A. A. Ghorbani and V. C. Bhavsar, "Incremental Communication for Multilayer Neural Networks," IEEE Transactions on Neural Networks, Vol. 6, No. 6, pp. 1375-1385, 1995.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....connections. Although the test time is clearly shorter for FANN structures with fewer parameters, the test time as a stand alone index is still useful for comparing different implementations of FANN structures that have the same numbers of parameters. The test time can be evaluated based on [68]. More specifically, the necessary time in the forward pass of data through the network in the test phase can be expressed as t test = 2 [M eff t t (M eff H eff ) t 0 ] P: 7) where t t , t 0 , and P are the time required by a single add or multiply operation, the transfer time between two ....

A.A. Ghorbani and V.C. Bhavsar, "Incremental communication for multilayer neural networks, " IEEE Trans. on Neural Networks, vol. 6, no. 6, pp. 1375--1385, Nov. 1995.

....in the input(s) output of a node does not enforce instability. However, when the precision of the incremental values falls below a certain level, the network fails to converge. The analysis is suported by simulation studies of two learning problems. 1 Introduction We have earlier proposed[2] the incremental communication method for inter node communication in the Artificial Neural Networks(ANNs) The proposed communication method is aimed at reducing the communication complexity of ANNs by limiting the node s input output bandwidth requirements. In incremental intercommunication, ....

....We have found that for some problems even four bit precision for Deltay in fixed and or floating point representations is sufficient for the network to converge. With the 8 12 bit precisions almost the same results are obtained as that with the conventional communication using 32 bit precision [2]. Limited precision reduces the accuracy and may degrade the performance of an ANN learning algorithm; in other words, it may lead to deviations in the performance of the learning algorithm. Note that in the incremental communication method all the operations inside a node may be carried out using ....

A. A. Ghorbani and V. C. Bhavsar, "Incremental Communication for Multilayer Neural Networks," IEEE Transactions on Neural Networks, Vol. 6, No. 6, pp. 1375-1385, 1995.

....networks. Keywords Artificial Neural Networks, Incremental Communication, Error Analysis, Multilayer Perceptrons, Finite Precision Computation. I. INTRODUCTION We have earlier proposed the incremental inter node communication method for internode communication in Artificial Neural Networks(ANNs)[11]. In the incremental communication method, instead of communicating the whole value of a variable, only the increment or decrement to its previous value is sent on a communication link. The incremental communication, when implemented using limited precision for the incremental values that are ....

....to deviations in the performance of the learning algorithm compared to the full precision implementation of incremental communication. In some circumstances, the limited precision of incremental values may even cause smaller output error than that of the full precision (see the parity problem in [11]) This is due to the fact that the errors that are caused by the limited precision representation, the representational errors, can assume both positive and negative values and therefore, some of the representational errors may cancel each other. The price paid in using reduced precision for the ....

[Article contains additional citation context not shown here]

A. A. Ghorbani and V. C. Bhavsar, "Incremental Communication for Multilayer Neural Networks," IEEE Transactions on Neural Networks, Vol. 6, No. 6, pp. 1375-1385, 1995.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC