Investigates the proposed implementation of neural networks on massively parallel hierarchical computer systems with hypernet topology. The proposed mapping scheme takes advantage of the inherent structure of hypernet...
详细信息
Investigates the proposed implementation of neural networks on massively parallel hierarchical computer systems with hypernet topology. The proposed mapping scheme takes advantage of the inherent structure of hypernets to process multiple copies of the neural network in the different subnets, each executing a portion of the training set. Finally, the weight changes in all the subnets are accumulated to adjust the synaptic weights in all the copies. An expression is derived to estimate the time for all-to-all broadcasting, the principal mode of communication in implementing neural networks on parallel computers. This is later used to estimate the time required to execute various execution phases in the neural network algorithm, and thus to estimate the speedup performance of the hypernet in implementing neural networks.
In this paper an algorithm is described, which is based on the Real-Time Recurrent Learning (RTRL) algorithm by Williams and Zipser [1], for Multilayer Real-Time Recurrent Learning Networks (MLRN). The general gradien...
详细信息
A number of fully and partially recurrent networks have been proposed to deal with temporally extended tasks. However, it is not yet clear which algorithms and network architectures are best suited to certain kinds of...
详细信息
A number of fully and partially recurrent networks have been proposed to deal with temporally extended tasks. However, it is not yet clear which algorithms and network architectures are best suited to certain kinds of problems. In this paper we report on experimental investigations of a quantitative nature, which address this particular need by comparing fully recurrent networks using learning algorithms such as backpropagation-through-time (BPTT), batch BPTT Quickprop-through-time, and real-time recurrent learning with Elman and Jordan partially recurrent networks on four benchmark problems: detection of three consecutive zeros, nonlinear plant identification, Turing machine emulation, and real-world distillation column modelling.
This paper presents a strategy for efficiently rendering time-varying volume data on a distributed-memory parallel computer. Visualizing time-varying volume data take both large storage space and long computation time...
详细信息
ISBN:
(纸本)0818682596
This paper presents a strategy for efficiently rendering time-varying volume data on a distributed-memory parallel computer. Visualizing time-varying volume data take both large storage space and long computation time. Instead of employing all processors to render one volume at a time, a pipelined rendering approach partitions processors into groups so that multiple volumes can be rendered concurrently. The overall rendering time is greatly minimized because rendering is overlapped with I/O required to load the volume data sets. Moreover, parallelization overhead may be reduced as a result of partitioning the processors. We modify an existing parallel volume renderer to exploit various levels of rendering parallelism and to study how the partitioning of processors may lead to optimal rendering performance. We find that two factors affecting the overall execution time are resource utilization efficiency and pipeline startup latency. The optimal partitioning configuration is the one that balances these two factors. Tests on Intel Paragon computers show that in general optimal partitionings do exist for a given rendering task and result in 40-50% saving in overall rendering time.
In this paper, we examine the effectiveness of absorbing layers as non-reflecting computational boundaries for the Euler equations. The absorbing-layer equations are simply obtained by splitting the governing equation...
详细信息
As image coders evolve from DCT-based to wavelet-based designs, the latter must be enhanced to include capabilities currently supported by standards such as JPEG. Said and Pearlman (see IEEE Transactions on Circuits a...
详细信息
As image coders evolve from DCT-based to wavelet-based designs, the latter must be enhanced to include capabilities currently supported by standards such as JPEG. Said and Pearlman (see IEEE Transactions on Circuits and Systems for Video Tech., vol.6, p.243-50, 1996) and Schwartz, Zandi and Boliek (see Proc. of SPIE, vol.2564, July 1995) have described approaches for incorporating progressive transmission capabilities within wavelet-based coders. All of these coders apply scalar quantization to wavelet transform coefficients and then apply sophisticated entropy coding methods to the quantized coefficients. In this paper, we present coding techniques that enable progressive transmission when trellis coded quantization (TCQ) is applied to the wavelet coefficients. While the trellis coded quantizer is more complex than the uniform scalar quantizer, comparable PSNR performance is achieved with a simple entropy coder. In addition, our use of sophisticated quantization and bit rate allocation algorithms enables the development of coders that are tuned for improved perceptual image quality.
作者：
McCormick, MEBhattacharyya, RMouring, SEDr. Michael E. McCormick:is a research professor of civil engineering at The Johns Hopkins University. Before joining the Hopkins faculty in 1994
he was a professor of ocean engineering for twenty-five years at the U.S. Naval Academy. In addition he has held full-time faculty positions at Swarthmore College Trinity College (Hartford) and the Catholic University of America. He was also a hydrodynamicist at the David Taylor Model Basin for more than four years. Prof. McCormick received his undergraduate degree in mathematics and physics from AmericanUniversity a masters degree in applied mechanics and a Ph.D. in mechanical engineering from Catholic University a Ph.D. in civil engineering and a Sc.D. in engineering science from Trinity College in Dublin Ireland. He has over 100 publications including two books in the areas of ocean engineering wave mechanics and ocean wave energy conversion. He has also edited two books dealing with ocean engineering. In addition he is co-editor of both the journal Ocean Engineering and the Elsevier book series in ocean engineering. Dr. Rameswar Bhattacharyya:is professor of naval architecture at the U.S. Naval Academy
where he has served for twenty-six years and adjunct professor of mechanical engineering at The Johns Hopkins University. Prior to joining the Naval Academy faculty he was a faculty member in the Department of Naval Architecture and Marine Engineering at the University of Michigan. His research experience includes ten years at both the Lubecker Flender-Werke and the Hamburg Ship Model Basin in Germany. His research has led to numerous publications including two books one in the area of ship dynamics and the other in the area of computer-aided ship design. Prof. Bhattacharyya received his undergraduate degree in naval architecture from the Indian Institute of Technology and his doctorate in engineering from the Technical University of Hanover Germany. In addition he holds an honorary doctorate from the University of Veracruz. With Prof. McCormick he co
Panels and all other structural components of surface ships and submarines vibrate when the vessel is underway. The vibratory motions are primarily excited by the power plant. At operational (design) speeds, panels vi...
详细信息
Panels and all other structural components of surface ships and submarines vibrate when the vessel is underway. The vibratory motions are primarily excited by the power plant. At operational (design) speeds, panels vibrate in their fundamental modes and those associated with their higher harmonic frequencies. The panel motions have rather well-defined energy spectra, which depend on both the structural design, position of the panel and the rotational speed of the single or multiple power plants. The panel motions will interact with the vortices in the adjacent turbulent boundary layer. The interaction can result in either an increase in the frictional drag or a decrease. Because of this, the argument is made that the designs of the panels and their support systems should include considerations of this hydroelastic effect.
Parallel direct execution simulation is an important tool for performance and scalability analysis of large message passing parallel programs executing on top of a virtual computer. However, detailed simulation of mes...
Compressible mixing layers are analyzed using a dilatational covariances model based on a pseudo-sound constitutive relation. The calculations are used to evaluate the different physical phenomena affecting compressib...
详细信息
暂无评论