Decentralized cluster data warehouses are popular owing to their scalability and availability. Performance and resource consumption must be considered while creating such systems. Amdahl's law argues that the prop...
Decentralized cluster data warehouses are popular owing to their scalability and availability. Performance and resource consumption must be considered while creating such systems. Amdahl's law argues that the proportion of the workload that cannot be parallelized limits a system's speedup. this law affects system design in decentralized cluster data warehousing. In creating a decentralized cluster data warehouse, Amdahl's law requires detecting and evaluating bottlenecks in data input, storage, processing, and query execution. Designers may improve parallelizable components by understanding non-parallelizable components. A decentralized cluster data warehouse needs data propagation. Keeping frequently accessible data close may impair system performance. this requires exact data partition, replication, and cluster node data placement. Decentralized cluster data warehouses need technical, organizational, and governance concerns. Data security, privacy, regulatory compliance, and accessibility are examples. John Gustafson and Edwin Barsis' Gustafson's Law stresses parallel computing's scalability and workload rise. Gustafson's Law allows scaling up the issue and using more computer resources to handle higher workloads, unlike Amdahl's Law. the law shows that as the issue size increases, the percentage of the program that may be processed in parallel becomes a bigger fraction of the entire execution time, improving efficiency and speedup. Gustafson's Law promotes reducing execution time by dividing task over numerous processors. this abstract compares Gustafson's Law with Amdahl's Law, revealing parallel computing's advantages for bigger problem instances and the impact of increasing workload in parallel efficiency.
Subsurface rendering has become essential for photo-realistic rendering. therefore, an accurate simulation is too costly due to the interactions between the particles composing the material and the amount of light pas...
详细信息
Subsurface rendering has become essential for photo-realistic rendering. therefore, an accurate simulation is too costly due to the interactions between the particles composing the material and the amount of light passing through it. In this paper, we propose a method to estimate the subsurface light scattering effects. Our method comprises two steps: the first step consists of generating synthetic images using a reference technique to train the neural network. the second step consists in generating an approximation of the reference image after a learning step. Our model is based on convolutional architecture. the neural network is fed in input by images retrieved from the G-buffer(depth map, normal map,albedo map), while in output, it generates an approximation of the subsurface diffusion. We show that our method produces effective approximations of subsurface scattering. Furthermore, our method generates images comparable to the reference method and does not require processing of the output image. therefore, it does not do any calculation in the three-dimensional space of 3D scene.
Small objects detection is a major challenge in computer vision and one well-known example of this is helmet detection. Helmets have a small size, which limits typical YOLO (You Only Look Once) object detection models...
详细信息
ISBN:
(数字)9798350367157
ISBN:
(纸本)9798350367164
Small objects detection is a major challenge in computer vision and one well-known example of this is helmet detection. Helmets have a small size, which limits typical YOLO (You Only Look Once) object detection models from efficiently detecting them. Recent studies have focused on modifying several YOLO architectures to propose a solution to this problem. this study provides a comprehensive evaluation of novel solutions that address the helmet detection problem which include modifying multiple YOLO versions. A comprehensive examination of 53 research publications, which were published the period between 2019 to 2023, out of which 30 were included investigates the contributions in terms of model modifications, datasets used for evaluation, loss function optimization, and resulting performance measures. Several YOLO versions are addressed including YOLOv3, YOLOv4, YOLOv5, YOLOv7, and YOLOX. the studies discuss various modifications to YOLO including architecture modifications, integration of attention mechanisms, addition of detection scales, optimization and regularization techniques, loss function Improvements and anchor box optimization. the findings provide insight into the evolution of helmet detection algorithms within the YOLO architecture and how these modifications helped in addressing the problem of small objects detection. this review aims to provide academics and practitioners with significant insights into cutting-edge techniques for overcoming the challenges of identifying small objects as helmets.
Due to the containerized deployment environment, microservice orchestration requires coordination of multiple independently running services. this highly decoupled and distributed architecture increases the complexity...
详细信息
ISBN:
(数字)9798350389418
ISBN:
(纸本)9798350389425
Due to the containerized deployment environment, microservice orchestration requires coordination of multiple independently running services. this highly decoupled and distributed architecture increases the complexity of orchestration, and existing static resource allocation strategies are difficult to provide sufficient fault tolerance in the face of sudden failures. therefore, this article introduced a dynamic fault-tolerant scheduling algorithm based on Qlearning, which monitored service status in real-time, predicted potential faults, and dynamically adjusted resource allocation to improve system stability and fault tolerance. Firstly, microservices are modularized using the container orchestration tool Kubernetes, and containerization technology is utilized to encapsulate each service in a separate Pod for operation. Elastic allocation of resources can be achieved through Kubernetes’ automated scheduling mechanism, while efficient service communication and load balancing can be achieved through service mesh; Secondly, a dynamic faulttolerant scheduling algorithm based on Q-learning is introduced, combined with real-time monitoring data of CPU (Central processing Unit), memory, network traffic, etc., deployed in containers, to establish a fault prediction model. Finally, long short-term memory networks can be used to dynamically analyze and predict the load status and historical fault records of each service node, identify potential faults in real time, and automatically adjust resource allocation strategies based on the prediction results, reallocate service loads or start backup instances. the experimental results show that the Q-learningbased dynamic fault-tolerant scheduling algorithm effectively improves the system’s fault tolerance and stability. the analysis of fault recovery time showed that the recovery time for CPU overload, memory leakage, and network congestion was significantly reduced within 48 hours, to 7 seconds, 9 seconds, and 12 seconds, respec
this paper aims to smoothen the movements of an under-constrained cable-suspended parallel robot which carries a camera for video capturing purposes, especially for video capturing of football games. this goal is achi...
