The Experts below are selected from a list of 213 Experts worldwide ranked by ideXlab platform
Narasimhan Sundararajan - One of the best experts on this subject based on the ideXlab platform.
-
parallel implementation of backpropagation neural networks on a heterogeneous array of Transputers
Systems Man and Cybernetics, 1997Co-Authors: Shou King Foo, Paramasivan Saratchandran, Narasimhan SundararajanAbstract:This paper analyzes parallel implementation of the backpropagation training algorithm on a heterogeneous transputer network (i.e., Transputers of different speed and memory) connected in a pipelined ring topology. Training-set parallelism is employed as the parallelizing paradigm for the backpropagation algorithm. It is shown through analysis that finding the optimal allocation of the training patterns amongst the processors to minimize the time for a training epoch is a mixed integer programming problem. Using mixed integer programming optimal pattern allocations for heterogeneous processor networks having a mixture of T805-20 (20 MHz) and T805-25 (25 MHz) Transputers are theoretically found for two benchmark problems. The time for an epoch corresponding to the optimal pattern allocations is then obtained experimentally for the benchmark problems from the T805-20, TS805-25 heterogeneous networks. A Monte Carlo simulation study is carried out to statistically verify the optimality of the epoch time obtained from the mixed integer programming based allocations. In this study pattern allocations are randomly generated and the corresponding time for an epoch is experimentally obtained from the heterogeneous network. The mean and standard deviation for the epoch times from the random allocations are then compared with the optimal epoch time. The results show the optimal epoch time to be always lower than the mean epoch times by more than three standard deviations (3/spl sigma/) for all the sample sizes used in the study thus giving validity to the theoretical analysis.
-
parallel implementations of backpropagation neural networks on Transputers a study of training set parallelism
1996Co-Authors: Paramasivan Saratchandran, Narasimhan Sundararajan, Shou King FooAbstract:Hardware and software aspects transputer topologies for parallel implementation comparison between serial and parallel implementation analysis and implementation for equal distribution of the training set in a homogeneous transputer array analysis and implementation for unequal distribution of the training set in a homogeneous transputer array analysis and implementation for unequal distribution of the training set in a heterogeneous transputer array conclusion.
-
Parallel implementation of backpropagation on Transputers
Proceedings of 1993 International Conference on Neural Networks (IJCNN-93-Nagoya Japan), 1Co-Authors: Shou King Foo, Paramasivan Saratchandran, Narasimhan SundararajanAbstract:Backpropagation algorithm is one of the most popular training algorithms for multilayer feedforward neural networks. However training the network with this algorithm has proved to be computationally intensive for a sequential machine. In this paper, parallel implementation of the backpropagation algorithm is investigated using Transputers hosted by a personal computer. Two methods of transputer implementations were considered. One method was the multi-tasking approach and the other the processor farming approach. Results showed that for all test cases, the training time for the neural network with the multi-tasking approach is shorter than the processor farming approach. Comparing with a serial 486-33 PC, it is found that as the problem size scales up, the improvement in training time from the parallel implementation becomes significant.
D P Obrien - One of the best experts on this subject based on the ideXlab platform.
-
a real time plasma boundary determination and display system using Transputers
Symposium On Fusion Technology, 1995Co-Authors: J J Ellis, E Van Der Goot, D P ObrienAbstract:A transputer based system has been developed to perform real time determination of the plasma boundary. The system calculates the plasma boundary in less than 2ms, using digitised data from a set of the magnetic pickup coils and saddle loops. The result of this analysis is passed on simultaneously to a display station and to a second transputer system for further analysis. The display station provides, on-line, an animated display of the plasma boundary cross-section during the pulse. Although this display only shows the results at a suitable rate for animation, all incoming data is stored. This makes it possible to replay all the data after the pulse for more detailed analysis (Fig. 1).
-
real time plasma boundary determination for display and control using Transputers
Symposium On Fusion Technology, 1993Co-Authors: E Van Der Goot, J J Ellis, D P ObrienAbstract:A transputer based system has been developed to demonstrate the feasibility of performing real time determination of the plasma boundary. This system has been used successfully towards the end of the last operational period of JET to produce, for the first time, an animated display of the plasma boundary cross-section. For this feasibility study the system was limited to producing the display of the plasma boundary only after the pulse had finished. To provide a real time, on-line calculation of the plasma boundary, a new transputer based system is being developed that will take its input directly from the magnetic pickup coils and perform the calculation in less than 2ms. The system will provide a real time on-line display and play-back facility. The boundary data is also suitable for use in plasma shape control, which plays an important role in optimising plasma performance, necessary for long pulse operation in future reactors. The implementation of the present system will be described in detail. Emphasis will be placed on the method used to calculate the plasma boundary position and the implementation of this method on a network of Transputers. The design and implementation of the software and the hardware for the new system will also be described.
R Harrop - One of the best experts on this subject based on the ideXlab platform.
-
use of Transputers in a 3 d positron emission tomograph
IEEE Transactions on Medical Imaging, 1991Co-Authors: M S Atkins, D Murray, R HarropAbstract:The use of a VME-bus-based transputer network as a parallel processing engine for positron volume imaging (PVI) is discussed. The authors find that the speedups of parallel networks depend on two major factors, the ratio of computation to communication for a task and the size of the task, and give a simple model to explore the limits on speedups. Through actual implementation it is shown that real-time PVI data acquisition can be achieved with about 20 transputer nodes, and it is estimated that three-dimensional (3-D) image reconstruction can be achieved within 10 min using 200 nodes. Larger images and a larger number of histograms can readily be accommodated using the same parallel algorithms, as the model presented places no limits on the size of the images. The versatility and scalability of Transputers makes them very suitable for use in PVI tomographs in that the same Transputers can be used for speeding up data acquisition, image reconstruction, and display. >
Shou King Foo - One of the best experts on this subject based on the ideXlab platform.
-
parallel implementation of backpropagation neural networks on a heterogeneous array of Transputers
Systems Man and Cybernetics, 1997Co-Authors: Shou King Foo, Paramasivan Saratchandran, Narasimhan SundararajanAbstract:This paper analyzes parallel implementation of the backpropagation training algorithm on a heterogeneous transputer network (i.e., Transputers of different speed and memory) connected in a pipelined ring topology. Training-set parallelism is employed as the parallelizing paradigm for the backpropagation algorithm. It is shown through analysis that finding the optimal allocation of the training patterns amongst the processors to minimize the time for a training epoch is a mixed integer programming problem. Using mixed integer programming optimal pattern allocations for heterogeneous processor networks having a mixture of T805-20 (20 MHz) and T805-25 (25 MHz) Transputers are theoretically found for two benchmark problems. The time for an epoch corresponding to the optimal pattern allocations is then obtained experimentally for the benchmark problems from the T805-20, TS805-25 heterogeneous networks. A Monte Carlo simulation study is carried out to statistically verify the optimality of the epoch time obtained from the mixed integer programming based allocations. In this study pattern allocations are randomly generated and the corresponding time for an epoch is experimentally obtained from the heterogeneous network. The mean and standard deviation for the epoch times from the random allocations are then compared with the optimal epoch time. The results show the optimal epoch time to be always lower than the mean epoch times by more than three standard deviations (3/spl sigma/) for all the sample sizes used in the study thus giving validity to the theoretical analysis.
-
parallel implementations of backpropagation neural networks on Transputers a study of training set parallelism
1996Co-Authors: Paramasivan Saratchandran, Narasimhan Sundararajan, Shou King FooAbstract:Hardware and software aspects transputer topologies for parallel implementation comparison between serial and parallel implementation analysis and implementation for equal distribution of the training set in a homogeneous transputer array analysis and implementation for unequal distribution of the training set in a homogeneous transputer array analysis and implementation for unequal distribution of the training set in a heterogeneous transputer array conclusion.
-
Parallel implementation of backpropagation on Transputers
Proceedings of 1993 International Conference on Neural Networks (IJCNN-93-Nagoya Japan), 1Co-Authors: Shou King Foo, Paramasivan Saratchandran, Narasimhan SundararajanAbstract:Backpropagation algorithm is one of the most popular training algorithms for multilayer feedforward neural networks. However training the network with this algorithm has proved to be computationally intensive for a sequential machine. In this paper, parallel implementation of the backpropagation algorithm is investigated using Transputers hosted by a personal computer. Two methods of transputer implementations were considered. One method was the multi-tasking approach and the other the processor farming approach. Results showed that for all test cases, the training time for the neural network with the multi-tasking approach is shorter than the processor farming approach. Comparing with a serial 486-33 PC, it is found that as the problem size scales up, the improvement in training time from the parallel implementation becomes significant.
M S Atkins - One of the best experts on this subject based on the ideXlab platform.
-
use of Transputers in a 3 d positron emission tomograph
IEEE Transactions on Medical Imaging, 1991Co-Authors: M S Atkins, D Murray, R HarropAbstract:The use of a VME-bus-based transputer network as a parallel processing engine for positron volume imaging (PVI) is discussed. The authors find that the speedups of parallel networks depend on two major factors, the ratio of computation to communication for a task and the size of the task, and give a simple model to explore the limits on speedups. Through actual implementation it is shown that real-time PVI data acquisition can be achieved with about 20 transputer nodes, and it is estimated that three-dimensional (3-D) image reconstruction can be achieved within 10 min using 200 nodes. Larger images and a larger number of histograms can readily be accommodated using the same parallel algorithms, as the model presented places no limits on the size of the images. The versatility and scalability of Transputers makes them very suitable for use in PVI tomographs in that the same Transputers can be used for speeding up data acquisition, image reconstruction, and display. >
