Search

The artificial intelligence inference model is currently the best mathematical model based on artificial neural networks. The intelligent reasoning mode based on artificial intelligence has a large construction scale. In a completed inference, many multiplication and addition operations need to be completed. To establish an effective artificial intelligence inference model, its computational complexity is dozens or even more times higher than traditional artificial intelligence algorithms. This article focused on the research on the isomerization acceleration of FPGA (Field Programmable Gate Array). This article was looking for a method that reduces both system changes and migration, while maximizing the flexibility of FPGA programmability. In order to build on this foundation, a fast interface for fast computation using FPGA was designed. The characteristic of this project was that it could combine the FPGA acceleration platform with ordinary computers or servers, and could effectively utilize traditional computer software and toolchains. This article compared the computational time consumption between the inference model and the CPU. In the scale calculation logic scenario of the inference model designed in this article, it could achieve the same computing power as the CPU when processing 70000 multiplication calculations. The experimental results indicated that the increase in CPU computing power was greater than that of the inference model. This indicated that the larger the amount of computational data, the more significant the acceleration effect of the inference model in this article would be. [ABSTRACT FROM AUTHOR]

This work provides an approach that automatically optimizes the component computations on graphics processing unit (GPU) devices from different vendors. The approach consists of a two-level optimization, where the first level considers the linear part of the computation for vectorization and applies mixed matrix formats to increase computational throughput further. Then, the second optimization level treats the combination of linear and non-linear parts as a black box and searches for the optimal configuration of parameters such as the degree of vectorization, the combination of matrix formats, and the group (of threads) sizes during parallel execution on GPU. Moreover, we also introduce constraints that reduce the optimization procedure's execution time. Finally, we select three different types of components that could be representative to computational tasks in power system and perform our optimization approach on these kernels. The computational performance is compared with unoptimized baseline and sparse linear algebra library based implementations, result shows that our optimization leads to better performance and more efficient memory utilization. • Automatic vectorization on linear algebraic operations. • Optimized memory access by combined matrix formats and storage strategies. • Automated benchmarking and code generation to allow automatic performance tuning. • Exploration space reduction algorithm to reduce benchmark time. • Performance outperforms the implementation based on linear algebra libraries. [ABSTRACT FROM AUTHOR]

[Display omitted] Parallel computing with multiple graphics processing units (GPUs) for adaptive mesh refinement (AMR) is a powerful method that improves the computational efficiency of phase-field (PF) simulations of dendrite growth. Octree-type block-structured AMR is often used in parallel computation because of its good parallelism; however, the process of filling the buffer region with physical data from adjacent blocks constitutes a major overhead cost that reduces computational performance. In this study, the mother-leaf (ML) method, which can reduce this overhead, was applied to accelerate parallel-GPU AMR PF simulations of dendrite growth. To evaluate the computational performance of the ML method, single- and parallel-GPU AMR PF simulations of dendrite growth were performed. The results show that single- and parallel-GPU AMR computations with the ML method accelerate the PF simulations of dendrite growth approximately 1.2–1.6 times and 1.2–1.4 times, respectively, compared to the conventional method without the ML method. In large-scale dendrite-growth PF simulations, such accelerations are significant for investigating various dendrite growth phenomena. [ABSTRACT FROM AUTHOR]

Computational holography involves creating complex holographic patterns, which is both fundamental and computationally intensive. This process presents significant challenges, particularly in achieving real-time hologram generation. This study presents a thorough comparison and analysis of computational efficiency for computing polygon-based computer-generated holograms (CGH) in terms of programming language (Python and MATLAB), execution hardware (CPU and GPU) and algorithms (interpolation-based and analytical-based). We open-sourced all the codes used for polygonal CGH executed in both MATLAB and Python, offering valuable insights into the performance suitability of different algorithms and languages. Basically, MATLAB demonstrates superior performance over Python, especially for CPU calculations, whereas it performs similarly when utilizing a graphics processing unit (GPU) and an accelerated algorithm like the wavefront recording plane (WRP) method. Analytical-based method and interpolation-based method are not consistently superior; the former performs well when addressing small matrices (e.g., using WRP), while the latter performs well when addressing large matrices. • Open-Source Contribution: We implemented CGH algorithms in MATLAB and Python, open-sourcing them for collaboration. • Comprehensive Comparison: We compared CGH computational efficiency in two programming languages across hardware platforms, and algorithmic methods. • Algorithm Efficiency: We objectively evaluated analytical-based and interpolation-based methods of polygon CGHs. In efficiency, the former excels with small matrices, while the latter performs better with large ones. • Guidance for Researchers: This work can help researchers choose the right tools and settings for their CGH applications. • Advancing Holography: This study highlights the pros and cons of current CGH tools, aiding future advancements in the field. [ABSTRACT FROM AUTHOR]

Tucker decomposition is used extensively for modeling multi-dimensional data represented as tensors. Owing to the increasing magnitude of nonzero values in real-world tensors, a growing demand has emerged for expeditious and scalable Tucker decomposition techniques. Several graphics processing unit (GPU)-accelerated techniques have been proposed for Tucker decomposition to decrease the decomposition speed. However, these approaches often encounter difficulties in handling extensive tensors owing to their huge memory demands, which exceed the available capacity of GPU memory. This study presents an expandable GPU-based technique for Tucker decomposition called GPUTucker. The proposed method meticulously partitions sizable tensors into smaller sub-tensors, which are referred to as tensor blocks , and effectively implements the GPU-based data pipeline by handling these tensor blocks asynchronously. Extensive experiments demonstrate that GPUTucker outperforms state-of-the-art Tucker decomposition methods in terms of the decomposition speed and scalability. [ABSTRACT FROM AUTHOR]

