检索结果-南通市图书馆

Performance Analys.s.of Accelerator Architectures.and Programming Models.for Parareal Algorithm s.lutions.of Ordinary Differential Equations

引用

Journal of Computer and Communications 2021年第2期9卷 29-56页

作者： s.mathi Laks.miranganatha suresh s. muknahallipatna Department of Electrical and Computer Engineering University of Wyoming Laramie USA

Increas.ng needs.for the s.udy of complex dynamical s.s.ems.require computing s.lutions.of a large number of ordinary and partial differential time-dependent equations.in near real-time. Numerical integration algorithms. which are computationally expens.ve and inherently s.quential, are typically us.d to compute s.lutions.of ordinary and partial differential time-dependent equations. This.pres.nts.challenges.to s.udy complex dynamical s.s.ems.in near real-time. This.paper examines.the challenges.of computing s.lutions.of ordinary differential time-dependent equations.us.ng the Parareal algorithm belonging to the clas. of parallel-in-time algorithms.on various.high-performance computing accelerator-bas.d architectures.and as.ociated programming models. The paper pres.nts.the code refactoring s.eps.and performance analys.s.of the Parareal algorithm on two accelerator computing architectures. the Intel Xeon Phi CPU and Graphics.Proces.ing Unit many-core architectures. and with OpenMP, OpenACC, and CUDA programming models. The s.eedup and s.aling performance analys.s.are us.d to demons.rate the s.itability of the Parareal to compute the s.lutions.of a s.ngle ordinary differential time-dependent equation and a family of interdependent ordinary differential time-dependent. The s.eedup, weak and s.rong s.aling res.lts.demons.rate the s.itability of Graphical Proces.ing Units.with the CUDA programming model as.the mos. efficient accelerator for computing s.lutions.of ordinary differential time-dependent equations.us.ng parallel-in-time algorithms. Cons.dering the time and effort required to refactor the code for execution on the accelerator architectures. the Graphical Proces.ing Units.with the OpenACC programming model is.the mos. efficient accelerator for computing s.lutions.of ordinary differential time-dependent equations.us.ng parallel-in-time algorithms.

关键词： Accelerators Many-Core Directive-Bas.d Time-Parallel s.aling s.eedup

来源：

维普期刊数据库评论

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Graphical Proces.ing Unit Bas.d Time-Parallel Numerical Method for Ordinary Differential Equations

引用

Journal of Computer and Communications 2020年第2期8卷 39-63页

作者： s.mathi Laks.miranganatha suresh s. muknahallipatna Department of Electrical and Computer Engineering University of Wyoming Laramie WY USA

On-line trans.ent s.ability analys.s.of a power grid is.crucial in determining whether the power grid will travers. to a s.eady s.ate s.able operating point after a dis.urbance. The trans.ent s.ability analys.s.involves.computing the s.lutions.of the algebraic equations.modeling the grid network and the ordinary differential equations.modeling the dynamics.of the electrical components.like s.nchronous.generators. exciters. governors. etc., of the grid in near real-time. In this.res.arch, we inves.igate the us. of time-parallel approach in particular the Parareal algorithm implementation on Graphical Proces.ing Unit us.ng Compute Unified Device Architecture to compute s.lutions.of ordinary differential equations. The numerical s.lution accuracy and computation time of the Parareal algorithm executing on the GPU are demons.rated on the s.ngle machine infinite bus.tes. s.s.em. Two types.of dynamic model of the s.ngle s.nchronous.generator namely the clas.ical and detailed models.are s.udied. The numerical s.lutions.of the ordinary differential equations.computed by the Parareal algorithm are compared to that computed us.ng the modified Euler’s.method demons.rating the accuracy of the Parareal algorithm executing on GPU. s.mulations.are performed with varying numerical integration time s.eps. and the s.itability of Parareal algorithm in computing near real-time s.lutions.of ordinary different equations.is.pres.nted. A s.eedup of 25× and 31× is.achieved with the Parareal algorithm for clas.ical and detailed dynamic models.of the s.nchronous.generator res.ectively compared to the s.quential modified Euler’s.method. The weak s.aling efficiency of the Parareal algorithm when required to s.lve a large number of ordinary differential equations.at each time s.ep due to the increas. in s.quential computations.and as.ociated memory trans.er latency between the CPU and GPU is.dis.us.ed.

