are made with several classical models over 9 MRA datasets, and the results show that the proposed model outperforms the classical models.

Computational optimization strategies for the simulation of random media and components

In this paper efficient computational strategies are presented to speed-up the analysis of random media and components. In particular, a Hybrid Stochastic Optimization (HSO) tool, based on the synergy between various algorithms, i.e. Genetic Algorithms, Simulated Annealing as well as Tabu-list is suggested to reconstruct a set of microstructures starting from probabilistic descriptors. The subsequent analysis (e.g. Finite Element analysis) can be performed to obtain the desired macroscopic quantity of interest and, providing a link between the micro- and the macro-scale. Different computational speed-up strategies are also presented. The proposed simulation approach is highly parallelizable, flexible and scalable. It can be adopted by other fields as well where an optimization analysis is required and a set of different solutions should be identified in order to perform computational experiments. Numerical examples demonstrate the applicability of the proposed strategies for realistic problems.

Efficient parallel algorithm for pixel classification in remote sensing imagery

An important approach for image classification is the clustering of pixels in the spectral domain. Fast detection of different land cover regions or clusters of arbitrarily varying shapes and sizes in satellite images presents a challenging task. In this article, an efficient scalable parallel clustering technique of multi-spectral remote sensing imagery using a recently developed point symmetry-based distance norm is proposed. The proposed distributed computing time efficient point symmetry based K-Means technique is able to correctly identify presence of overlapping clusters of any arbitrary shape and size, whether they are intra-symmetrical or inter-symmetrical in nature. A Kd-tree based approximate nearest neighbor searching technique is used as a speedup strategy for computing the point symmetry based distance. Superiority of this new parallel implementation with the novel two-phase speedup strategy over existing parallel K-Means clustering algorithm, is demonstrated both quantitatively and in computing time, on two SPOT and Indian Remote Sensing satellite images, as even K-Means algorithm fails to detect the symmetry in clusters. Different land cover regions, classified by the algorithms for both images, are also compared with the available ground truth information. The statistical analysis is also performed to establish its significance to classify both satellite images and numeric remote sensing data sets, described in terms of feature vectors.

A Scatter Search Algorithm for a Multistage Production Scheduling Problem With Blocking and Semi-Continuous Batching Machine

This paper studies a multistage production scheduling problem with blocking and semi-continuous batching machine, which is abstracted from the integrated hot rolling production in iron and steel industry. One major characteristic of the problem is that the first machine is a semi-continuous batching machine that can simultaneously process multiple jobs and the jobs do not enter and leave it in a batch mode, but one by one and continuously. Furthermore, this problem considers multiple production stages of hot rolling line while previous problems in the literature focused on only the single hot rolling stage or the two stages of reheating and hot rolling. This problem can be treated as a generalized permutation flowshop scheduling problem with blocking to minimize the makespan (i.e., the maximum completion time of all jobs), which is a NP-hard problem. We formulate this problem as a mixed integer linear programming model and propose a scatter search (SS) algorithm to solve it. To further improve the performance of the SS, the reference set is divided into three parts to balance the solution quality and diversity, and a modified stochastic variable neighborhood search is developed as the local search, where two kinds of speedup strategies based on the problem's characteristics are incorporated. Computational results on practical production data and randomly generated instances of our problem show that the SS algorithm outperforms the commercial software named CPLEX and some other meta-heuristics. In addition, further tests using benchmark instances of the traditional permutation flowshop scheduling problem with blocking also demonstrate that our SS algorithm is superior to previous meta-heuristics in the literature.

A Speed-Up Strategy for Finite Volume WENO Schemes for Hyperbolic Conservation Laws

In this paper, a speed-up strategy for finite volume WENO schemes is developed for solving hyperbolic conservation laws. It adopts p-adaptive like reconstruction, which automatically adjusts from fifth order WENO reconstruction to first order constant reconstruction when nearly constant solutions are detected by the undivided differences. The corresponding order of accuracy for the solutions is shown to be the same as obtained by original WENO schemes. The strategy is implemented with both WENO and mapped WENO schemes. Numerical examples in different space dimensions show that the strategy can reduce the computational cost by 20---40%, especially for problems with large fraction of constant regions.

On-chip sensor-driven efficient thermal profile estimation algorithms

This article addresses the problem of chip-level thermal profile estimation using runtime temperature sensor readings. We address the challenges of: (a) availability of only a few thermal sensors with constrained locations (sensors cannot be placed just anywhere); (b) random chip power density characteristics due to unpredictable workloads and fabrication variability. Firstly we model the random power density as a probability density function. Given such statistical characteristics and the runtime thermal sensor readings, we exploit the correlation in power dissipation among different chip modules to estimate the expected value of temperature at each chip location. Our methods are optimal if the underlying power density has Gaussian nature. We give a heuristic method to estimate the chip-level thermal profile when the underlying randomness is non-Gaussian. An extension of our method has also been proposed to address the dynamic case. Several speedup strategies are carefully investigated to improve the efficiency of the estimation algorithm. Experimental results indicated that, given only a few thermal sensors, our method can generate highly accurate chip-level thermal profile estimates within a few milliseconds.

Evolutionary Rough Parallel Multi-Objective Optimization Algorithm

A hybrid unsupervised learning algorithm, which is termed as Parallel Rough-based Archived Multi-Objective Simulated Annealing (PARAMOSA), is proposed in this article. It comprises a judicious integration of the principles of the rough sets theory and the scalable distributed paradigm with the archived multi-objective simulated annealing approach. While the concept of boundary approximations of rough sets in this implementation, deals with the incompleteness in the dynamic classification method with the quality of classification coefficient as the classificatory competencemeasurement, the time-efficient parallel approach enables faster convergence of the Pareto-archived evolution strategy. It incorporates both the rough set-based dynamic archive classifi- cation method and the distributed implementation as a two-phase speedup strategy in this algorithm. A measure of the amount of domination between two solutions has been incorporated in this work to determine the acceptance probability of a new solution with an improvement in the spread of the non-dominated solutions in the Pareto-front by adopting rough sets theory. A complexity analysis of the proposed algorithm is provided. An extensive comparative study of the proposed algorithm with three other existing and well-known Multi-Objective Evolutionary Algorithms (MOEAs) demonstrate the effectiveness of the former with respect to four existing performance metrics and eleven benchmark test problems of varying degrees of difficulties. The superiority of this new parallel implementation over other algorithms also has been demonstrated in timing, which achieves a near optimal speedup with a minimal communication overhead.

Collaborative assignment using belief-desire-intention agent modeling and negotiation with speedup strategies

In this paper, we propose a distributed agent model that applies belief-desire-intention (BDI) reasoning and negotiation for addressing the linear assignment problem (LAP) collaboratively. In resource allocation, LAP is viewed as seeking a concurrent allocation of one different resource for every task to optimize a linear sum objective function. The proposed model provides a basic agent-based foundation needed for efficient resource allocation in a distributed environment. A distributed agent algorithm that has been developed based on the BDI negotiation model is examined both analytically and experimentally. To improve performance in terms of average negotiation speed and solution quality, two initialization heuristics and two different reasoning control strategies are applied, with the latter yielding different variants of the basic algorithm. Extensive simulations suggest that all the heuristic-algorithm combinations can produce a near optimal solution soon enough in some specific sense. The significance and applicability of the research work are also discussed.