In order to efficiently perform the direct numerical simulation for shock wave boundary layer interference, a large grid computing scale is needed, and a single graphics processing unit cannot meet the urgent needs of computing. This paper proposes a parallel algorithm based on the architecture of multiple graphics processing units to effectively deal with the direct numerical simulation for shock wave boundary layer interference. The algorithm achieves a high overlap between graphics processing unit computation and boundary data exchange to improve the parallel efficiency when the grid computing scale is large. The working principle of the multi-graphics-processing-unit parallel algorithm is introduced, and the accuracy and feasibility of the algorithm are verified. Finally, through the application of the algorithm in the research of the compression-expansion slope boundary layer, fine flow structures such as large-scale vortex structure, shocklet, and attached shock wave can be captured. As a conclusion, the multi-graphics-processing-unit parallel algorithm proposed in this paper can accurately display the flow field of shock wave boundary layer interference. • A parallel algorithm based on multiple GPUs is proposed. • The accuracy, feasibility, and efficiency of the parallel algorithm are verified. • The DNS of the SWBLI based on the algorithm is tested and analyzed. [ABSTRACT FROM AUTHOR]

Spinal instrumentation and surgical manipulations may cause loss of navigation accuracy requiring an efficient re-alignment of the patient anatomy with pre-operative images during surgery. While intra-operative ultrasound (iUS) guidance has shown clear potential to reduce surgery time, compared with clinical computed tomography (CT) guidance, rapid registration aiming to correct for patient misalignment has not been addressed. In this article, we present an open-source platform for pedicle screw navigation using iUS imaging. The alignment method is based on rigid registration of CT to iUS vertebral images and has been designed for fast and fully automatic patient re-alignment in the operating room. Two steps are involved: first, we use the iUS probe's trajectory to achieve an initial coarse registration; then, the registration transform is refined by simultaneously optimizing gradient orientation alignment and mean of iUS intensities passing through the CT-defined posterior surface of the vertebra. We evaluated our approach on a lumbosacral section of a porcine cadaver with seven vertebral levels. We achieved a median target registration error of 1.47 mm (100% success rate, defined by a target registration error <2 mm) when applying the probe's trajectory initial alignment. The approach exhibited high robustness to partial visibility of the vertebra with success rates of 89.86% and 88.57% when missing either the left or right part of the vertebra and robustness to initial misalignments with a success rate of 83.14% for random starts within ±20° rotation and ±20 mm translation. Our graphics processing unit implementation achieves an efficient registration time under 8 s, which makes the approach suitable for clinical application. [ABSTRACT FROM AUTHOR]

Monte Carlo ray tracing (MCRT) is a fundamental simulation method for central receiver systems(CRSs). MCRT is an effective method to describe the radiative flux distribution on the receiver surface reflected by either a single heliostat or all heliostats in a heliostat field. In this paper, a GPU-based ray-tracing simulation method, namely, quasi-Monte Carlo ray tracing (QMCRT), is proposed to address problems of both efficiency and accuracy. First, QMCRT, as a bidirectional approach, can avoid unnecessary intersection calculations. This method also facilitates sunshape sampling and heliostat surface slope error sampling, which can achieve memory and run-time efficiency. Second, in the traditional approaches, the simulated maximum radiative flux (MaxRF) is randomly higher than the reference value, even if tens of millions of rays are traced. In QMCRT, the problem is solved by applying a trimmed mean smoothing operation to the generated radiative flux distribution. As a result, a stable MaxRF value approaching the reference value is obtained, while the total power remains almost unchanged. The results obtained for both synthetic and real heliostats obtained using QMCRT are substantially in keeping with the results obtained using established computational tools. QMCRT is two orders of magnitude faster than the traditional MCRT method when addressing traditional one-reflection CRS case. Compared with the state-of-the-art GPU-based grid ray tracing (GRT) approach, QMCRT is equally fast but generates a more accurate result. QMCRT also has an advantage in terms of efficiency for CRS compared with two well-known simulation software tools, i.e., SolTrace and Tonatiuh. • An efficient and accurate Monte Carlo ray tracing algorithm, QMCRT, is proposed. • The proposed ray tracer is applicable to both synthetic and measured heliostats. • A GPU-based inverse transform sampling is adopted to sample Buie's sun shape model. • The problem of instability of simulated maximum flux density is well addressed. [ABSTRACT FROM AUTHOR]

This paper presents and compares the implementation of a digital predistortion (DPD) linearizer for radiofrequency power amplifiers (PAs) considering two different hardware acceleration platforms. Both graphics processing unit (GPU)-based and field programmable gate arrays (FPGA)-based implementations of the DPD are capable to meet the high throughput requirements for linearizing the PA with current 5G new radio (NR) wideband signals. A comparison in terms of hardware resources usage, precision, throughput and linearization performance is provided for both GPU and FPGA implementations of a DPD based on the generalized memory polynomial (GMP) behavioral model. [Display omitted] • This paper focuses on the implementation of the GMP DPD model in both FPGA and GPU. • A trade-off among precision, resources usage and throughput performance is given. • The DPD performance is validated with a SLMBA with a hardware-in-loop test setup. [ABSTRACT FROM AUTHOR]

Abstract Geolocation is increasingly employed to reconstruct the movements of demersal fishes using data retrieved from electronic archival tags. However, geolocation methods commonly suffer from limitations such as low horizontal resolution of locations, flawed land boundary treatment, and extensive computation time. We addressed these issues using a state-space approach based on the particle filter (PF), and developed a geolocation package with graphics processing unit (GPU) acceleration. Our method focused on application to demersal fish and utilizes comparison of the tag-recorded depth and temperature to the same variables from an unstructured grid regional oceanographic model. A rigorous boundary treatment scheme was implemented to handle regions with complex coastline geometry. Validation exercises using stationary mooring tags and double-electronic-tagged (archival and acoustic tags) Atlantic cod in the Gulf of Maine resulted in <10 km median errors of the estimated tracks. Sensitivity analyses suggest that using 200,000 particles was adequate to stabilize the location track estimation. Acceleration of the particle filter using GPUs resulted in faster processing than the single threaded CPU (central processing unit) implementation, enabling rapid geolocations using consumer grade computer hardware. The geolocation output of each tagged fish includes the most probable track and the associated spatial probability distribution. The resulting PF geolocation package enables high resolution and accelerated geolocation analyses to be performed on affordable consumer-grade computer hardware, resolving the time intensiveness problem of the PF that may have prevented its adoptions in marine animal geolocation. Expanded application of geolocation will yield more reliable migration information to support management. Geolocation results from archival tagging will contribute to our understanding of the spatial ecology of marine species. [ABSTRACT FROM AUTHOR]