详细信息
this paper considers the possibility of massively parallel solution to the Lambert problem on graphics processing units. Several of the most popular solution algorithms were used for this problem. Software implementat...
详细信息
In recent years, remote sensing and other applications have used hyperspectral image processing in a variety of ways. For more precise and in-depth information extraction, hyperspectral images offer a wealth of spectr...
详细信息
In recent years, remote sensing and other applications have used hyperspectral image processing in a variety of ways. For more precise and in-depth information extraction, hyperspectral images offer a wealth of spectral information to recognize and discriminate spectrally identical materials. Numerous cutting-edge methods based on spectral and spatial data are available for hyperspectral image classification. Convolutional neural network (CNN), a subclass of artificial neural networks has gained popularity in a number of fields, including hyperspectral image classification. CNN is built to automatically and adaptively learn spatial hierarchies of data by backpropagation using a variety of building blocks, including convolution layers, pooling layers, and fully connected layers. In this paper, different CNN architectures such as IDCNN, 2D-CNN, 3D-CNN and 3D2D-CNN are evaluated to classify hyperspectral images. Experiments are performed on Indiana Pines and Pavia University images. Experimental results show that 3D2D-CNN gives highest classification accuracy.
In the traditional semantic segmentation of street scene, it is difficult to extract target features due to the different sizes of images, the occlusion of targets and the difficulty of recognizing small targets. Imag...
详细信息
ISBN:
(数字)9798350350890
ISBN:
(纸本)9798350350906
In the traditional semantic segmentation of street scene, it is difficult to extract target features due to the different sizes of images, the occlusion of targets and the difficulty of recognizing small targets. Images are difficult to be semantically segmented efficiently by these models. To solve the above problems, a pyramid network using spatial attention mechanism based on atrous convolutional pyramid (AMF-SPNet) is proposed. Firstly, the feature pyramid fused with dilated convolution is used to extract the features of the target. In the original ASPP model, we change the pooling layer to atrous convolution, and change the interior of the pyramid to atrous convolution with expansion rate 1,3,6,12,18. this model can expand the receptive field of the convolution kernel without increasing the number of parameters. Secondly, the spatial attention mechanism is added to the network, so that the model can better pay attention to the street view target information. Finally, the cityspaces dataset is used for verification, and the MIoU reaches 88.3%, the MAP reaches 89.6%. Compared with other network models, the proposed model can extract the location information and spatial information of street view images more accurately, so that the accuracy of semantic segmentation is improved.
MapReduce supports the processing of large data sets in parallel. It has been shown that MapReduce is an example for the use of the bulk synchronous parallel (BSP) bridging model, a model for parallel computation on a...
详细信息
ISBN:
(纸本)9783030775421;9783030775438
MapReduce supports the processing of large data sets in parallel. It has been shown that MapReduce is an example for the use of the bulk synchronous parallel (BSP) bridging model, a model for parallel computation on a fixed set of processors comprising alternating computation and communication phases. In this article we extend the normal execution of MapReduce from processing large finite data sets to processing stream queries with input data stream assumed to continue indefinitely. We classify stream queries into three classes, memoryless, semi-memoryless and memorable, and provide the model for each class using MapReduce based on BSP. In addition, as some stream queries require large amounts of computing sources, the BSP computation model is extended to a model with unbounded many agents, but preserving the barrier synchronization. A behavioral theory is developed for this model extending the behavioral theory of the BSP model. this comprises an axiomatization, the definition of Infinite-Agent BSP abstract state machines (Inf-Ag-BSP-ASM) and the proof that such ASMs capture the unbounded synchronized computations. Finally, we show how MapReduce processing can be further improved on grounds of the unbounded extension.
Since the emergence of multi-core systems, many efforts must be taken to make existing software take advantage of these new architectures. We exploit dense matrix kernels and node parallelism in the sparse LU factoriz...
详细信息
ISBN:
(纸本)9781665426251
Since the emergence of multi-core systems, many efforts must be taken to make existing software take advantage of these new architectures. We exploit dense matrix kernels and node parallelism in the sparse LU factorization, at the same time, also relying on third-party optimized multithreaded BLAS libraries. We introduce multi-threaded unified management mode and Task parallel optimization, targeting multi-core architectures. Our approach avoids a deep redesign and fully benefits from the numerical kernels and features of the original code. In this context, we propose simple approaches to take advantage of NUMA architectures. the performance gains are analyzed in detail on test problems, compared with MKL libraries.
暂无评论