关键词： Time-Parallel Differential Equation Numerical Integration Graphic Proces.ing Unit

来源：

维普期刊数据库评论

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

An Evaluation of Deep Learning Models.for Clas.ifying Time s.ries.Individual Data Ins.ances

引用

Journal of Computer and Communications 2024年第11期12卷 187-206页

作者： Jos.ua A. Blaney suresh s. muknahallipatna Department of Electrical Engineering and Computer s.ience University of Wyoming Laramie WY USA

Deep learning for time s.ries.s.quence individual data ins.ance clas.ification can revolutionize computer as.is.ed navigation by providing s.rgeons.with accurate, real-time ins.rument locality through automatic ins.rument localization. This.paper pres.nts.an evaluation of Deep Learning models.to perform individual data ins.ance clas.ification of time s.ries.data. The models.explored include convolution and recurrent networks. as.well as.s.ate-of-the-art res.dual and inception architectures. The time s.ries.data us.d to evaluate the models.cons.s.s.of depth and force meas.rements.from a drill. Four recurrent neural network models.us.ng long s.ort-term memory and gated recurrent units. known as.bas.line models. and four models.us.ng 1D convolution with Res.et and Inception architectures. known as.advanced models. were evaluated by determining the data ins.ance members.ip of the four clas.es. The four clas.es.repres.nt four dis.inct regions.in a bone travers.d by the drill bit during a s.rgical procedure. Firs., the time s.ries.data is.preproces.ed, identifying the four clas.es.or regions.of the bone. Next, the paper pres.nts.a dis.us.ion of the network architecture and modifications.of both the bas.c and advanced deep learning models. followed by the training proces. and hyperparameters.tuning. The performance of the models.was.evaluated us.ng the precis.on and recall performance parameters. Out of the eight models.evaluated, the recurrent neural network with gated recurrent units.has.the bes. performance. The paper als. demons.rates.the importance of the feature depth over the feature force in clas.ifying the data ins.ances. followed by the effects.of the imbalanced datas.t on the performance of the models.

关键词： Recurrent Neural Networks Long s.ort Term Memory Gated Recurrent Units Res.dual Networks Inception Networks

来源：

维普期刊数据库评论

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

A Comparis.n of PPO, TD3 and s.C Reinforcement Algorithms.for Quadruped Walking Gait Generation

引用

Journal of Intelligent Learning s.s.ems.and Applications 2023年第1期15卷 36-56页

作者： James.W. Mock suresh s. muknahallipatna Department of Electrical Engineering and Computer s.ience University of Wyoming Laramie Wyoming USA

Deep reinforcement learning (deep RL) has.the potential to replace clas.ic robotic controllers. s.ate-of-the-art Deep Reinforcement algorithms.s.ch as.Proximal Policy Optimization, Twin Delayed Deep Determinis.ic Policy Gradient and s.ft Actor-Critic Reinforcement Algorithms. to mention a few, have been inves.igated for training robots.to walk. However, conflicting performance res.lts.of thes. algorithms.have been reported in the literature. In this.work, we pres.nt the performance analys.s.of the above three s.ate-of-the-art Deep Reinforcement algorithms.for a cons.ant velocity walking tas. on a quadruped. The performance is.analyzed by s.mulating the walking tas. of a quadruped equipped with a range of s.ns.rs.pres.nt on a phys.cal quadruped robot. s.mulations.of the three algorithms.acros. a range of s.ns.r inputs.and with domain randomization are performed. The s.rengths.and weaknes.es.of each algorithm for the given tas. are dis.us.ed. We als. identify a s.t of s.ns.rs.that contribute to the bes. performance of each Deep Reinforcement algorithm.

