We explore methods to automatically detect the quality in individual or batches of pharmaceutical products as they are manufactures. The goal is to detect 100% of the defects, not just statistically sample a small percentage of the products and draw conclusions that may not be 100% accurate. Removing all of the defective products, or halting production in extreme cases, will reduce costs and eliminate embarrassing and expensive recalls. We use the knowledge that experts have accumulated over many years, dynamic data derived from networks of smart sensors using both audio and chemical spectral signatures, multiple scales to look at individual products and larger quantities of products, and finally adaptive models and algorithms.

Hyperspectral band image selection is a fundamental problem for hyperspectral remote sensing data processing. Accepting its importance, several information-based band selection methods have been proposed, which apply Shannon entropy to measure image information. However, the Shannon entropy is not accurate in measuring image information since it neglects the spatial distribution of pixels and is computed only from a histogram. This paper investigates the potential of spatial entropy in measuring image information and proposes a new mutual information (MI) band selection method based on the spatial entropy. Then selected band images are validated for supervised classification via Support Vector Machine (SVM). Using a hyperspectral AVIRIS 92AV3C dataset, experiment results show that with 20 images selection from 220 bands, the supervised classification accuracy can reach 90.6%. Comparison with a previous Shannon entropy-based band selection method shows that the proposed method selects band images which can achieve more accurate classification results.

Branch Prediction is a common function in nowadays microprocessors. Branch predictor is duplicated in each core of a multi/many-core processor and makes prediction for multiple concurrent running programs respectively. To evaluate the parallel branch prediction in a multi/many-core processor, existing schemes generally use a parallel simulator running on a CPU that does not have a real massive parallel running environment to support the simulation and thus has a bad simulating performance. In this paper, we use a real many-core platform, GPU, to perform a parallel simulation of branch prediction for the future general purpose multi/many-core processor design. We verify the correctness of the GPU based parallel branch predictor against the traditional CPU based branch predictor. Experiment result shows that the GPU based parallel simulation scheme obtains a two to ten times of speedup over the CPU platform when the issue rate ranging from one to four instructions per cycle, and it shows that the GPU based scheme is a promising way to improve the simulation speed for future multi/many-core processor research.

Solving a super-high dimensional equations group is widely used in science and engineering, but the slow solution speed is the biggest problem researchers face. Research on FPGA based evolvable hardware chips for solving the super-high dimensional equations group (SHDESC) is proposed in this paper. These chips can be implemented on a million-gate scale FPGA chip. The core architecture of SHDESC is a systolic array which consists of thousands of special arithmetic units and can execute many super-high dimensional matrix operations parallelly in short time as well as really achieve the purpose of high speed solution in hardware/software codesign. The experiments show that these chips can achieve high precision results in a short period of time to solve a super-high dimensional equations group.

The Facility Location Problem is an important research topic in spatial analysis. This paper focuses on the Static and Mobile Facility Location (SMFL) Problem, which aims to identify those static and mobile facility locations that serve a target area most efficiently and equally. This paper formalizes the SMFL problem as a bi-objective model and then solves the model by using a novel heuristic algorithm, named Static and Mobile Facility Location Searching (SMFLS). The algorithm consists of two steps: static facility location searching and mobile facility location searching. In order to solve the model for large datasets efficiently, a clustering-based heuristic method is proposed for the static facility location searching while the mobile facility location searching is implemented using a greedy heuristic method. Experiments on synthetic datasets demonstrate the efficiency of the SMFLS algorithm. In addition, with the aim of conducting facility location decision-making conveniently and efficiently, in this paper, an interactive geospatial analysis platform, named Geospatial Analysis Platform using Interactive Maps (GAPIM) is proposed by combining the bi-objective models and the SMFLS algorithm with an interactive map. Experiments on Alberta public health service data are conducted, with the results demonstrating the efficiency and practicality of the platform.

Performance is a key feature of parallel system. However, there is a great gap between the peak performance and performance attainable by a practical application. The model-based performance evaluation may be used to support the performance-oriented program development for parallel system. In this paper a hybrid TCPN model is proposed to describe the parallel program and the resources respectively. This method can bring less effect to modify the program structure because of running environment changes. And the performance engineering activities based on this model ranges from performance prediction in early development stages, performance analysis in the coding phase, to locate the performance bottleneck and modify it. After the correctness verification of the TCPN model, a reachable graph can be got. Then the further performance-tuning can be done by summing the execution time of corresponding action in the critical path.