The proposed work develops a rapid and automatic method for brain tumour detection and segmentation using multi-sequence magnetic resonance imaging (MRI) datasets available from BraTS. The proposed method consists of three phases: tumourous slice detection, tumour extraction and tumour substructures segmentation. In phase 1, feature blocks and SVM classifier are used to classify the MRI slices into normal or tumourous. Phase 2 contains fuzzy c means (FCM) algorithm to extract the tumour region from slices identified by phase 1. In addition, graphics processing unit (GPU) based FCM method has been implemented for reducing the processing time which is major overhead with FCM processing of MRI volumes. For phase 3, a novel probabilistic local ternary patterns (PLTP) technique is used to segment the tumour substructures based on the probability density value of histogram bins. Quantitative measures such as sensitivity, specificity, accuracy and dice values are used to analyses the performance of the proposed method and compare with state-of-art-methods. As post processing, the tumour volume estimation and 3D visualization were done for analyzing the nature and location of the tumour to the medical experts. Further, the availability of the GPU reduces the processing time up to 18× than serial CPU processing. [ABSTRACT FROM AUTHOR]

Abstract Spatter formation is a longstanding issue in laser welding. This research addresses this topic and proposes a machine vision algorithm executed on a graphics processing unit to detect spatters in real time. Using this approach, a control system detecting spatter at a rate of 1 kHz and with a resolution of 900 x 900 pixels was implemented. Based on an experimental series, it is shown that the variation of the process parameters has a significant influence on the formation of spatter. It was also possible to quantify the variance of the spatter formation for a given set of process parameters. [ABSTRACT FROM AUTHOR]

Recently, cone-beam computed tomography (CBCT) equipped with image-guided radiation therapy (IGRT) has been used to precisely identify the location of target lesion. However, the treatment accuracy for respiratory-sensitive regions is still low due to long scanning time and target motion, and the imaging dose is also relatively high compared to other imaging systems. These issues can be solved by using 4D IGRT with digital tomosynthesis (DT). Also, an application of graphics processing unit (GPU) computing can reduce reconstruction time. The purpose of this study was to develop a 4D DT imaging technique for IGRT and compare image quality between 3D DT and 4D DT. And, we also compared reconstruction time between CPU and GPU computing. Projections were separately obtained through 3 phases for simulating 4D DT imaging. We measured profile, normalized root-mean-square error (NRMSE), contrast-to-noise ratio (CNR), noise-power spectrum (NPS) and figure-of-merit (FOM). NRMSEs and CNRs of 4D DT images were averagely 0.56 and 0.74 times lower than those of 3D DT images, respectively. The NPSs of 4D DT images were higher than 3D DT images. The FOMs of 4D DT images were averagely 1.65 times higher than 3D DT images. The GPU computing reduced reconstruction time in 3D and 4D imaging by a factor of 0.11 and 0.56, respectively. As a results, the 4D DT imaging enables the high-precision IGRT for respiratory-sensitive regions, and the real-time 4D imaging would be possible by using the GPU computing. [ABSTRACT FROM AUTHOR]

Abstract Optical Burst Switching (OBS) is a promising technology for next generation of Transparent Optical Networks (TON). However, many scientific challenges remain to be overcome such as the problem of Burst Routing and Wavelength Assignment (BRWA) with several conflicting objectives and constraints. In this paper, we first formulate the BRWA as a Multi Objective Integer Linear Programming (MO-ILP) optimization problem. In the formulated problem, the proposed BRWA policy will satisfy several constraints in order to guarantee a high-speed management of processes, required by the transparent optical traffic. Then, since the obtained ILP problem contains a large number of optical constraints and conflicting objectives, we propose to use an exact parallel Neural Hierarchical (epNH) MO-ILP solution with Graphics Processing Unit (GPU) parallel implementation using Compute Unified Device Architecture (CUDA). This also allows doing a concurrent search for multiple solutions, reducing processing cost, making hybrid interfaces to other search techniques, and achieving better overall effectiveness. In addition, our architecture based on Artificial Neural Networks (ANN) allows flexibility and scalability. The processing time remains fixed regardless of the input size. Our BRWA GPU-based epNH MO-ILP solver is based on the joint use of advanced MO-ILP optimization methods, ANN large-scale inherent parallelism and CUDA-GPU High-Performance Computing (HPC) architecture. [ABSTRACT FROM AUTHOR]

A proof of concept is presented on how to produce uncertainty-aware near real-time coastal flood inundation Web maps from water-level observations and predictions, which have been computed for tide gauge sites and made publicly accessible. The stochastic inundation simulation takes into account several sources of uncertainty, which have until now not been employed in either bathtub models or hydrodynamic models. The simulation is based on the Monte Carlo method. The feasibility of the proposed approach is demonstrated by an implementation using general-purpose computing on graphics processing units. The outcome of the research is that the current technologies enable the building of a novel system that connects to official data sources and that takes into account sources of uncertainty whose inclusion in the past has been avoided by being either weakly known or computationally too expensive. [ABSTRACT FROM AUTHOR]

Among the geological features, karst is the one that has received special attention in oil and gas exploration for being a strong indicator of the potential existence of hydrocarbon reservoirs. The integration of automatic pattern recognition methods and Graphics Processing Units (GPU) provides a powerful tool to help geological interpretation of seismic data. In order to provide insightful information for interpreters, this work investigates the usage of GPUs in addition to image segmentation by means of unsupervised classification for the identification of karst features in 3D seismic data. For this purpose, an implementation of the robust Self-Organizing Map for GPUs (SOM/GPU) is provided, and a comparison against a Central Processing Unit (CPU)-based SOM (SOM/CPU) is performed to assess the speeding-up provided by GPU. Experiments have shown promising results for geological interpretation using seismic data. [ABSTRACT FROM AUTHOR]

During the last decade, sampling based methods for motion and path planning have gained more interest. Specifically, in the field of robotics, approaches based on the Rapidly-exploring Random Tree (RRT) algorithm have become the customary technique for solving the single-query motion planning problem. However, dynamic large maps still represent a challenging scenario for these methods to produce fast enough results. Taking advantage of an NVidia CUDA-enabled Graphic Processing Unit (GPU), we present quad-RRT, an extension of the bi-directional strategy to speed up the RRT when dealing with large-scale, bidimensional (2D) maps. Designed for modern GPUs, quad-RRT computes four trees instead of the two ones built by the bidirectional approaches. This modification aims balancing the direct searching ability of these methods with the parallel exploration of those parts of the map at both sides of the path joining the initial and goal poses. Experimental results demonstrate that the proposed algorithm provides a significant speedup dealing with large-scale maps densely populated by obstacles, when compared to other implementations of the RRT. Hence, the algorithm can have a high impact in the field of inspection path planning for distributed infrastructure. It is also a promising approach to allow new generation robots, designed to work in unconstrained environments, dynamically plan large-scale paths. [ABSTRACT FROM AUTHOR]

Miniaturized grippers that possess an untethered structure are suitable for a wide range of tasks, ranging from micromanipulation and microassembly to minimally invasive surgical interventions. In order to robustly perform such tasks, it is critical to properly estimate their overall configuration. Previous studies on tracking and control of miniaturized agents estimated mainly their 2D pixel position, mostly using cameras and optical images as a feedback modality. This paper presents a novel solution to the problem of estimating and tracking the 3D position, orientation and configuration of the tips of submillimeter grippers from marker-less visual observations. We consider this as an optimization problem, which is solved using a variant of the Particle Swarm Optimization algorithm. The proposed approach has been implemented in a Graphics Processing Unit (GPU) which allows a user to track the submillimeter agents online. The proposed approach has been evaluated on several image sequences obtained from a camera and on B-mode ultrasound images obtained from an ultrasound probe. The sequences show the grippers moving, rotating, opening/closing and grasping biological material. Qualitative results obtained using both hydrogel (soft) and metallic (hard) grippers with different shapes and sizes ranging from 750 microns to 4 mm (tip to tip), demonstrate the capability of the proposed method to track the agent in all the video sequences. Quantitative results obtained by processing synthetic data reveal a tracking position error of 25 ± 7 μ m and orientation error of 1.7 ± 1.3 degrees. We believe that the proposed technique can be applied to different stimuli responsive miniaturized agents, allowing the user to estimate the full configuration of complex agents from visual marker-less observations. [ABSTRACT FROM AUTHOR]

The Traveling Thief Problem (TTP) is a multi-component combinatorial optimization problem that combines two well-known problems in the literature: the Traveling Salesman Problem (TSP) and the Knapsack Problem (KP). This paper proposes a novel list-constrained local search process inspired in Variable Neighborhood Descent (VND) for multiple neighborhood structures, combined with a metaheuristic Greedy Randomized Adaptive Search Procedure (GRASP). The local search implementation was made in a Graphics Processing Unit (GPU) architecture in order to explore the massive number of computing cores to simultaneously explore neighbor solutions, while the GRASP was implemented exploring the natural parallelism of a multi-core CPU. The computational results were compared to state-of-the-art results in literature and indicate promising research directions for the design of novel search algorithms in high performance architectures. [ABSTRACT FROM AUTHOR]

Acceleration of an existing MPI-based, particle-laden turbulent flow simulation code is achieved using up to four NVIDIA GPU devices. The overall design is to transfer the entire flow velocity, temperature, and humidity fields to each device, and compute particle trajectories entirely on the GPU hardware. For one-way coupled turbulent flow simulations, accurate simulations can be achieved for less computational cost than the original CPU implementation for particle numbers above 10 6 . Above 10 7 particles, the GPU version is roughly 14 times faster than the original CPU implementation. The effects of interpolation order, precision of the transferred Eulerian fields, and sub-time-stepping for fast particle dynamics are discussed. [ABSTRACT FROM AUTHOR]

Transparent Optical Networks has been known as a potential solution for high speed flexible optical backbone networks since a long time ago. However, many scientific challenges remain to be overcome such as the problem of Optical Burst Grooming (OBG) with several conflicting objectives and constraints. OBG is a fundamental problem in the engineering, control, and management of optical traffic networks, and arises in most network design applications, including optical burst routing, survivability design, and traffic scheduling. The traffic grooming problem is to coalesce several high and low speed sub bursts close in time together to form a larger burst that will be switched as one unit. In this paper, we first formulate OBG as a Multi Objective Integer Linear Programming (MO-ILP) optimization problem. Then we propose to use a parallel and hierarchical solver based on Artificial Neural Networks (ANN) with Graphics Processing Unit (GPU) parallel implementation using Compute Unified Device Architecture (CUDA). The processing time remains fixed regardless of the input size and the number of optical constraints and conflicting objectives. Our OBG solver benefits of the joint use of advanced ILP-MO optimization methods, ANN large-scale inherent parallelism and CUDA-GPU High-Performance Computing (HPC) architecture. Through a comprehensive simulation study, we show that, our proposed grooming approach can significantly improve the optical traffic performance, resulting in less contention and low burst lost probability. [ABSTRACT FROM AUTHOR]

As the new generation of smart sensors is evolving towards high sampling acquisitions systems, the amount of information to be handled by learning algorithms has been increasing. The Graphics Processing Unit (GPU) architecture provides a greener alternative with low energy consumption for mining big data, bringing the power of thousands of processing cores into a single chip, thus opening a wide range of possible applications. In this paper (a substantial extension of the short version presented at REM2016 on April 19–21, Maldives [1]), we design a novel parallel strategy for time series learning, in which different parts of the time series are evaluated by different threads. The proposed strategy is inserted inside the core a hybrid metaheuristic model, applied for learning patterns from an important mini/microgrid forecasting problem, the household electricity demand forecasting. The future smart cities will surely rely on distributed energy generation, in which citizens should be aware about how to manage and control their own resources. In this sense, energy disaggregation research will be part of several typical and useful microgrid applications. Computational results show that the proposed GPU learning strategy is scalable as the number of training rounds increases, emerging as a promising deep learning tool to be embedded into smart sensors. [ABSTRACT FROM AUTHOR]