关键词： Reinforcement Learning Machine Learning Markov Decis.on Proces. Domain Randomization

来源：

维普期刊数据库评论

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Real Time Thermal Image Bas.d Machine Learning Approach for Early Collis.on Avoidance s.s.em of s.owplows

引用

Journal of Intelligent Learning s.s.ems.and Applications 2024年第2期16卷 107-142页

作者： Fletcher Wads.orth suresh s. muknahallipatna Khaled Ks.ibati Department of Electrical Engineering and Computer s.ience University of Wyoming Laramie WY USA Department of Civil and Architectural Engineering University of Wyoming Laramie WY USA

In an effort to reduce vehicle collis.ons.with s.owplows.in poor weather conditions. this.paper details.the development of a real time thermal image bas.d machine learning approach to an early collis.on avoidance s.s.em for s.owplows. which intends.to detect and es.imate the dis.ance of trailing vehicles. Due to the operational conditions.of s.owplows. which include heavy-blowing s.ow, traditional optical s.ns.rs.like LiDAR and vis.ble s.ectrum cameras.have reduced effectivenes. in detecting objects.in s.ch environments. Thus. we propos. us.ng a thermal infrared camera as.the primary s.ns.r along with machine learning algorithms. Firs., we curate a large datas.t of thermal images.of vehicles.in heavy s.ow conditions. Us.ng the curated datas.t, two machine-learning models.bas.d on the modified Res.et architectures.were trained to detect and es.imate the trailing vehicle dis.ance us.ng real-time thermal images. The trained detection network was.capable of detecting trailing vehicles.99.0% of the time at 1500.0 ft dis.ance from the s.owplow. The trained trailing dis.ance network was.capable of es.imating dis.ance with an average es.imation error of 10.70 ft. The inference performance of the trained models.is.dis.us.ed, along with the interpretation of the performance.

关键词： Convolutional Neural Networks Res.dual Networks Object Detection Image Proces.ing Thermal Imaging

来源：

维普期刊数据库评论

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

s.m-to-Real: A Performance Comparis.n of PPO, TD3, and s.C Reinforcement Learning Algorithms.for Quadruped Walking Gait Generation

引用

Journal of Intelligent Learning s.s.ems.and Applications 2024年第2期16卷 23-43页

作者： James.W. Mock suresh s. muknahallipatna Department of Electrical Engineering and Computer s.ience University of Wyoming Laramie WY USA

The performance of the s.ate-of-the-art Deep Reinforcement algorithms.s.ch as.Proximal Policy Optimization, Twin Delayed Deep Determinis.ic Policy Gradient, and s.ft Actor-Critic for generating a quadruped walking gait in a virtual environment was.pres.nted in previous.res.arch work titled “A Comparis.n of PPO, TD3, and s.C Reinforcement Algorithms.for Quadruped Walking Gait Generation”. We demons.rated that the s.ft Actor-Critic Reinforcement algorithm had the bes. performance generating the walking gait for a quadruped in certain ins.ances.of s.ns.r configurations.in the virtual environment. In this.work, we pres.nt the performance analys.s.of the s.ate-of-the-art Deep Reinforcement algorithms.above for quadruped walking gait generation in a phys.cal environment. The performance is.determined in the phys.cal environment by trans.er learning augmented by real-time reinforcement learning for gait generation on a phys.cal quadruped. The performance is.analyzed on a quadruped equipped with a range of s.ns.rs.s.ch as.pos.tion tracking us.ng a s.ereo camera, contact s.ns.ng of each of the robot legs.through force res.s.ive s.ns.rs. and proprioceptive information of the robot body and legs.us.ng nine inertial meas.rement units. The performance comparis.n is.pres.nted us.ng the metrics.as.ociated with the walking gait: average forward velocity (m/s., average forward velocity variance, average lateral velocity (m/s., average lateral velocity variance, and quaternion root mean s.uare deviation. The s.rengths.and weaknes.es.of each algorithm for the given tas. on the phys.cal quadruped are dis.us.ed.

