Computational Science – ICCS 2018

In soft complex grounds, earthquakes cause damages with large deformation such as landslides and subsidence. Use of elasto-plastic models as the constitutive equation of soils is suitable for evaluation of nonlinear wave propagation with large ground deformation. However, there is no example of elasto-plastic nonlinear wave propagation analysis method capable of simulating a large-scale soil deformation problem. In this study, we developed a scalable elasto-plastic nonlinear wave propagation analysis program based on three-dimensional nonlinear finite-element method. The program attains 86.2% strong scaling efficiency from 240 CPU cores to 3840 CPU cores of PRIMEHPC FX10 based Oakleaf-FX [1], with 8.85 TFLOPS (15.6% of peak) performance on 3840 CPU cores. We verified the elasto-plastic nonlinear wave propagation program through convergence analysis, and conducted an analysis with large deformation for an actual soft ground modeled using 47,813,250 degrees-of-freedom.

We report on a two-scale approach for efficient matrix-free finite element simulations. The proposed method is based on surrogate element matrices constructed by low-order polynomial approximations. It is applied to a Stokes-type PDE system with variable viscosity as is a key component in mantle convection models. We set the ground for a rigorous performance analysis inspired by the concept of parallel textbook multigrid efficiency and study the weak scaling behavior on SuperMUC, a peta-scale supercomputer system. For a complex geodynamical model, we achieve a parallel efficiency of 95% on up to 47 250 compute cores. Our largest simulation uses a trillion ($$\mathcal {O}(10^{12})$$) degrees of freedom for a global mesh resolution of 1.7 km.

The computation of crustal deformation following a given fault slip is important for understanding earthquake generation processes and reduction of damage. In crustal deformation analysis, reflecting the complex geometry and material heterogeneity of the crust is important, and use of large-scale unstructured finite-element method is suitable. However, since the computation area is large, its computation cost has been a bottleneck. In this study, we develop a fast unstructured finite-element solver for GPU-based large-scale computers. By computing several times steps together, we reduce random access, together with the use of predictors suitable for viscoelastic analysis to reduce the total computational cost. The developed solver enabled 2.79 times speedup from the conventional solver. We show an application example of the developed method through a viscoelastic deformation analysis of the Eastern Mediterranean crust and mantle following a hypothetical M 9 earthquake in Greece by using a 2,403,562,056 degree-of-freedom finite-element model.

Based on previous work on in-situ data transfer infrastructure and compiler-based software analysis, we have designed a virtual observation system for real time computer simulations. This paper presents an event detection framework for a virtual observation system. By using signal processing and detection approaches to the memory-based data streams, this framework can be reconfigured to capture high-frequency events and low-frequency events. These approaches used in the framework can dramatically reduce the data transfer needed for in-situ data analysis (between distributed computing nodes or between the CPU/GPU nodes). In the paper, we also use a terrestrial ecosystem system simulation within the Earth System Model to demonstrate the practical values of this effort.

Adaptive mesh refinement (AMR) can be used to improve climate simulations since these exhibit features on multiple scales which would be too expensive to resolve using non-adaptive meshes. In particular, long-term climate simulations only allow for low resolution simulations using current computational resources. We apply AMR to single components of the existing earth system model (ESM) instead of constructing a complex ESM based on AMR. In order to compatibly incorporate AMR into an existing model, we explore the applicability of a tree-based data structure. Using a numerical scheme for tracer transport in ECHAM6, we test the performance of AMR with our data structure utilizing an idealized test case. The numerical results show that the augmented data structure is compatible with the data structure of the original model and also demonstrate improvements of the efficiency compared to non-adaptive meshes.

This article proposes a bivariate compensated Volk and Schumaker (CompVSTP) algorithm, which extends the compensated Volk and Schumaker (CompVS) algorithm, to evaluate Bèzier tensor product surfaces with floating-point coefficients and coordinates. The CompVSTP algorithm is obtained by applying error-free transformations to improve the traditional Volk and Schumaker tensor product (VSTP) algorithm. We study in detail the forward error analysis of the VSTP, CompVS and CompVSTP algorithms. Our numerical experiments illustrate that the Comp-VSTP algorithm is much more accurate than the VSTP algorithm, relegating the influence of the condition numbers up to second order in the rounding unit of the computer.

In this paper a novel hybridization of agent-based evolutionary system (EMAS, a metaheuristic putting together agency and evolutionary paradigms) is presented. This method assumes utilization of particle swarm optimization (PSO) for upgrading certain agents used in the EMAS population, based on agent-related condition. This may be perceived as a method similar to local-search already used in EMAS (and many memetic algorithms). The obtained and presented in the end of the paper results show the applicability of this hybrid based on a selection of a number of 500 dimensional benchmark functions, when compared to non-hybrid, classic EMAS version.

In this paper, an agent-based data-driven model that focuses on path planning layer of origin/destination popularities and route choice is developed. This model improves on the existing mathematical modeling and pattern recognition approaches. The paths and origins/destinations are extracted from a video. The parameters are calibrated from density map generated from the video. We carried out validation on the path probabilities and densities, and showed that our model generates better results than the previous approaches. To demonstrate the usefulness of the approach, we also carried out a case study on capacity analysis of a building layout based on video data.

In this article, the researcher develops an image coding technique which is based on the wolf-sheep predation model. In the design, images are converted to virtual worlds of sheep, routes and wolves. Wolves in this model wander around searching for sheep while the algorithm tracks their movement. A wolf has seven movements which capture all the directions of the wolf. In addition, the researcher introduces one extra move of the wolf the purpose of which is to provide a shorter string of movements and to enhance the compression ratio. The first coordinates and the movements of the wolf are tracked and recorded. Then, arithmetic coding is applied on the string of movements to further compress it. The algorithm was applied on a set of images and the results were compared with other algorithms in the research community. The experimental results reveal that the size of the compressed string of wolf movements offer a higher reduction in space and the compression ratio is higher than those of many existing compression algorithms including G3, G4, JBIG1, JBIG2 and the recent agent-based model of ant colonies.