Large-scale tridiagonal matrix solvers based on heterogeneous systems currently cannot balance computational efficiency and numerical stability when solving a non-diagonally dominant matrix. A tridiagonal solver combined central processing unit with graphics processing unit is proposed, based on SPIKE 2 as a solver framework, a simplified SPIKE algorithm as a central processing unit component, and a diagonal pivot algorithm as a graphics processing unit component. The solver performance is further improved by using a data-layout-transformation mechanism to obtain continuous addresses, reducing memory communication using constant memory to store unchanged data in the calculation process, and employing a kernel-fusion mechanism to reduce power consumption of graphics processing unit. For a diagonally dominant matrix, extended Thomas algorithms and cycle reduction to replace the graphics processing unit component are proposed in the solver. Experimental results show that the tridiagonal matrix solver in this paper can effectively consider both numerical stability and computational efficiency, and reduce total power consumption while improving memory efficiency. [ABSTRACT FROM AUTHOR]

This paper proposes a parallel computation model for the self-organizing map (SOM) neural network applied to Euclidean traveling salesman problems (TSP). This model is intended for implementation on the graphics processing unit (GPU) platform. The Euclidean plane is partitioned into an appropriate number of cellular units, called cells, and each cell is responsible of a certain part of the data and network. Compared to existing GPU implementations of optimization metaheuristics, which are often based on data duplication or mixed sequential/parallel solving, the advantage of the proposed model is that it is decentralized and based on data decomposition. Designed for handling large-scale problems in a massively parallel way, the required computing resources grow linearly along the problem size. Experiments are conducted on 52 publicly available Euclidean TSP instances with up to 85,900 cities for the largest TSPLIB instance and 71,009 cities for the largest National TSP instance. Experimental results show that our GPU implementations of the proposed model run significantly faster than the currently best-performing neural network approaches, to obtain results of similar quality. [ABSTRACT FROM AUTHOR]

Accurate estimation of peak wall stress (PWS) is the crux of biomechanically motivated rupture risk assessment for abdominal aortic aneurysms aimed to improve clinical outcomes. Such assessments often use the finite element (FE) method to obtain PWS, albeit at a high computational cost, motivating simplifications in material or element formulations. These simplifications, while useful, come at a cost of reliability and accuracy. We achieve research-standard accuracy and maintain clinically applicable speeds by using novel computational technologies. We present a solution using our custom finite element code based on graphics processing unit (GPU) technology that is able to account for added complexities involved with more physiologically relevant solutions, e.g. strong anisotropy and heterogeneity. We present solutions up to 17 × faster relative to an established finite element code using state-of-the-art nonlinear, anisotropic and nearly-incompressible material descriptions. We show a realistic assessment of the explicit GPU FE approach by using complex problem geometry, biofidelic material law, double-precision floating point computation and full element integration. Due to the increased solution speed without loss of accuracy, shown on five clinical cases of abdominal aortic aneurysms, the method shows promise for clinical use in determining rupture risk of abdominal aortic aneurysms. [ABSTRACT FROM AUTHOR]

In order to avoid the complex explicit feature extraction process and the problem of low-level data operation involved in traditional facial expression recognition, we proposed a method of Faster R-CNN (Faster Regions with Convolutional Neural Network Features) for facial expression recognition in this paper. Firstly, the facial expression image is normalized and the implicit features are extracted by using the trainable convolution kernel. Then, the maximum pooling is used to reduce the dimensions of the extracted implicit features. After that, RPNs (Region Proposal Networks) is used to generate high-quality region proposals, which are used by Faster R-CNN for detection. Finally, the Softmax classifier and regression layer is used to classify the facial expressions and predict boundary box of the test sample, respectively. The dataset is provided by Chinese Linguistic Data Consortium (CLDC), which is composed of multimodal emotional audio and video data. Experimental results show the performance and the generalization ability of the Faster R-CNN for facial expression recognition. The value of the mAP is around 0.82. [ABSTRACT FROM AUTHOR]

A new Infrared (IR) image analysis system will be deployed for real-time divertor protection during the upcoming Operational Phase (OP) 2.1 in the Wendelstein 7-X (W7-X) stellarator. Its primary objective is to prevent thermal overloads from permanently damaging Plasma-Facing Components (PFCs), resulting in machine downtimes and repair costs. The real-time constraint for this system is 110 ms, which is the maximum allowed delay entailing the acquisition, calibration, processing, and interlock, while all processing steps have to complete within 10 ms allowing for processing longer than the acquisition time of 10 ms at 100 Hz. This paper describes the implementation, real-time processing performance and detection effectiveness of Thermal Overload Detection (TOD). The implemented and evaluated TOD system fulfils real-time constraints. It reduces the total system delay to 50 ms and provides high detection sensitivity of 0.97 for archived discharge sequences from the OP1.2 campaign. The attained acceleration is significant, i.e., a 95% and 99% decrease in runtime for the sequential Central Processing Unit (CPU) and parallel Graphics Processing Unit (GPU) implementations, respectively, compared to the initial Python prototype. For the first time, the presented results confirm the feasibility of protecting W7-X in real-time comprising fundamentals for further advanced protection and control. • The first real-time protection of the plasma-facing components in Wendelstein 7-X. • The primary OP2.1 system to provide safe operation with the water-cooled components. • Accelerated full-frame processing of high-resolution infrared images with GPU and CPU. • Highly sensitive overload prediction algorithm for effective machine protection. • Evaluation and analysis of the algorithm on the archived discharges from OP1.2. [ABSTRACT FROM AUTHOR]

As the new generation of smart sensors is evolving towards high sampling acquisitions systems, the amount of information to be handled by learning algorithms has been increasing. The Graphics Processing Unit (GPU) architectures provide a greener alternative with low energy consumption for mining big-data, harnessing the power of thousands of processing cores in a single chip, opening a widely range of possible applications. Here, we design a novel evolutionary computing GPU parallel function evaluation mechanism, in which different parts of time series are evaluated by different processing threads. By applying a metaheuristics fuzzy model in a low-frequency data for household electricity demand forecasting, results suggested that the proposed GPU learning strategy is scalable as the number of training rounds increases. [ABSTRACT FROM AUTHOR]