Solute transport in the subsurface is often considered to be a nonequilibrium process. Nonequilibrium during transport of solutes in porous medium has been categorized as either transport-related or sorption-related. For steady state flow in a homogeneous soil and assuming a linear sorption process, we will consider advection-diffusion adsorption equations. In this paper, numerical methods are considered for the mathematical model for steady state flow in a homogeneous soil with a linear sorption process. The modified upwind finite difference method is adopted to approximate the concentration in mobile regions and immobile regions. Optimal order $l^2$-error estimate is derived. Numerical results are supplied to justify the theoretical work.

The multilevel fast multipole algorithm (MLFMA) has shown great efficiency in solving large scale electromagnetic scattering problems. However, when unknowns become up to tens of millions, it is not trivial to keep high performance because of the complicated structure and calculation of MLFMA. In order to get rid of the bottleneck caused by load balancing, a parallel data partitioning strategy is proposed based on the hierarchical structure of an oct-tree of MLFMA. We present our data partitioning strategy in the light of different layers' properties including the processing of three kinds of layers in the tree and a fine-grained decomposition. We also put forward a solution of a coexisting data correlating problem, using a transition layer. Meanwhile, with the purpose of minimizing communication time in distributed memory system, a redundant technique is applied in the distributed layer. Parallel efficiency analysis demonstrates that the computational cost in parallelization of MLFMA can be asymptotically cut, and a high parallel efficiency can be obtained in our implementation.

Recently, Zhang and Lu proposed the generalized symmetric SOR (GSSOR) method for solving the nonsingular augmented systems and studied the convergence of the GSSOR method. In this paper, we prove the semi-convergence of the GSSOR method when it is applied to solve the singular augmented systems, which is the generalization of the GSSOR iteration method.

Standard compact scheme or upwind compact scheme have high order accuracy and high resolution, but cannot capture the shock without oscillations. In this paper, modified compact scheme is developed by using an effective shock detector to block upwinding compact scheme to cross the shock, a control function, and an adaptive scheme which uses some WENO flux near the shock. The new scheme makes the original upwinding compact scheme able to capture the shock sharper than WENO and, more important, keep high order accuracy and high resolution in the smooth area which is particularly important for shock, shock boundary layer interaction and shock acoustic interaction. The scheme is robust and has no case-related coefficients.

Metadata-based optimizations are the common methods to improve small files performance in local file systems. However, several problems will be introduced when applying the similar optimizations for small files in cluster file systems. In this paper, we study the tradeoffs between the performance of metadata and small files in metadata-based optimizations for a cluster file system. Our method aims to guarantee the metadata performance by adaptively migrating small files among file system nodes. We establish a theory model to analyze the small files load need to be migrated. To compute the migrated load in advance, a novel forecasting method is devised to accurately predict the one-step-ahead load of metadata and small files on a MDS. Then we propose a adaptive small file threshold model to decide the small files to be migrated. In the model, we consider the long-term and short-term tradeoffs respectively. To reduce the migration overhead, we discuss the migration tradeoffs for small files and present methods and schemes to eliminate unnecessary overheads. Finally, experiments are performed on a cluster file system and the results show the efficiency of our method in terms of promoting the load forecasting accuracy, trading off the performance of metadata and small files, and reducing migration overhead.

A TPCA (tri-valued programmable cellular automata) is proposed in this paper. Implemented based on TOC (Ternary Optical Computer) the TPCA has three advantages over other automata, that is the high programmability, the parallelism of computing and the tri-valued logic implementation. The programmability means that the transformation rules of each cell in CA can be modified at will and be any functions both linear and nonlinear. The parallelism comes from the advantage of optical computing and it can guarantee that CA even with very large-scale can be constructed in parallel and efficient. And the tri-valued implementation would make the CA be more flexible and complex than the counterparts in binary. Combining the characteristics of TOC, the TPCA is discussed in detail. Studied results show that the time complexity has nothing to do with the number of the cells in CA and it is just related to the complexity of the transformation functions. This means that it would be easy to construct more powerful and complicated CA and be widely used in many other fields.

To study the water entry ballistic trajectory of a torpedo, the wind tunnel experiment has been done based on the similarity principle. Then the drag coefficient of the torpedo is got when it enters the water, which is amended by the introduction of continuous supercavitation factor and local cavity effect factor. The vertical plane motion equations are established to get the torpedo's trajectory. The large scale nonlinear transient finite element commercial software MSC. dytran is also used to simulate the initial water entry impact of the Disk-Ogive-Head [1] torpedo, including four special high-speed water entry attitude angles. Then the kinematics parameters as the tail of torpedo submerges in water are input into the motion equation as the initial conditions. Finally, two parts of the data are combined to get the whole kinematic and kinetic parameters. During the calculation, the ballistic modeling uses the cavitation number to determine the torpedo's moving status: in the supercavitation stage, in partial cavity stage or in full wet navigation stage. The simulation results will do reference use to the following trajectory design. In addition, the water impact load and over load calculation of high-speed oblique water entry impact will help to design the intensity of torpedo's shell.