关键词： Reinforcement Learning Reality Gap Pos.tion Tracking Action s.aces Domain Randomization

来源：

维普期刊数据库评论

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Improving Accuracy and Computational Burden of Bundle Adjus.ment Algorithm Us.ng GPUs

引用

Engineering（科研） 2023年第10期15卷 663-690页

作者： Pranay R. Kommera suresh s. muknahallipatna John E. McInroy Department of Electrical Engineering and Computer s.ience University of Wyoming Laramie Wyoming USA

Bundle adjus.ment is.a camera and point refinement technique in a 3D s.ene recons.ruction pipeline. The camera parameters.and the 3D points.are refined by minimizing the difference between computed projection and obs.rved projection of the image points.formulated as.a non-linear leas.-s.uare problem. Levenberg-Marquardt method is.us.d to s.lve the non-linear leas.-s.uare problem. s.lving the non-linear leas.-s.uare problem is.computationally expens.ve, proportional to the number of cameras. points. and projections. In this.paper, we implement the Bundle Adjus.ment (BA) algorithm and analyze techniques.to improve algorithmic performance by reducing the mean s.uare error. We inves.igate us.ng an additional radial dis.ortion camera parameter in the BA algorithm and demons.rate better convergence of the mean s.uare error. We als. demons.rate the us. of explicitly computed analytical derivatives. In addition, we implement the BA algorithm on GPUs.us.ng the CUDA parallel programming model to reduce the computational time burden of the BA algorithm. CUDA s.reams. atomic operations. and cuBLAs.library in the CUDA programming model are propos.d, implemented, and demons.rated to improve the performance of the BA algorithm. Our implementation has.demons.rated better convergence of the BA algorithm and achieved a s.eedup of up to 16× on the us. of the BA algorithm on various.datas.ts.

关键词： Bundle Adjus.ment Levenberg-Marquardt s.ene Recons.ruction Radial Dis.tortion Coefficient Explicit Jacobian CUDA Optimization

来源：

维普期刊数据库评论

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Optimized CUDA Implementation to Improve the Performance of Bundle Adjus.ment Algorithm on GPUs

引用

Journal of s.ftware Engineering and Applications 2024年第4期17卷 172-201页

作者： Pranay R. Kommera suresh s. muknahallipatna John E. McInroy Department of Electrical Engineering and Computer s.ience University of Wyoming Laramie Wyoming USA

The 3D recons.ruction pipeline us.s.the Bundle Adjus.ment algorithm to refine the camera and point parameters. The Bundle Adjus.ment algorithm is.a compute-intens.ve algorithm, and many res.archers.have improved its.performance by implementing the algorithm on GPUs. In the previous.res.arch work, “Improving Accuracy and Computational Burden of Bundle Adjus.ment Algorithm us.ng GPUs.” the authors.demons.rated firs. the Bundle Adjus.ment algorithmic performance improvement by reducing the mean s.uare error us.ng an additional radial dis.orting parameter and explicitly computed analytical derivatives.and reducing the computational burden of the Bundle Adjus.ment algorithm us.ng GPUs. The naïve implementation of the CUDA code, a s.eedup of 10× for the larges. datas.t of 13,678 cameras. 4,455,747 points. and 28,975,571 projections.was.achieved. In this.paper, we pres.nt the optimization of the Bundle Adjus.ment algorithm CUDA code on GPUs.to achieve higher s.eedup. We propos. a new data memory layout for the parameters.in the Bundle Adjus.ment algorithm, res.lting in contiguous.memory acces.. We demons.rate that it improves.the memory throughput on the GPUs. thereby improving the overall performance. We als. demons.rate an increas. in the computational throughput of the algorithm by optimizing the CUDA kernels.to utilize the GPU res.urces.effectively. A comparative performance s.udy of explicitly computing an algorithm parameter vers.s.us.ng the Jacobians.ins.ead is.pres.nted. In the previous.work, the Bundle Adjus.ment algorithm failed to converge for certain datas.ts.due to s.veral block matrices.of the cameras.in the augmented normal equation, res.lting in rank-deficient matrices. In this.work, we identify the cameras.that caus. rank-deficient matrices.and preproces. the datas.ts.to ens.re the convergence of the BA algorithm. Our optimized CUDA implementation achieves.convergence of the Bundle Adjus.ment algorithm in around 22 s.conds.for the larges. datas.t compared

关键词： s.ene Recons.ruction Bundle Adjus.ment Levenberg-Marquardt Non-Linear Leas. s.uares Memory Throughput Computational Throughput Contiguous.Memory Acces. CUDA Optimization

来源：

维普期刊数据库评论

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Adaptive Region Cons.ruction for Efficient Us. of Radio Propagation Maps

引用

Journal of Computer and Communications 2017年第8期5卷 21-51页

作者： Vinay B. Ramakris.naiah suresh s. muknahallipatna Robert F. Kubichek Department of Electrical & Computer Engineering University of Wyoming Laramie WY USA

The efficient cons.ruction of contours.of a radio propagation map is.crucial in us.ng radio propagation maps.in a number of real-time communication and network applications. In this.res.arch work, we firs. propos. an adaptive region cons.ruction (ARC) technique capable of cons.ructing contours.of different res.lutions.of a radio propagation map. Next, the proces. of implementing the ARC technique for real-time execution on a GPU is.pres.nted. The drawbacks.of the firs. implementation us.ng only the global memory are dis.us.ed, and optimization techniques.to improve the performance are dis.us.ed and implemented. s.mulations.are performed with varying s.zes.of radio propagation maps. and the s.itability of the ARC technique for real-time operation is.pres.nted. A s.eedup of 25× is.achieved with the s.ared vers.on of the GPU compared to the s.quential CPU implementation. Als., the contour cons.ructed us.ng the ARC technique is.compared to that cons.ructed us.ng the convex hull approach demons.rating the higher accuracy of the contour from the ARC technique.

关键词： Radio Propagation s.mpling Parallel Adaptive Accelerator Contours

来源：

维普期刊数据库评论

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Nelder-Mead Bas.d Iterative Algorithm for Optimal Antenna Beam Patterns.in Ad Hoc Networks

引用

Journal of Computer and Communications 2017年第7期5卷 117-134页

作者： Vinay B. Ramakris.naiah Robert F. Kubichek suresh s. muknahallipatna Department of Electrical & Computer Engineering University of Wyoming Laramie WY USA

Directional antennas.s.ape trans.is.ion patterns.to provide greater coverage dis.ance and reduced coverage angle. Us. of adaptive directional antenna arrays.can minimize interference while als. being more energy efficient. When us.d in an ad-hoc network, this.reduces.interference among trans.itting nodes.and thereby increas.s.throughput. s.ch “s.art antennas. us. digital beamforming bas.d on s.gnal proces.ing algorithms.to compute the appropriate weights.to form effective antenna patterns. s.art antennas.require the knowledge of the s.gnal received at each antenna in the antenna array, thereby increas.ng the complexity of hardware and cos.. Als., conventional s.art antennas.optimize res.lts.for each individual node, while it is.preferable to have a global optimal s.lution. A problem that has.not been addres.ed is.how to compute individual beam patterns.that maximize s.me meas.re of global network performance. His.orically, the focus.has.been on finding node antenna patterns.that give locally optimal performance. In this.paper, we inves.igate a low hardware complexity beamforming approach aimed at improving global performance that us.s.average Nois.-to-s.gnal ratio as.the performance meas.re. Given a multi-hop route from s.urce to des.ination, beam patterns.are s.aped to maximize average s.gnal-to-nois. ratio acros. all nodes.on the route, which reduces.bit-error rates.and extends.battery and network lifetime. The antenna weights.are s.quentially adjus.ed acros. all nodes.in the route to achieve optimization acros. the network. By us.ng phas.-only weights. hardware cos.s.are minimized. The performance of the algorithm us.ng different path los. models.is.explored.

关键词： Phas.d Array Adaptive Antennas Beamforming Ad Hoc Networks Network-Wide Optimization Low Hardware Complexity Global Performance Improvement

来源：

维普期刊数据库评论

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

欢迎您,

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

请选择保存的检索档案：

请选择收藏分类：

通借通还

欢迎您,

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：