• Subpixel displacement fields can be measured from pairs of reference/deformed speckle images with dedicated Convolutional Neural Networks (CNNs) • Simplifying such CNNs limits memory size and improves computational efficiency • This simplification can be made by limiting the number of levels and filters per level • Removing the deepest levels improves the displacement resolution but impairs the measurement resolution • Simplified CNNs have global metrological performance equivalent to that of non-simplified ones Convolutional Neural Networks (CNNs) are now commonly used in the computer vision community, in particular for optical flow estimation. Some attempts to use such tools to measure displacement and strain fields from pairs of reference/deformed speckle images (like Digital Image Correlation) have been recently reported in the literature. The aim of this work is twofold. The first one is to customize a state-of-the-art CNN dedicated to optical flow estimation to reach better performance when processing speckle images. This is mainly obtained by removing the deepest levels. The second one is to further simplify the CNN by reducing as much as possible the number of filters in the remaining levels while keeping equivalent metrological performance to the original version, in order to accelerate image processing on a power-efficient compact Graphics Processing Unit (GPU). Synthetic images deformed through a suitable displacement field are used to assess the metrological performance of the different versions of the CNN tested in this study. We focus the sub-pixel part of the displacement is considered for this first attempt, this part being much more challenging to determine than integer displacements obtained at the pixel scale. The latter can be found by cross-correlation or with a rough version of DIC. Real images are tested with the simplest version of the CNN and obtained results are compared with those provided by classic subset-based Digital Image Correlation. The two main conclusions are i- that the customization procedure improves the metrological performance of the original version, and that ii- the metrological performance of the ultimate simplified version of the CNN is globally equivalent to the one of the initial version despite the drastic simplification obtained at the end of the procedure. This performance lies between that of DIC used with first- and second-order subset shape functions. [ABSTRACT FROM AUTHOR]

We introduce a new resampling method, named Uphill, that is free from numerical instability and suitable for parallel implementation on graphics processing unit (GPU). Common resampling algorithms such as Systematic suffer from numerical instability when single precision floating point numbers are used. This is due to cumulative summation over the weights of particles when the weights differ widely or the number of particles is large. The Metropolis and Rejection resampling algorithms do not suffer from numerical instability as they only calculate the ratios of weights pairwise rather than perform collective operations over the weights. They are more suitable for the GPU implementation of the particle filter. However, they undergo non-coalesced global memory access patterns which cause their speed deteriorate rapidly as the number of particles gets large. Uphill also does not suffer from numerical instability but, experiences the same non-coalesced global memory access problem with Metropolis and Rejection. We introduce its faster version named Uphill-Fast which eliminates this problem. We make comparisons of Uphill and Uphill-Fast with the Systematic, Metropolis, Metropolis-C2 and Rejection resampling methods with respect to quality and speed. We also compare them on a highly non-linear system. Uphill-Fast runs faster and attains similar quality, in terms of RMSE, in comparison with Metropolis and Rejection when the number of particles is very large. Uphill-Fast runs with roughly same speed as Metropolis-C2 with better variance and MSE when the number of particles is very large. • A new fast resampling algorithm that is numerically stable with low variance & MSE. • It does not suffer from non-coalesced global memory access patterns on the GPU. • It achieves similar quality (RMSE) with the other resampling methods. • It is faster than Metropolis and Rejection when the number of particles is large. • It is as fast as Metropolis-C2 along with better variance and MSE. [ABSTRACT FROM AUTHOR]

The interaction noise caused by the periodic impingement of the forward blade wake on the backward blade widely exists in underwater rotating machinery. The rod-hydrofoil configuration is a model problem for interaction noise. Based on our in-house code of lattice Boltzmann method (LBM) solver accelerated by graphics processing unit (GPU), a direct numerical simulation (DNS) of rod-hydrofoil configuration with angles of attack (AoA) from 0° to 30° is carried out in the present work to study acoustic characteristics of the rod-hydrofoil configuration. The results show that GPU-accelerated LBM solver can achieve a DNS of hydrodynamics and acoustics in a short time and obtain the characteristics of the global noise field of rod-hydrofoil configuration accurately and efficiently. With the increase of AoA, the compound vortex structure of the rod-hydrofoil configuration changes greatly, and the hydrodynamic and acoustics characteristics also fluctuate greatly. When AoA <15°, the pure tonal noise caused by Karmen vortex street plays a dominant role. When AoA >15°, the effect of wake noise caused by the shedding of compound vortex structure in the wake region of hydrofoil becomes strong. • The GPU-acceleration LBM-DNS with high computing efficiency is verified. • Hydrodynamic and the vortex shedding characteristics of the rod-hydrofoil at different AoA are investigated. • The noise generation and propagation mechanism of rod-hydrofoil at different AoA are studied. • The relationship between global noise and complex vortex of the rod-hydrofoil configuration are revealed. [ABSTRACT FROM AUTHOR]

The technology of X-ray dual spectral computed tomography (DSCT) scans an object using two different x-ray energy spectra. The acquired DSCT raw data from this procedure can be used to measure the object by reconstructing its basis material images. Dual energy iterative algorithms can be applied to produce highly quantitative basis material images. However, they are not commonly employed in current medical applications due to the extensive computational demands required, such as for 3D cone-beam DSCT reconstruction. In order to improve the performance of the iterative algorithms for high-resolution cone-beam DSCT reconstruction, a graphics processing unit (GPU)-based multi-resolution approach is proposed. This approach divides the reconstructed volume into multiple sub-volumes, and each sub-volume is reconstructed at high resolution from the corrective polychromatic projections evaluated by a pre-reconstructed low-resolution volume. After all the sub-volumes are reconstructed, the sub-volumes are combined to yield the whole volume at high resolution. This approach requires less GPU memory, and significantly improves the performance of iterative reconstruction while maintaining a high image quality. The advantages of this approach were verified by numerical experiments in which FORBILD head phantom and thorax phantom were iteratively reconstructed from noise-free and noisy polychromatic projections. [ABSTRACT FROM AUTHOR]