Similarity analysis of biomolecular networks among different species or within one species is an efficient approach to understand evolution or disease. The more data from biological experiment, the larger networks. Sequential computational limitation on single PC or workstation have to be considered when methods are developed. The Immediate Neighbors-in-first Method is a method for querying the subnetwork which is most similar to the target in a biomolecular network. Parallel algorithm for it to treat large-scale networks is developed and the parallel performance is evaluated in this paper. Moreover, we apply the present method to two groups of tests on real biological data including protein interaction networks of Fly and Yeast and metabolic networks of Yeast and E. coli. Several conserved protein interactions and metabolic pathways are found and some new protein interactions and functions are predicted.

This article focuses on evaluating the quality of sea surface temperature(SST) observed by satellite remote sensing. Under the premises that the scarcity of field measurement data and the abundant but overlapped multiple satellites detect information, in this article the consistency of multiple source information is used to verify the accuracy and reliability of satellite remote sensing data. Due to the limitation of Grubbs test when analyzing multi-source satellites SST, an improved algorithm is proposed, which is found to be more effectively than the traditional variance method when quantifying the differences and conflicts of SST. And the method is applied to the data extracted from 11 SST products in East China Sea, so a large amount of points set with high consistent can be confirmed, the outlying data can be discovered and eliminated, the waters (not include the outlying data) with confliction can be dig out and the conflicting level also can be quantized. It provides reference for the subsequent researchers to evaluate the quality of marine information.

Virtualization technology is widely used in server consolidation, high performance computing, and cloud data center due to its benefits on high resource utilization, flexible manageability, and dynamically scalability. However, it also introduces additional performance overheads. It’s worthy to evaluate the overheads and to find the bottleneck of virtualization in different scenarios. In this paper, we propose a combinative evaluation method to analyze the performance from single virtual machine to multiple virtual machine systems that measures and analyzes both the macro-performance and micro-performance. By correlating the analysis results of two-granularity performance data, some potential performance bottlenecks come out.

Particle Swarm Optimization (PSO) has shown its fast search speed in many complicated optimization and search problems. However, PSO often easily falls into local optima because the particles would quickly get closer to the best particle. Under these circumstances, the best particle could hardly be improved. This paper proposes a new hybrid PSO (HPSO) to solve this problem by combining space transformation search (STS) with a new modified velocity model. Experimental studies on 8 benchmark functions demonstrate that the HPSO holds good performance in solving both unimodal and multimodal functions optimization problems.

Image simplification, which reduces the information content of an image to suppress undesired details such as noise, is a very important basic ingredient of a lot of practical applications. The simplification of human brain MRI (Magnetic Resonance Imaging) is one of essential pre-processing steps for medical researches and clinical applications. Usually, the process of image simplification requires multiple iterations of reconstruction. Therefore, the efficiency of the reconstruction algorithm is a key problem. This paper has proposed an efficient reconstruction algorithm for MRI brain image simplification based on downhill filter. The main contribution of this paper is to use the regional maxima concept to modify the initialization condition of downhill filter algorithm. Experimental results show that the efficiency of this algorithm is much better than that of fast hybrid reconstruction algorithm, and it can achieve good result when it is used to the contour extraction from the MRI of human brain.

The Radar Cross Section (RCS) prediction for cavities is significant to measure a target's radar detection ability. For electrically large, deep, arbitrary shaped cavities, this paper presents a hybrid algorithm based parallel solution using Message Passing Interface on distributed memory computers. The meaning of `Hybrid' here is threefold. First, the RCS for cavities is modeled and calculated with a hybrid algorithm of IPO (Iterative Physical Optics), FMM (Fast Multipole Method) and Generalized Reciprocit Integral (GRI) combined by a cascading segmentation technique. Second, a hybrid approach is applied to the two phases of parallelization. On phase of geometrical preprocessing, all parallel processes assume a whole workload to construct the cavity geometry independently. On the other phase of electromagnetic computing, the workload is distributed by domain decomposition. Third, the decomposition scheme is hybrid as facets are decomposed to compute near-field interation while angle samples are used to distribute far-field interaction. The superposition of electromagnetic measures and permutability of math vector operations are fully exploited to do partial computation in order to minimize the communication overhead. The hybrid parallel solution can achieve very good tradeoff between memory and time. It yields a good load balance while still keeping the parallel code pretty concise. Numerical results show near-linear scalability and over 90% parallel efficiency.

This paper proposes a new means to complete the optical vector-matrix multiplication (OVMM). The OVMM is implemented on a novel optical computing architecture, ternary optical computer (TOC) by using modified signed-digit (MSD) number system. In addition, this study realizes optical addition in three steps, independent of the length of the operands, by four transformations in MSD, and discusses the multiplication of two MSD numbers in detail. A preliminary experimentation about the OVMM is performed on TOC. It illustrates that the proposed method for OVMM is feasible and correct.

Lattice Boltzmann Method (LBM) is a novel numerical method for flows simulations. Compared with classic methods of Finite Difference Method, Finite Volume Method and Finite Element Method, LBM has numerous advantages, including inherent parallelization and simplicity of boundary condition treatment. The LBM usually has a constraint of incompressible fluid (Mach number less than 0.4). A variant of the LBM is studied and used to deal with compressible fluid with Mach number up to 0.9 in this paper. Special emphasis is placed on mesh generation of 3-D complete geometry in Cartesian coordinate system. Numerical experiments are fulfilled in 2-D and 3-D compressible flows. Performance evaluation of the algorithm demonstrates high parallel efficiency and prefect scalability. Numerical results indicate that the LBM is successful with the simulation of compressible fluid.

We first analyze a stabilized finite volume method for the three-dimensional stationary Navier-Stokes equations. This method is based on local polynomial pressure projection using low order elements that do not satisfy the inf-sup condition. Then we derive a general superconvergent result for the stabilized finite volume approximation of the stationary Navier-Stokes equations by using a $L^2$-projection. The method is a postprocessing procedure that constructs a new approximation by using the method of least squares. The superconvergent results have three prominent features. First, they are established for any quasi-uniform mesh. Second, they are derived on the basis of the domain and the solution for the stationary Navier-Stokes problem by solving sparse, symmetric positive definite systems of linear algebraic equations. Third, they are obtained for the finite elements that fail to satisfy the inf-sup condition for incompressible flow. Therefore, this method presented here is of practical importance in scientific computation.

This work faces the problem of quality and prediction time assessment in a Dynamic Data Driven Application System (DDDAS) for predicting natural hazard evolution. Natural hazard management is undoubtedly a relevant area where systems modeling and numerical analysis take a great prominence.
Modeling such systems is a very hard problem to tackle. Besides, the results obtained by simulators usually don't provide accurate information, mostly due to the underlying uncertainty in the input parameters that define the actual environmental conditions at the very beginning of the simulation.
For this reason, we have developed a two-stage prediction strategy, which, first of all, carries out a parameter adjustment process by comparing the results provided by the simulator and the real observed hazard evolution. It has been demonstrated that this method improves notably the quality of the predictions. Furthermore, we have designed data injection techniques that allow us to take advantage from real-time acquired information, so that our strategy fits the DDDAS paradigm.
Nevertheless, because of the urgent nature of the systems we deal with, it is also necessary to assess the time incurred in applying the above mentioned strategy, in order for it to be useful and applicable in a real emergency situation. In this sense, we have developed a new methodology for prediction time assessment under this kind of prediction environments, based on Artificial Intelligence techniques.
In this research work, we have chosen forest fires as a representative study case, although the exposed methods can be extrapolated to any kind of natural hazard.

In this paper, we study a local stabilized nonconforming finite element method for the generalized Stokes equations. This nonconforming method is based on two local Gauss integrals, and uses the equal order pairs of mixed finite elements on quadrilaterals. Optimal order error estimates are obtained for velocity and pressure. Numerical experiments performed agree with the theoretical results.

Standard Fourier spectral method can be used to solve a lot of problems with periodic boundary conditions. However, for non-periodic boundary condition problems, standard Fourier spectral method is not efficient or even useless. This work has developed a new way to use Fourier spectral method for non-periodic boundary condition problems. First, the original function is normalized and then a smooth buffer polynomial is developed to extend the normalized function domain. The new function will be smooth and periodic with both function values and derivatives, which is easy to be treated by standard FFT for high resolution. This method has obtained high order accuracy and high resolution with a penalty of 25% over standard Fourier spectral method, as shown by our examples. The scheme demonstrates to be robust. The method will be further used for simulation of transitional and turbulent flow.