A. Roy, S. Govil, and R. Miranda, "An algorithm to generate radial basis function (RBF)-like nets for classification problems," Neural Networks, vol. 8, pp. 179--201, 1995.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....to these parameters. 1) Experimental Results: Tables I and II show our experimental results for the three problems over 30 runs. The error rate in the table refers to the percentage of wrong classifications produced by the EANN s. In comparison with the error rates obtained by other methods [32] [34], EPNet s results are quite competitive. The size of EANN s, measured by the number of hidden nodes and connections, is very small. This will be useful in ANN s execution phase and for hardware implementation. Table III compares EPNet s results with those produced by a hand designed BP network ....

A. Roy, S. Govil, and R. Miranda, "An algorithm to generate radial basis function (RBF)-like nets for classification problems," Neural Networks, vol. 8, pp. 179--201, 1995.

....0.2135 (21.35 ) while EPNet achieved the testing error rate of 0.1927 (19.27 ) The largest ANN evolved by EPNet among 30 runs had only six hidden nodes. The average was 3.4. The heart disease problem: Table X shows results from EPNet and other neural and nonneural algorithms. The GM algorithm [55] is used to construct RBF networks. It produced a RBF network of 24 Gaussians with 18.18 testing error. Bennet and Mangasarian [56] reported a testing error rate of 16.53 with their MSM1 method, 25.92 with their MSM method, and about 25 with BP, which is much worse than the worst ANN evolved ....

A. Roy, S. Govil, and R. Miranda, "An algorithm to generate radial basis function (RBF)-like nets for classification problems," Neural Networks, vol. 8, pp. 179--201, 1995.

....has polynomial time complexity is discussed. The algorithm uses binary data during training and overcomes the local minima problem that backpropagation experiences. 1 Introduction It is desirable for a neural network to be able to determine it s own network size and topology during training [2] [7]. The algorithm presented here determines which LTGs are performing what tasks in the learning problem at hand. The method proposed in this paper is much less complex than the method proposed by other algorithms which determine their own topology. The algorithm trains the network in a single pass ....

....all LTGs in the first layer are connected to those in the next layer, which are in turn connected to the subsequent layer. Whether to reuse LTGs is a philosophical one. In backpropagation it is desirable to produce the smallest network possible to reduce the moving target problem [2] Roy et al. [7] argue that the algorithms should be more brain like and I hypothesis that the brain is not a minimum network based on the grandmother cell problem [1] Though for software or hardware implementation, reusability may be desirable. While this algorithm overcomes the complexity problems in [3] it ....

Roy, A., Govil, S., Miranda, R. (1995). "An Algorithm to Generate Radial Basis Functions (RBF)-Like Nets for Classification Problems". Neural Networks, Vol. 8, No. 2,

....of 0.2135 (21.35 ) while EPNet achieved the testing error rate of 0.1927 (19.27 ) The largest ANN evolved by EPNet among 30 runs had only 6 hidden nodes. The average was 3.4. The Heart Disease Problem Table X shows results from EPNet and other neural and non neural algorithms. The GM algorithm [55] is used to construct RBF networks. It produced a RBF network of 24 Gaussians with 18.18 testing error. Bennet and Mangasarian [56] reported a testing error rate of 16.53 with their MSM1 method, 25.92 with their MSM method, and about 25 with BP, which is much worse than the worst ANN evolved ....

A. Roy, S. Govil, and R. Miranda, "An algorithm to generate radial basis function (RBF)-like nets for classification problems," Neural Networks, vol. 8, pp. 179--201, 1995.

....parameters. C.1 Experimental Results Tables I and II show our experimental results for the three problems over 30 runs. The error rate in the table refers to the percentage of wrong classifications produced by the EANNs. In comparison with the error rates obtained by other methods [32] 33] [34], EPNet s results are quite competitive. The size of EANNs, measured by the number of hidden nodes and connections, is very small. This will be useful in ANN s execution phase and for hardware implementation. Table III compares EPNet s results with those produced by a hand designed BP network (the ....

A. Roy, S. Govil, and R. Miranda, "An algorithm to generate radial basis function (RBF)-like nets for classification problems," Neural Networks, vol. 8, pp. 179--201, 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