2-D electromagnetic model with a line source uses Preconditioned Conjugate Gradient to solve the stiffness matrix and Herrmann's pseudo-delta function to simulate the neighborhood of the singular source. We employ this Finite Element Method (FEM) on Fermi-GPU, which is the third generation GPU for the general computing purpose. Our algorithm takes advantage of the GPU's particular characteristics of Single Instruction Multiple Thread (SIMT) architecture and runs entirely on the GPU to minimize the transfer time between CPU and GPU. Meanwhile, we design a reasonable pre-conditioner (Diagonal Pre-conditioner) on the GPU and inject a mixed scheme into our algorithm for obtaining the good performance. In the numerical experiments, the correctness of the algorithm is acknowledged, and it achieves high performance by outperforming CPU solvers over hundreds of times. Indeed, it allows finer mesh generation so as to obtain more accurate resolutions and advance the ability of data interpretation of geophysical prospecting. [ABSTRACT FROM AUTHOR]

Microstructure reconstruction and compression techniques are designed to identify microstructures with desired properties. While a microstructure reconstruction involves searching for a microstructure with prescribed statistical properties, a microstructure compression focuses on efficient representation of material morphology for the purpose of multiscale modelling. Successful application of these techniques, nevertheless, requires proper understanding of the underlying statistical descriptors quantifying morphology of a material. In this paper, we focus on a lineal path function designed to capture short-range effects and phase connectedness, which can hardly be handled by the commonly used two-point probability function. Usage of the lineal path function is, however, significantly limited because of huge computational requirements. So as to examine the properties of the lineal path function during computationally exhaustive compression and reconstruction processes, we start with an acceleration of the lineal path evaluation, namely by porting part of its code to a graphics processing unit using the CUDA (Compute Unified Device Architecture) programming environment. This allows us to present a unique comparison of the entire lineal path function with the commonly used rough approximation based on the Monte Carlo and/or sampling template. Moreover, this accelerated version of the lineal path function is then compared to the two-point probability function during the compression and reconstruction of two-phase morphologies. Their significant features are discussed and illustrated using a set of artificial periodic as well as real-world random microstructures. [ABSTRACT FROM AUTHOR]

In this paper, we describe parallel versions of two different variants (classical and alternating tree) of the Hungarian algorithm for solving the Linear Assignment Problem (LAP). We have chosen Compute Unified Device Architecture (CUDA) enabled NVIDIA Graphics Processing Units (GPU) as the parallel programming architecture because of its ability to perform intense computations on arrays and matrices. The main contribution of this paper is an efficient parallelization of the augmenting path search phase of the Hungarian algorithm. Computational experiments on problems with up to 25 million variables reveal that the GPU-accelerated versions are extremely efficient in solving large problems, as compared to their CPU counterparts. Tremendous parallel speedups are achieved for problems with up to 400 million variables, which are solved within 13 seconds on average. We also tested multi-GPU versions of the two variants on up to 16 GPUs, which show decent scaling behavior for problems with up to 1.6 billion variables and dense cost matrix structure. [ABSTRACT FROM AUTHOR]

In this paper, a graphics processing unit (GPU) based matching technique is proposed to perform fast feature matching between different images under various image conditions with viewpoint changes. Most recently, general-purpose graphics processing units (GPGPUs or GPUs) have become commonplace within high performance supercomputers. GPUs allow developers to effectively exploit the computational power for high performance computing. This paper focuses on improving the performance of feature matching based on self-organizing map by porting it onto the GPUs. GPU optimization has been applied for the fast computation of keypoints to make the system fast and efficient. This scheme has enhanced the overall performance and is much more efficient compared to other methods without degradation of detection results. The proposed matching algorithm is partitioned between the CPU and GPU in a way that best exploits the parallelism and perform matching between the different images. Experimental results demonstrate that fast feature matching is achieved using the graphics processing units, and its computational efficiency is checked by comparing its results and run times with those of some standard software (MATLAB) run on central processing unit (CPU). [ABSTRACT FROM AUTHOR]

Memory bound applications such as solvers for large sparse systems of equations remain a challenge for GPUs. Fast solvers should be based on numerically efficient algorithms and implemented such that global memory access is minimised. To solve systems with trillions ( O ( 10 12 ) ) unknowns the code has to make efficient use of several million individual processor cores on large GPU clusters. We describe the multi-GPU implementation of two algorithmically optimal iterative solvers for anisotropic PDEs which are encountered in (semi-) implicit time stepping procedures in atmospheric modelling. In this application the condition number is large but independent of the grid resolution and both methods are asymptotically optimal, albeit with different absolute performance. In particular, an important constant in the discretisation is the CFL number; only the multigrid solver is robust to changes in this constant. We parallelise the solvers and adapt them to the specific features of GPU architectures, paying particular attention to efficient global memory access. We achieve a performance of up to 0.78 PFLOPs when solving an equation with 0.55 · 10 12 unknowns on 16384 GPUs; this corresponds to about 3% of the theoretical peak performance of the machine and we use more than 40% of the peak memory bandwidth with a Conjugate Gradient (CG) solver. Although the other solver, a geometric multigrid algorithm, has a slightly worse performance in terms of FLOPs per second, overall it is faster as it needs less iterations to converge; the multigrid algorithm can solve a linear PDE with half a trillion unknowns in about one second. [ABSTRACT FROM AUTHOR]

In this paper, we will introduce a new method to investigate the best visualization of remote sensing imagery data sets on graphics processing units (GPU). The main shortcoming of classical dimension reduction methods is that their visualizations inevitably lose a significant amount of information in remote sensing imagery. Our proposed method (Enhancement Trustworthy (ET)) preserves the color distance as noticeably as possible for high dimension space by defining parallel-supervised extra step to enhance the quality of visualization. The ET classifies the visualization colors into false and true, and constructs a trustworthiness of CIE color space to improve the trustworthy value of false colors. The trustworthiness value of points inside CIE color space illustrate the locations of false and true projected colors in the visualization. Our method is supervised, where the user can select the interested points to enhance by ET. The proposed method is general enhancement step which can be used to enhance any visualization, for example, we used it to enhance the visualizations of remote sensing imagery data sets, which are carried by PCA, SPE and Isomap methods. The results showed the ability of ET to overcome the false colors and return the more reliable true colors. The colors of enhance visualizations showed more details which represent the real neighborhood relation in original data sets. [ABSTRACT FROM AUTHOR]