Mistakes in pedestrian infrastructure design in modern cities decrease transfer comfort for people, impact greenery due to appearance of desire paths, and thus increase the amount of dust in the air because of open ground. These mistakes can be avoided if optimal path networks are created considering behavioral aspects of pedestrian traffic, which is a challenge. In this article, we introduce Ant Road Planner, a new method of computer simulation for estimation and creation of optimal path networks which not only considers pedestrians’ behavior but also helps minimize the total length of the paths so that the area is used more efficiently. The method, which includes a modeling algorithm and its software implementation with a user-friendly web interface, makes it possible to predict pedestrian networks for new territories with high precision and detect problematic areas in existing networks. The algorithm was successfully tested on real territories and proved its potential as a decision making support system for urban planners.

To describe the diversity of opinions and dynamics of their changes in a society, there exist different approaches—from macroscopic laws of political processes to individual-based cognition and perception models. In this paper, we propose mesoscopic individual-based model of opinion dynamics which tackles the role of context by considering influence of different sources of information during life cycle of agents. The model combines several sub-models such as model of generation and broadcasting of messages by mass media, model of daily activity, contact model based on multiplex network and model of information processing. To show the applicability of the approach, we present two scenarios illustrating the effect of the conflicting strategies of informational influence on a population and polarization of opinions about topical subject.

In the paper, a heuristic algorithm for tensor product approximation with B-spline basis functions of three-dimensional material data is presented. The algorithm has an application as a preconditioner for implicit dynamics simulations of a non-linear flow in heterogeneous media using alternating directions method. As the simulation use-case, a non-stationary problem of liquid fossil fuels exploration with hydraulic fracturing is considered. Presented algorithm allows to approximate the permeability coefficient function as a tensor product what in turn allows for implicit simulations of the Laplacian term in the partial differential equation. In the consequence the number of time steps of the non-stationary problem can be reduced, while the numerical accuracy is preserved.

We consider a Cartesian decomposition of datasets, i.e. finding datasets such that their unordered Cartesian product yields the source set, and some natural generalization of this decomposition. In terms of relational databases, this means reversing the SQL CROSS JOIN and INNER JOIN operators (the last is equipped with a test verifying the equality of a tables attribute to another tables attribute). First we outline a polytime algorithm for computing the Cartesian decomposition. Then we describe a polytime algorithm for computing a generalized decomposition based on the Cartesian decomposition. Some applications and relating problems are discussed.

Manifold learning is a main stream research track used for dimensionality reduction as a method to select features. Many variants have been proposed with good performance. A novel graph-based algorithm for supervised image classification is introduced in this paper. It makes the use of graph embedding to increase the recognition accuracy. The proposed algorithm is tested on four benchmark datasets of different types including scene, face and object. The experimental results show the validity of our solution by comparing it with several other tested algorithms.

Relation classification is one of the most important topics in Natural Language Processing (NLP) which could help mining structured facts from text and constructing knowledge graph. Although deep neural network models have achieved improved performance in this task, the state-of-the-art methods still suffer from the scarce training data and the overfitting problem. In order to solve this problem, we adopt the adversarial training framework to improve the robustness and generalization of the relation classifier. In this paper, we construct a bidirectional recurrent neural network as the relation classifier, and append word-level attention to the input sentence. Our model is an end-to-end framework without the use of any features derived from pre-trained NLP tools. In experiments, our model achieved higher F1-score and better robustness than comparative methods.

Given a microblog post and a topic, it is an important task to judge the sentiment towards that topic: positive or negative, and has important theoretical and application value in the public opinion analysis, personalized recommendation, product comparison analysis, prevention of terrorist attacks, etc. Because of the short and irregular messages as well as containing multifarious features such as emoticons, and sentiment of a microblog post is closely related to its topic, most existing approaches cannot perfectly achieve cooperating analysis of topic and sentiment of messages, and even cannot know what factors actually determined the sentiment towards that topic. To address the issues, MB-LDA model and attention network are applied to Bi-RNN for topic-based microblog polarity classification. Our cascaded model has three distinctive characteristics: (i) a strong relationship between topic and its sentiment is considered; (ii) the factors that affect the topic’s sentiment are identified, and the degree of influence of each factor can be calculated; (iii) the synchronized detection of the topic and its sentiment in microblog is achieved. Extensive experiments show that our cascaded model outperforms state-of-the-art unsupervised approach JST and supervised approach SSA-ST significantly in terms of sentiment classification accuracy and F1-Measure.

Recommending the best and optimal content to user is the essential part of digital space activities and online user interactions. For example, we like to know what items should be sent to a user, what promotion is the best one for a user, what web design would fit a specific user, what ad a user would be more susceptible to or what creative cloud package is more suitable to a specific user.In this work, we use deep learning (autoencoders) to create a new model for this purpose. The previous art includes using Autoencoders for numerical features only and we extend the application of autoencoders to non-numerical features.Our approach in coming up with recommendation is using “matrix completion” approach which is the most efficient and direct way of finding and evaluating content recommendation.

AI and machine learning are mathematical modeling methods for learning from data and producing intelligent models based on this learning. The data these models need to deal with, is normally a mixed of data type where both numerical (continuous) variables and categorical (non-numerical) data types. Most models in AI and machine learning accept only numerical data as their input and thus, standardization of mixed data into numerical data is a critical step when applying machine learning models. Having data in the standard shape and format that models require often a time consuming, nevertheless very significant step of the process.

The world of AI and Machine Learning is the world of data and learning from data so the insights could be used for analysis and prediction. Almost all data sets are of mixed variable types as they may be quantitative (numerical) or qualitative (categorical). The problem arises from the fact that a long list of methods in Machine Learning such as “multiple regression”, “logistic regression”, “k-means clustering”, and “support vector machine”, all to be as examples of such models, designed to deal with numerical data type only. Though the data, that need to be analyzed and learned from, is almost always, a mixed data type and thus, standardization step must be undertaken for all these data sets. The standardization process involves the conversion of qualitative (categorical) data into numerical data type.

Latent Dirichlet Allocation (LDA) is a statistical technique for topic modeling. Since it is very computationally demanding, its parallelization has garnered considerable interest. In this paper, we systematically analyze the data access patterns for LDA and devise suitable algorithmic adaptations and parallelization strategies for GPUs. Experiments on large-scale datasets show the effectiveness of the new parallel implementation on GPUs.

We develop a reinforcement learning based search assistant which can assist users through a sequence of actions to enable them realize their intent. Our approach caters to subjective search where user is seeking digital assets such as images which is fundamentally different from the tasks which have objective and limited search modalities. Labeled conversational data is generally not available in such search tasks, to counter this problem we propose a stochastic virtual user which impersonates a real user for training and obtaining bootstrapped agent. We develop A3C algorithm based context preserving architecture to train agent and evaluate performance on average rewards obtained by the agent while interacting with virtual user. We evaluated our system with actual humans who believed that it helped in driving their search forward with appropriate actions without being repetitive while being more engaging and easy to use compared to conventional search interface.

The Adapteva Epiphany many-core architecture comprises a scalable 2D mesh Network-on-Chip (NoC) of low-power RISC cores with minimal uncore functionality. The Epiphany architecture has demonstrated significantly higher power-efficiency compared with other more conventional general-purpose floating-point processors. The original 32-bit architecture has been updated to create a 1,024-core 64-bit processor recently fabricated using a 16 nm process. We present here our recent work in developing an emulation and simulation capability for future many-core processors based on the Epiphany architecture. We have developed an Epiphany SoC device emulator that can be installed as a virtual device on an ordinary x86 platform and utilized with the existing software stack used to support physical devices, thus creating a seamless software development environment capable of targeting new processor designs just as they would be interfaced on a real platform. These virtual Epiphany devices can be used for research in the area of many-core RISC array processors in general.

Heterogeneous computing systems with multiple CPUs and GPUs are increasingly popular. Today, heterogeneous platforms are deployed in many setups, ranging from low-power mobile systems to high performance computing systems. Such platforms are usually programmed using OpenCL which allows to execute the same program on different types of device. Nevertheless, programming such platforms is a challenging job for most non-expert programmers. To enable an efficient application runtime on heterogeneous platforms, programmers require an efficient workload distribution to the available compute devices. The decision how the application should be mapped is non-trivial. In this paper, we present a new approach to build accurate predictive-models for OpenCL programs. We use a machine learning-based predictive model to estimate which device allows best application speed-up. With the LLVM compiler framework we develop a tool for dynamic code-feature extraction. We demonstrate the effectiveness of our novel approach by applying it to different prediction schemes. Using our dynamic feature extraction techniques, we are able to build accurate predictive models, with accuracies varying between 77% and 90%, depending on the prediction mechanism and the scenario. We evaluated our method on an extensive set of parallel applications. One of our findings is that dynamically extracted code features improve the accuracy of the predictive-models by 6.1% on average (maximum 9.5%) as compared to the state of the art.

Detection of Cardiac Arrhythmia (CA) is performed using the clinical analysis of the electrocardiogram (ECG) of a patient to prevent cardiovascular diseases. Machine Learning Algorithms have been presented as promising tools in aid of CA diagnoses, with emphasis on those related to automatic classification. However, these algorithms suffer from two traditional problems related to classification: (1) excessive number of numerical attributes generated from the decomposition of an ECG; and (2) the number of patients diagnosed with CAs is much lower than those classified as “normal” leading to very unbalanced datasets. In this paper, we combine in a coordinate way several data mining techniques, such as clustering, feature selection, oversampling strategies and automatic classification algorithms to create more efficient classification models to identify the disease. In our evaluations, using a traditional dataset provided by the UCI, we were able to improve significantly the effectiveness of Random Forest classification algorithm achieving an accuracy of over 88%, a value higher than the best already reported in the literature.

In the field of CT medical image reconstruction, there are two approaches you can take to reconstruct the images: the analytical methods, or the algebraic methods, which can be divided into iterative or direct.Although analytical methods are the most used for their low computational cost and good reconstruction quality, they do not allow reducing the number of views and thus the radiation absorbed by the patient.In this paper, we present two direct algebraic approaches for CT reconstruction: performing the Sparse QR (SPQR) factorization of the system matrix or carrying out a singular values decomposition (SVD). We compare the results obtained in terms of image quality and computational time cost and analyze the memory requirements for each case.

In recent years interest in blood viscosity has increased significantly in different biomedical areas. Blood viscosity, a measure of the resistance of blood flow, related to its thickness and stickiness, is one of the main biophysical properties of blood. Many factors affect blood viscosity, both in physiological and in pathological conditions.The aim of this study is to estimate blood viscosity by using the regression equation of viscosity which is based on hematocrit and total plasma proteins. It can be used to perform several observations regards the main factors which can influence blood viscosity. The main contribution regards the correlation between viscosity values and other important biological parameters such as cholesterol. This correlation has been supported by performing statistical tests and it suggest that the viscosity could be the main risk factor in cardiovascular diseases. Moreover, it is the only biological measure being correlated with the other cardiovascular risk factors. Results obtained are compliant with values obtained by using the standard viscosity measurement through a viscometer.

This paper proposes a robust high-quality finite element mesh generation method which is capable of modeling problems with complex geometries and multiple materials and suitable for the use in biomedical simulation. The previous octree-based method can generate a high-quality mesh with complex geometries and multiple materials robustly allowing geometric approximation. In this study, a robust mesh optimization method is developed combining smoothing and topology optimization in order to correct geometries guaranteeing element quality. Through performance measurement using sphere mesh and application to HTO tibia mesh, the validity of the developed mesh optimization method is checked.

Logistic regression as implemented in PLINK is a powerful and commonly used framework for assessing gene-gene (GxG) interactions. However, fitting regression models for each pair of markers in a genome-wide dataset is a computationally intensive task. Performing billions of tests with PLINK takes days if not weeks, for which reason pre-filtering techniques and fast epistasis screenings are applied to reduce the computational burden.Here, we demonstrate that employing a combination of a Xilinx UltraScale KU115 FPGA with an Nvidia Tesla P100 GPU leads to runtimes of only minutes for logistic regression GxG tests on a genome-wide scale. In particular, a dataset of 53,000 samples genotyped at 130,000 SNPs was analyzed in 8 min, resulting in a speedup of more than 1,000 when compared to PLINK v1.9 using 32 threads on a server-grade computing platform. Furthermore, on-the-fly calculation of test statistics, p-values and LD-scores in double-precision make commonly used pre-filtering strategies obsolete.

This paper presents improvements in financial time series prediction using a Deep Neural Network (DNN) in conjunction with a Discrete Wavelet Transform (DWT). When comparing our model to other three alternatives, including ARIMA and other deep learning topologies, ours has a better performance. All of the experiments were conducted on High-Frequency Data (HFD). Given the fact that DWT decomposes signals in terms of frequency and time, we expect this transformation will make a better representation of the sequential behavior of high-frequency data. The input data for every experiment consists of 27 variables: The last 3 one-minute pseudo-log-returns and last 3 one-minute compressed tick-by-tick wavelet vectors, each vector is a product of compressing the tick-by-tick transactions inside a particular minute using a DWT with length 8. Furthermore, the DNN predicts the next one-minute pseudo-log-return that can be transformed into the next predicted one-minute average price. For testing purposes, we use tick-by-tick data of 19 companies in the Dow Jones Industrial Average Index (DJIA), from January 2015 to July 2017. The proposed DNN’s Directional Accuracy (DA) presents a remarkable forecasting performance ranging from 64% to 72%.

As far as we know, Inverse Extreme Learning Machine (IELM) is the first work extending ELM to LLP problem. Due to basing on extreme learning machine (ELM), it obtains the fast speed and achieves competitive classification accuracy compared with the existing LLP methods. Kernel extreme learning machine (KELM) generalizes basic ELM to the kernel-based framework. It not only solves the problem that the node number of the hidden layer in basic ELM depends on manual setting, but also presents better generalization ability and stability than basic ELM. However, there is no research based on KELM for LLP. In this paper, we apply KELM and design the novel method LLP-KELM for LLP. The classification accuracy is greatly improved compared with IELM. Lots of numerical experiments manifest the advantages of our novel method.

Recurrent neural network are a type of deep learning units that are well studied to extract features from sequential samples. They have been extensively applied in forecasting univariate financial time series, however their application to high frequency multivariate sequences has been merely considered. This paper solves a classification problem in which recurrent units are extended to deep architecture to extract features from multi-variance market data in 1-minutes frequency and extreme market are subsequently predicted for trading signals. Our results demonstrate the abilities of deep recurrent architecture to capture the relationship between the historical behavior and future movement of high frequency samples. The deep RNN is compared with other models, including SVM, random forest, logistic regression, using CSI300 1-minutes data over the test period. The result demonstrates that the capability of deep RNN generating trading signal based on extreme movement prediction support more efficient market decision making and enhance the profitability.

Multi-view Multi-task (MVMT) Learning, a novel learning paradigm, can be used in extensive applications such as pattern recognition and natural language processing. Therefore, researchers come up with several methods from different perspectives including graph model, regularization techniques and feature learning. SVMs have been acknowledged as powerful tools in machine learning. However, there is no SVM-based method for MVMT learning. In order to build up an excellent MVMT learner, we extend PSVM-2V model, an excellent SVM-based learner for MVL, to the multi-task framework. Through experiments we demonstrate the effectiveness of the proposed method.

Based on the media news of Alibaba and improvement of L&M dictionary, this study transforms unstructured text into structured news sentiment through dictionary matching. By employing data of Alibaba’s opening price, closing price, maximum price, minimum price and volume in Thomson Reuters database, we build a fifth-order VAR model with lags. The AR test indicates the stability of VAR model. In a further step, the results of Granger causality tests, impulse response function and variance decomposition show that VAR model is successful to forecast variables dopen, dmax and dmin. What’s more, news sentiment contributes to the prediction of all these three variables. At last, MAPE reveals dopen, dmax and dmin can be used in the out-sample forecast. We take dopen sequence for example, document how to predict the movement and rise of opening price by using the value and slope of dopen.

Corner detection is a fundamental step for many image processing applications including image enhancement, object detection and pattern recognition. Recent years, the quality and the number of images are higher than before, and applications mainly perform processing on videos or image flow. With the popularity of embedded devices, the real-time processing on the limited computing resources is an essential problem in high-performance computing. In this paper, we study the parallel method of Harris corner detection and implement it on a heterogeneous architecture using OpenCL. We also adopt some optimization strategy on the many-core processor. Experimental results show that our parallel and optimization methods highly improve the performance of Harris algorithm on the limited computing resources.

In this paper, we present a new method JKSE+ for structured learning. Compared with some classical methods such as SSVM and CRFs, the optimization problem in JKSE+ is a convex quadratical problem and can be easily solved because it is based on JKSE. By incorporating the privileged information into JKSE, the performance of JKSE+ is improved. We apply JKSE+ to the problem of object detection, which is a typical one in structured learning. Some experimental results show that JKSE+ performs better than JKSE.

P2P online lending platforms have become increasingly developed. However, these platforms may suffer a serious loss caused by default behaviors of borrowers. In this paper, we present an effective default behavior prediction model to reduce default risk in P2P lending. The proposed model uses mobile phone usage data, which are generated from widely used mobile phones. We extract features from five aspects, including consumption, social network, mobility, socioeconomic, and individual attribute. Based on these features, we propose a joint decision model, which makes a default risk judgment through combining Random Forests with Light Gradient Boosting Machine. Validated by a real-world dataset collected by a mobile carrier and a P2P lending company in China, the proposed model not only demonstrates satisfactory performance on the evaluation metrics but also outperforms the existing methods in this area. Based on these results, the proposed model implies the high feasibility and potential to be adopted in real-world P2P online lending platforms.

Performance appraisal has always been an important research topic in human resource management. A reasonable performance appraisal plan lays a solid foundation for the development of an enterprise. Traditional performance appraisal programs are labor-based, lacking of fairness. Furthermore, as globalization and technology advance, in order to meet the fast changing strategic goals and increasing cross-functional tasks, enterprises face new challenges in performance appraisal. This paper proposes a data mining-based performance appraisal framework, to conduct an automatic and comprehensive assessment of the employees on their working ability and job competency. This framework has been successfully applied in a domestic company, providing a reliable basis for its human resources management.

Word representation is one foundational research in natural language processing which full of challenges compared to other fields such as image and speech processing. It embeds words to a dense low-dimensional vector space and is able to learn syntax and semantics at the same time. But this representation only get one single vector for a word no matter it is polysemy or not. In order to solve this problem, sense information are added in the multiple sense language models to learn alternative vectors for each single word. However, as the most popular measuring method in single sense language models, word similarity did not get the same performance in multiple situation, because word similarity based on cosine distance doesn’t match annotated similarity scores. In this paper, we analyzed similarity algorithms and found there is obvious gap between cosine distance and benchmark datasets, because the negative internal in cosine space does not correspond to manual scores space and cosine similarity did not cover semantic relatedness contained in datasets. Based on this, we proposed a new similarity methods based on mean square error and the experiments showed that our new evaluation algorithm provided a better method for word vector similarity evaluation.

This paper presents a hybrid algorithm employed to reduce the weight of an electric motor, designed for electric vehicle (EV) propulsion. The approach uses a hybridization between Cuckoo Search and CMAES to generate an initial population. Then, the population is transferred to a new procedure which adaptively switches between two search strategies, i.e. one for exploration and one for exploitation. Besides the electric motor optimization, the proposed algorithm performance is also evaluated using the 15 functions of the CEC 2015 competition benchmark. The results reveal that the proposed approach can show a very competitive performance when compared with different state-of-the-art algorithms.

This work presents computations of electric current distributions inside an industrial submerged arc furnace. A 3D model has been developed in ANSYS Fluent that solves Maxwell’s equations based on scalar and vector potentials approach that are treated as transport equations. In this paper, the approach is described in detail and numerical simulations are performed on an industrial three-phase submerged arc furnace. The current distributions within electrodes due to skin and proximity effects are presented. The results show that the proposed method adequately models these phenomena.

Deep Convolutional Neural Network (CNN), which is one of the prominent deep learning methods, has shown a remarkable success in a variety of computer vision tasks, especially image classification. However, tuning CNN hyper-parameters requires expert knowledge and a large amount of manual effort of trial and error. In this work, we present the use of CNN on classifying good quality images versus bad quality images without understanding the image content. The well known data-sets were used for performance evaluation. More importantly we propose a hyper-heuristic approach for tuning CNN hyper-parameters. The proposed hyper-heuristic encompasses of a high level strategy and various low level heuristics. The high level strategy utilises search performance to determine how to apply low level heuristics to automatically find an appropriate set of CNN hyper-parameters. Our experiments show the effectiveness of this hyper-heuristic approach which can achieve high accuracy even when the training size is significantly reduced and conventional CNNs can no longer perform well. In short the proposed hyper-heuristic approach does enhance CNN deep learning.

Shared bikes are wildly welcomed and becoming increasing popular in the world, as a result, quite a few bike sharing systems have been conducted to provide services for bike users. However, current bike sharing systems are not flexible and considerate enough for public bike users because of the fixed stations and not well emphasized about user’s satisfactions. In this paper, an agent-based distributed approach for bike sharing systems is proposed, this approach aims at helping users obtain a needed shared bike successfully and efficiently. We pay more attention on user’s preferences to improve the satisfaction to the target shared bike, meanwhile, trust and probability are considered to improve the efficiency and success rate. To the end, results from simulation studies demonstrate the effectiveness of our proposed method.

The prize-collecting Steiner tree problem (PCSTP) is one of the important topics in computational science and operations research. The vertex-swap operation, which involves removal and addition of a pair of vertices based on a given minimum spanning tree (MST), has been proven very effective for some particular PCSTP instances with uniform edge costs. This paper extends the vertex-swap operator to make it applicable for solving more general PCSTP instances with varied edge costs. Furthermore, we adopt multiple dynamic data structures, which guarantee that the total time complexity for evaluating all the $$O(n^2)$$ possible vertex-swap moves is bounded by $$O(n)\cdot O(m\cdot \,$$log$$\,n)$$, where n and m denote the number of vertices and edges respectively (if we run Kruskal’s algorithm with a Fibonacci heap from scratch after swapping any pair of vertices, the total time complexity would reach $$O(n^2) \cdot O(m + n\cdot \,$$log$$\,n)$$). We also prove that after applying the vertex-swap operation, the resulting solutions are necessarily MSTs (unless infeasible).

CSS-Sprite is a technique of regrouping small images of a web page, called tiles, into images called sprites in order to reduce network transfer time. CSS-sprite packing problem is considered as an optimization problem. We approach it as a probabilistic non-oriented two-dimensional bin packing problem (2PBPP|R). Our main contribution is to allow tiles rotation while packing them in sprites. An experimental study evaluated our solution, which outperforms current solutions.

In this work, we introduce slot selection and co-allocation algorithms for parallel jobs in distributed computing with non-dedicated and heterogeneous resources (clusters, CPU nodes equipped with multicore processors, networks etc.). A single slot is a time span that can be assigned to a task, which is a part of a parallel job. The job launch requires a co-allocation of a specified number of slots starting and finishing synchronously. The challenge is that slots associated with different heterogeneous resources of distributed computing environments may have arbitrary start and finish points, different pricing policies. Some existing algorithms assign a job to the first set of slots matching the resource request without any optimization (the first fit type), while other algorithms are based on an exhaustive search. In this paper, algorithms for effective slot selection are studied and compared with known approaches. The novelty of the proposed approach is in a general algorithm selecting a set of slots efficient according to the specified criterion.

In this paper, an explicit size reduction of a compact rat-race coupler implemented in a microstrip technology is considered. The coupler circuit features a simple topology with a densely arranged layout that exploits a combination of high- and low-impedance transmission line sections. All relevant dimensions of the structure are simultaneously optimized in order to explicitly reduce the coupler size while maintaining equal power split at the operating frequency of 1 GHz and sufficient bandwidth for return loss and isolation characteristics. Acceptable levels of electrical performance are ensured by using a penalty function approach. Two designs with footprints of 350 mm2 and 360 mm2 have been designed and experimentally validated. The latter structure is characterized by 27% bandwidth. For the sake of computational efficiency, surrogate-based optimization principles are utilized. In particular, we employ an iterative construction and re-optimization of the surrogate model involving a suitably corrected low-fidelity representation of the coupler structure. This permits rapid optimization at the cost corresponding to a handful of evaluations of the high-fidelity coupler model.

Probability of detection (POD) is used for reliability analysis in nondestructive testing (NDT) area. Traditionally, it is determined by experimental tests, while it can be enhanced by physics-based simulation models, which is called model-assisted probability of detection (MAPOD). However, accurate physics-based models are usually expensive in time. In this paper, we implement a type of stochastic polynomial chaos expansions (PCE), as alternative of actual physics-based model for the MAPOD calculation. State-of-the-art least-angle regression method and hyperbolic sparse technique are integrated within PCE construction. The proposed method is tested on a spherically-void-defect benchmark problem, developed by the World Federal Nondestructive Evaluation Center. The benchmark problem is added with two uncertainty parameters, where the PCE model usually requires about 100 sample points for the convergence on statistical moments, while direct Monte Carlo method needs more than 10000 samples, and Kriging based Monte Carlo method is oscillating. With about 100 sample points, PCE model can reduce root mean square error to be within 1% standard deviation of test points, while Kriging model cannot reach that level of accuracy even with 200 sample points.

We present an efficient way to solve the Bethe–Salpeter equation (BSE), a method for the computation of optical absorption spectra in molecules and solids that includes electron–hole interactions. Standard approaches to construct and diagonalize the Bethe–Salpeter Hamiltonian require at least $$\mathcal {O}(N_e^5)$$ operations, where $$N_e$$ is the number of electrons in the system, limiting its application to smaller systems. Our approach is based on the interpolative separable density fitting (ISDF) technique to construct low rank approximations to the bare exchange and screened direct operators associated with the BSE Hamiltonian. This approach reduces the complexity of the Hamiltonian construction to $$\mathcal {O}(N_e^3)$$ with a much smaller pre-constant, and allows for a faster solution of the BSE. Here, we implement the ISDF method for BSE calculations within the Tamm–Dancoff approximation (TDA) in the BerkeleyGW software package. We show that this novel approach accurately reproduces exciton energies and optical absorption spectra in molecules and solids with a significantly reduced computational cost.

The paper presents a computational framework for assessing quantitatively the detection capability of structural health monitoring (SHM) systems for flat plates. The detection capability is quantified using the probability of detection (POD) metric, developed within the area of nondestructive testing, which accounts for the variability of the uncertain system parameters and describes the detection accuracy using confidence bounds. SHM provides the capability of continuously monitoring the structural integrity using multiple sensors placed sensibly on the structure. It is important that the SHM can reliably and accurately detect damage when it occurs. The proposed computational framework models the structural behavior of flat plate using a spring-mass system with a lumped mass at each sensor location. The quantity of interest is the degree of damage of the plate, which is defined in this work as the difference in the strain field of a damaged plate with respect to the strain field of the healthy plate. The computational framework determines the POD based on the degree of damage of the plate for a given loading condition. The proposed approach is demonstrated on a numerical example of a flat plate with two sides fixed and a load acting normal to the surface. The POD is estimated for two uncertain parameters, the plate thickness and the modulus of elasticity of the material, and a damage located in one spot of the plate. The results show that the POD is close to zero for small loads, but increases quickly with increasing loads.

With the popularization of automobiles, more and more algorithms have been proposed in the last few years for the anomalous trajectory detection. However, existing approaches, in general, deal only with the data generated by GPS devices, which need a great deal of pre-processing works. Moreover, without the consideration of region’s local characteristics, those approaches always put all trajectories even though with different source and destination regions together. Therefore, in this paper, we devise a novel framework for anomalous trajectory detection between regions of interest by utilizing the data captured by Automatic Number-Plate Recognition (ANPR) system. Our framework consists of three phases: abstraction, detection, classification, which is specially engineered to exploit both spatial and temporal features. In addition, extensive experiments have been conducted on a large-scale real-world datasets and the results show that our framework can work effectively.

An effective nature disaster early warning system often relies on widely deployed sensors, simulation based predicting components, and a decision making system. In many cases, the simulation components require advanced infrastructures such as Cloud for performing the computing tasks. However, effectively customizing the virtualized infrastructure from Cloud based time critical constraints and locations of the sensors, and scaling it based on dynamic loads of the computation at runtime is still difficult. The suitability of a Dynamic Real-time Infrastructure Planner (DRIP) that handles the provisioning within cloud environments of the virtual infrastructure for time-critical applications is demonstrated with respect to disaster early warning systems. The DRIP system is part of the SWITCH project (Software Workbench for Interactive, Time Critical and Highly self-adaptive Cloud applications).

In the paper we propose ontology based scientific visualization tools to calibrate and monitor various IoT devices in a uniform way. We suggest using ontologies to describe associated controllers, chips, sensors and related data filters, visual objects and graphical scenes to provide self-service solutions for IoT developers and device makers. High-level interface of these solutions enables composing data flow diagrams defining both the behavior of the IoT devices and rendering features. According to the data flow diagrams and the set of ontologies the firmware for IoT devices is automatically generated incorporating both the data visualization and device behavior code. After the firmware loading, it’s possible to connect to these devices using desktop computer or smartphone/tablet, get the visualization client code over HTTP, monitor the data and calibrate the devices taking into account monitoring results. To monitor the distributed IoT networks a new visualization model based on circle graph is presented. We demonstrate the implementation of suggested approach within ontology based scientific visualization system SciVi. It was tested in a real world project of an interactive Permian Antiquities Museum exhibition creating.

As the mobile device gaining increasing popularity, Acoustic Speech Recognition on it is becoming a leading application. Unfortunately, the limited battery and computational resources on a mobile device highly restrict the potential of Speech Recognition systems, most of which have to resort to a remote server for better performance. To improve the performance of local Speech Recognition, we propose C-1-G-2-Blstm. This model shares Convolutional Neural Network’s ability of learning local feature and Recurrent Neural Network’s ability of learning sequence data’s long dependence. Furthermore, by adopting the Gated Convolutional Neural Network instead of a traditional CNN, we manage to greatly improve the model’s capacity. Our tests demonstrate that C-1-G-2-Blstm can achieve a high accuracy at 90.6% on the Google Speech Commands data set, which is 6.4% higher than the state-of-art methods.

Due to the development of IoT solutions, we can observe the constantly growing number of these devices in almost every aspect of our lives. The machine learning may improve increase their intelligence and smartness. Unfortunately, the highly regarded programming libraries consume to much resources to be ported to the embedded processors. Thus, in the paper the concept of source code generation of machine learning models is presented as well as the generation algorithms for commonly used machine learning methods. The concept has been proven in the use cases.

Very large climate data repositories provide a consistent view of weather conditions over long time periods. In some applications and studies, given a current weather pattern (e.g. today’s weather), it is useful to identify similar ones (weather analogues) in the past. Looking for similar patterns in an archive using a brute force approach requires data to be retrieved from the archive and then compared to the query, using a chosen similarity measure. Such operation would be very long and costly. In this work, a wavelet-based fingerprinting scheme is proposed to index all weather patterns from the archive. The scheme allows to answer queries by computing the fingerprint of the query pattern, then comparing them to the index of all fingerprints more efficiently, in order to then retrieve only the corresponding selected data from the archive. The experimental analysis is carried out on the ECMWF’s ERA-Interim reanalyses data representing the global state of the atmosphere over several decades. Results shows that 32 bits fingerprints are sufficient to represent meteorological fields over a 1700 km $${\times }$$ 1700 km region and allow the quasi instantaneous retrieval of weather analogues.

Assimilation of data into a fire-spread model is formulated as an optimization problem. The level set equation, which relates the fire arrival time and the rate of spread, is allowed to be satisfied only approximately, and we minimize a norm of the residual. Previous methods based on modification of the fire arrival time either used an additive correction to the fire arrival time, or made a position correction. Unlike additive fire arrival time corrections, the new method respects the dependence of the fire rate of spread on diurnal changes of fuel moisture and on weather changes, and, unlike position corrections, it respects the dependence of the fire spread on fuels and terrain as well. The method is used to interpolate the fire arrival time between two perimeters by imposing the fire arrival time at the perimeters as constraints.

Complex systems (e.g., volcanoes, debris flows, climate) commonly have many models advocated by different modelers and incorporating different modeling assumptions. Limited and sparse data on the modeled phenomena does not permit a clean discrimination among models for fitness of purpose, and, heuristic choices are usually made, especially for critical predictions of behavior that has not been experienced. We advocate here for characterizing models and the modeling assumptions they represent using a statistical approach over the full range of applicability of the models. Such a characterization may then be used to decide the appropriateness of a model for use, and, perhaps as needed weighted compositions of models for better predictive power. We use the example of dense granular representations of natural mass flows in volcanic debris avalanches, to illustrate our approach.

With the development of technology, the technology foresight becomes more and more important. Delphi method as the core method of technology foresight is increasingly questioned. This paper propose a new technology foresight method based on intelligent convergence in open network environment. We put a large number of scientific and technological innovation topics into the open network technology community. Through the supervision and guidance to stimulate the discussion of expert groups, a lot of interactive information can be generated. Based on the accurate topic delivery, effective topic monitoring, reasonable topic guiding, comprehensive topic recovering, and interactive data mining, we get the technology foresight result and further look for the expert or team engaged in relevant research.

In order to predict blasting vibration intensity accurately, support vector machine regression (SVR) was adopted to predict blasting vibration velocity, vibration frequency and vibration duration. The mutation operation of genetic algorithm (GA) is used to avoid the local optimal solution of particle swarm optimization (PSO). The improved PSO algorithm is used to search for the best parameters of SVR model. In the experiments, the improved PSO-SVR algorithm was realized on the Apache Spark platform. The execution time and prediction accuracy of the sadovski method, the traditional SVR algorithm, the neural network (NN) algorithm and the improved PSO-SVR algorithm were compared. The results show that the improved PSO-SVR algorithm on Spark is feasible and efficient, and the SVR model can predict the blasting vibration intensity more accurately than other methods.

The hp-adaptive Finite Element Method (hp-FEM) generates a sequence of adaptive grids with different polynomial orders of approximation and element sizes. The hp-FEM delivers exponential convergence of the numerical error with respect to the mesh size. In this paper, we propose a heuristic algorithm to construct element partition trees. The trees can be transformed directly into the orderings, which control the execution of the multi-frontal direct solvers during the hp refined finite element method. In particular, the orderings determine the number of floating point operations performed by the solver. Thus, the quality of the orderings obtained from the element partition trees is important for good performance of the solver. Our heuristic algorithm has been implemented in 3D and tested on a sequence of hp-refined meshes. We compare the quality of the orderings found by the heuristic algorithm to those generated by alternative state-of-the-art algorithms. We show 50% reduction in flops number and execution time.

Ellagic acid (EA) is a polyphenolic compound with antiviral activity against chikungunya, a rapidly spreading new tropical disease transmitted to humans by mosquitoes and now affecting millions worldwide. The most common symptoms of chikungunya virus infection are fever and joint pain. Other manifestations of infection can include encephalitis and an arthritic joint swelling with pain that may persist for months or years after the initial infection. The disease has recently spread to the U.S.A., with locally-transmitted cases of chikungunya virus reported in Florida. There is no approved vaccine to prevent or medicine to treat chikungunya virus infections. In this study, the Estimated Daily Intake (EDI) of EA from the food supply established using the National Health and Nutrition Examination Survey (NHANES) is used to set a maximum dose of an EA formulation for a high priority clinical trial.

Parallel computing using MPI has become ubiquitous on multi-node computing clusters. A common problem while developing parallel codes is determining whether or not a deadlock condition can exist. Ideally we do not want to have to run a large number of examples to find deadlock conditions through trial and error procedures. In this paper we describe a methodology using both static analysis and symbolic execution of a MPI program to make a determination when it is possible. We note that using static analysis by itself is insufficient for realistic cases. Symbolic execution has the possibility of creating a nearly infinite number of logic branches to investigate. We provide a mechanism to limit the number of branches to something computable. We also provide examples and pointers to software necessary to test MPI programs.

Roulette Wheel Sampling, sometimes referred to as Fitness Proportionate Selection, is a method to sample from a set of objects each with an associated weight. This paper introduces a distributed version of the method designed for message passing environments. Theoretical bounds are derived to show that the presented method has better scalability than naive approaches. This is verified empirically on a test cluster, where improved speedup is measured. In all tested configurations, the presented method performs better than naive approaches. Through a renumbering step, communication volume is minimized. This step also ensures reproducibility regardless of the underlying architecture.

The paper seeks to introduce a new algorithm for computation of interpolating spline surfaces over non-uniform grids with $$C^2$$ class continuity, generalizing a recently proposed approach for uniform grids originally based on a special approximation property between biquartic and bicubic polynomials. The algorithm breaks down the classical de Boor’s computational task to systems of equations with reduced size and simple remainder explicit formulas. It is shown that the original algorithm and the new one are numerically equivalent and the latter is up to 50% faster than the classic approach.

A discrete ordinates method has been developed to approximate the neutron transport equation for the computation of the lambda modes of a given configuration of a nuclear reactor core. This method is based on discrete ordinates method for the angular discretization, resulting in a very large and sparse algebraic generalized eigenvalue problem. The computation of the dominant eigenvalue of this problem and its corresponding eigenfunction has been done with a matrix-free implementation using both, the power iteration method and the Krylov-Schur method. The performance of these methods has been compared solving different benchmark problems with different dominant ratios.

In this paper, we summarize on an approach which couples the multiscale method with the homogenization theory to model the pre-treatment depth filtration process in desalination facilities. By first coupling the fluid and solute problems, we systematically derive the homogenized equations for the effective filtration process while introducing appropriate boundary conditions to account for the deposition process occurring on the spheres’ boundaries. Validation of the predicted results from the homogenized model is achieved by comparing with our own experimentally-derived values from a lab-scale depth filter. Importantly, we identify a need to include a computational approach to resolve for the non-linear concentration parameter within the defined periodic cell at higher orders of reaction. The computational values can then be introduced back into the respective homogenized equations for further predictions which are to be compared with the obtained experimental values. This proposed hybrid methodology is currently in progress.

High efficient methods are required for the computation of several lambda modes associated with the neutron diffusion equation. Multiple iterative eigenvalue solvers have been used to solve this problem. In this work, three different block methods are studied to solve this problem. The first method is a procedure based on the modified block Newton method. The second one is a procedure based on subspace iteration and accelerated with Chebyshev polynomials. Finally, a block inverse-free Krylov subspace method is analyzed with different preconditioners. Two benchmark problems are studied illustrating the convergence properties and the effectiveness of the methods proposed.

Failure of peripheral endovascular interventions occurs at the intersection of vascular biology, biomechanics, and clinical decision making. It is our hypothesis that most of the endovascular treatments share the same driving mechanisms during post-surgical follow-up, and accordingly, a deep understanding of them is mandatory in order to improve the current surgical outcome. This work presents a versatile model of vascular adaptation post vein graft bypass intervention to treat arterial occlusions. The goal is to improve the computational models developed so far by effectively modeling the cell-cell and cell-membrane interactions that are recognized to be pivotal elements for the re-organization of the graft’s structure. A numerical method is here designed to combine the best features of an Agent-Based Model and a Partial Differential Equations model in order to get as close as possible to the physiological reality while keeping the implementation both simple and general.

In this work I reflect on the development of a multiscale simulation approach for forced migration, and present two prototypes which extend the existing Flee agent-based modelling code. These include one extension for parallelizing Flee and one for multiscale coupling. I provide an overview of both extensions and present performance and scalability results of these implementations in a desktop environment.