A sub-pixel digital image correlation (DIC) method with a path-independent displacement tracking strategy has been implemented on NVIDIA compute unified device architecture (CUDA) for graphics processing unit (GPU) devices. Powered by parallel computing technology, this parallel DIC (paDIC) method, combining an inverse compositional Gauss–Newton (IC-GN) algorithm for sub-pixel registration with a fast Fourier transform-based cross correlation (FFT-CC) algorithm for integer-pixel initial guess estimation, achieves a superior computation efficiency over the DIC method purely running on CPU. In the experiments using simulated and real speckle images, the paDIC reaches a computation speed of 1.66×10 5 POI/s (points of interest per second) and 1.13×10 5 POI/s respectively, 57–76 times faster than its sequential counterpart, without the sacrifice of accuracy and precision. To the best of our knowledge, it is the fastest computation speed of a sub-pixel DIC method reported heretofore. [ABSTRACT FROM AUTHOR]

In order to accelerate fast multipole boundary element method (FMBEM), in terms of the intrinsic parallelism of boundary elements and the FMBEM tree structure, a series of CUDA based GPU parallel algorithms for different parts of FMBEM with level-skip M2L for 3D elasticity are presented. A rigid body motion method (RBMM) for the FMBEM is proposed based on special displacement boundary conditions to deal with strongly singular integration and free term coefficients. The numerical example results show that our parallel algorithms obviously accelerates the FMBEM and can be used in large scale engineering problems with wide applications in the future. [ABSTRACT FROM AUTHOR]

Minimum variance (MV) beamforming has emerged as an adaptive apodization approach to bolster the quality of images generated from synthetic aperture ultrasound imaging methods that are based on unfocused transmission principles. In this article, we describe a new high-speed, pixel-based MV beamforming framework for synthetic aperture imaging to form entire frames of adaptively apodized images at real-time throughputs and document its performance in swine eye imaging case examples. Our framework is based on parallel computing principles, and its real-time operational feasibility was realized on a six-GPU (graphics processing unit) platform with 3,072 computing cores. This framework was used to form images with synthetic aperture imaging data acquired from swine eyes (based on virtual point-source emissions). Results indicate that MV-apodized image formation with video-range processing throughput (>20 fps) can be realized for practical aperture sizes (128 channels) and frames with λ/2 pixel spacing. Also, in a corneal wound detection experiment, MV-apodized images generated using our framework revealed apparent contrast enhancement of the wound site (10.8 dB with respect to synthetic aperture images formed with fixed apodization). These findings indicate that GPU-based MV beamforming can, in real time, potentially enhance image quality when performing synthetic aperture imaging that uses unfocused firings. [ABSTRACT FROM AUTHOR]

Control of autonomous systems subject to stochastic uncertainty is a challenging task. In guided airdrop applications, random wind disturbances play a crucial role in determining landing accuracy and terrain avoidance. This paper describes a stochastic parafoil guidance system which couples uncertainty propagation with optimal control to protect against wind and parameter uncertainty in the presence of impact area obstacles. The algorithm uses real-time Monte Carlo simulation performed on a graphics processing unit (GPU) to evaluate robustness of candidate trajectories in terms of delivery accuracy, obstacle avoidance, and other considerations. Building upon prior theoretical developments, this paper explores performance of the stochastic guidance law compared to standard deterministic guidance schemes, particularly with respect to obstacle avoidance. Flight test results are presented comparing the proposed stochastic guidance algorithm with a standard deterministic one. Through a comprehensive set of simulation results, key implementation aspects of the stochastic algorithm are explored including tradeoffs between the number of candidate trajectories considered, algorithm runtime, and overall guidance performance. Overall, simulation and flight test results demonstrate that the stochastic guidance scheme provides a more robust approach to obstacle avoidance while largely maintaining delivery accuracy. [ABSTRACT FROM AUTHOR]

Application of techniques for modelling of boundary value problems implies two conflicting requirements: obtaining high accuracy of the results and speed of the solution. Accurate results can be obtained only by using appropriate models and algorithms. In the previous papers the authors applied the parametric integral equations system (PIES) in modelling and solving boundary value problems. The first requirement was satisfied – the results were obtained with very high accuracy. This paper fulfils the second requirement by novel approach to accelerate PIES. Graphics cards (GPUs) programming for numerical calculations in general purpose applications (GPGPU) using NVIDIA CUDA is used for this purpose. The speed of calculations increased up to 80 times whereas high accuracy of the solutions was maintained. Examples included in this paper concern solving elasticity problems which are modelled by three-dimensional Navier–Lamé equations. [ABSTRACT FROM AUTHOR]

Membrane systems are parallel distributed computing models that are used in a wide variety of areas. Use of a sequential machine to simulate membrane systems loses the advantage of parallelism in Membrane Computing. In this paper, an innovative classification algorithm based on a weighted network is introduced. Two new algorithms have been proposed for simulating membrane systems models on a Graphics Processing Unit (GPU). Communication and synchronization between threads and thread blocks in a GPU are time-consuming processes. In previous studies, dependent objects were assigned to different threads. This increases the need for communication between threads, and as a result, performance decreases. In previous studies, dependent membranes have also been assigned to different thread blocks, requiring inter-block communications and decreasing performance. The speedup of the proposed algorithm on a GPU that classifies dependent objects using a sequential approach, for example with 512 objects per membrane, was 82×, while for the previous approach (Algorithm 1), it was 8.2×. For a membrane system with high dependency among membranes, the speedup of the second proposed algorithm (Algorithm 3) was 12×, while for the previous approach (Algorithm 1) and the first proposed algorithm (Algorithm 2) that assign each membrane to one thread block, it was 1.8×. [ABSTRACT FROM AUTHOR]