#Sankar, A., and R.J. Mammone, Growing and Pruning Neural Tree Networks, IEEE Trans. Comput. 42(3) 291-299 (1993).  Home/Search   Document Not in Database   Summary   Related Articles   Check

....trees is presented. The neural trees are hybrid concepts between the decision trees and the neural networks characterized by neural networks instead of decision nodes in the inner nodes of the tree structure; many different models of neural trees have been developed in the past years (see e.g. [7,8,9]. The neural tree adopted here (for a more detailed description see [10] is supervised and is characterized by a tree structure with perceptrons in the internal nodes; at the root, the perceptron tries to subdivide the training set into groups corresponding to the K classes of the problem, until ....

A. Sankar and R.J. Mammone, "Growing and pruning neural tree networks," IEEE Transaction on Computers, Vol. 42, No. 3, pp. 291-299, 1993

.... on line learning [17,21,22,28,31,35,36,42,44,61,66,67,69] constructivist structural learning [15,19,11,14,9] that is supported by biological facts [14,62,73,77,82] selectivist structural learning [26,29,49,56,59,64,50,32] hybrid constructivist selectivist structural learning December,2001 3 [52,66,70,31]; knowledge based learning neural networks (KBNN) 57,24,25,30,33,38,44, 45,51,63,76,77,83] The EFuNN model presented in the paper has elements from all the groups above. The model is called evolving because of the nature of the structural growth and structural adaptation of the whole evolving ....

Sankar, A., and R.J. Mammone, "Growing and pruning neural tree networks", IEEE Trans. Comput. 42(3) 291-299 (1993).

....that is adequate to the expected accuracy of the system. Reducing the structure of a KBNN can be achieved through regular pruning of nodes and connections thus allowing for knowledge to emerge in the structure, or through aggregating nodes into bigger rule clusters. The former approach is used in [19,21,38,39,41,44,46,47,53]. The latter one is explored in this paper. It is based on a regular aggregation of rule nodes in the KBNN structure, which is equivalent to aggregating rules into rule clusters before new data and new knowledge is accommodated in the system. This is the case with the EFuNNs. Different KBNNs are ....

A. Sankar, R.J. Mammone, Growing and pruning neural tree networks, IEEE Trans. Comput. 42 (3) (1993) 291-299.

.... learning [69,35,36,82] on line [17,21,22,28,31,35,36,42,44,61,66,67,69] constructivist structural [ 15,19,11,14,9] that is supported by biological facts [ 14,62,73,77, 82] selectivist structural learning [26,29,49,56,59,64,50,32] hybrid constructivist selectivist structural learning 2 [52,66,70,31]; knowledge based learning neural networks (KBNN) 57,24,25,30,33,38,44, 45,51,63,76,77,83] The EFuNN model presented in the paper has elements from all the groups above. The model is called evolving because of the nature of the structural growth and structural adaptation of the whole evolving ....

Sankar, A., and R.J. Mammone, "Growing and pruning neural tree networks", IEEE Trans. Comput. 42(3) 291-299 (1993).

....that is adequate to the expected accuracy of the system. Reducing the structure of a KBNN can be achieved through regular pruning of nodes and connections thus allowing for knowledge to emerge in the structure, or through aggregating nodes into bigger rule clusters. The former approach is used in [19,21,38,39,41,44,46,47,53]. The latter one is explored in this paper. It is based on a regular aggregation of rule nodes in the KBNN structure, which is equivalent to aggregating rules into rule clusters before new data and new knowledge is accommodated in the system. This is the case with the EFuNNs. Di erent KBNNs are ....

A. Sankar, R.J. Mammone, Growing and pruning neural tree networks, IEEE Trans. Comput. 42 (3) (1993) 291}299.

....2 Delta Delta Delta x i K that is, an expansion up to order K of an input vector of length N . The task of a selfstructuring algorithm is to determine which of these terms is important to solve the problem at hand. Most conventional self structuring algorithms are based on growing and pruning [87]. A growing algorithm is a method for adding terms to a small network [25] whilst a pruning algorithm removes terms from a large one [63] In many cases considerable computational effort is required to determine the change to be made to the network, and very often suboptimal results are obtained. ....

A. Sankar and R. J. Mammone. Growing and Pruning Neural Tree Networks. IEEE Trans. Computers, 42(3):291--299, 1993.

....greatly. The experiment is conducted using a database containing 32 classes of real logos, which are used in [12] A Baird noise simulation program with the parameters in Table 3.4 is used to generate learning and test images. The experiments are organized to construct a Neural Tree (NT) [73, 74, 75] in order to assess the performances of recursive neural networks. The neural tree is a special decision tree and has been recently proposed for solving classification tasks. As shown in Figure 3.10, the data is splitted in each vertex and the classification of a pattern, according to the ....

A. Sankar and R. J. Mammone, "Growing and pruning neural tree networks," IEEE Transactions on Computers, vol. 42, pp. 291--299, 1993.

....i 2 Delta Delta Delta x i k that is, an expansion up to order k of an input vector of length n. The task of a self structuring algorithm is to determine which of these terms is important to solve the problem at hand. Most conventional self structuring algorithms are based on growing and pruning [7]. A growing algorithm is a method for adding terms to a small network [8] whilst a pruning algorithm removes terms from a large one [9] In many cases considerable computational effort is required to determine the change to be made to the network, and very often sub optimal results are obtained. ....

A. Sankar and R. J. Mammone. Growing and Pruning Neural Tree Networks. IEEE Trans. Computers, 42(3):291--299, 1993.

....decomposed into a sequence of simpler local classification rules. Decision trees can be straightforwardly combined with neural networks since neural network modules can be used to make the decisions at the branch nodes in the tree. For this reason a number of neural tree models have been proposed [121, 132, 19, 128, 36, 119, 127, 87]. For neural computing the neural tree architecture has the advantage that it can be readily implemented in hardware because of its recursive nature. At a branch tree node in a classification tree the outcome of the decision causes the loading of one of two sets of weights (Figure 20) This ....

....using a single single layered neural network. At a branch tree node the output of the network determines which of two sets of weights to load from the register. The output of a leaf tree node is the output from the entire neural tree. A straightfoward way of generating a neural tree is as follows [121]: A8. Construction of a Neural Decision Tree 1. Attempt to separate the pattern set using the pocket al..gorithm, etc. The solution will divide the data into two sets S 1 and S 0 according to the outputs 1 and Gamma1 respectively. 2. For the pattern set S 1 check if all the patterns correctly ....

[Article contains additional citation context not shown here]

A. Sankar and R.J. Mammone. Growing and pruning neural tree networks. IEEE Transactions on Computers, 42:291--299, 1993.

....search methods for the CNeT are introduced and used for training and recall. The influence of different search methods on the performance of the CNeT is experimentally evaluated. 1 Introduction Neural tree architectures were recently introduced for pattern classification [1] 2] 4] [6]. The main idea is to combine advantages of neural networks and decision trees. Neural tree architectures are decision trees with a neural network in each node. The existing neural tree architectures can be grouped according to the learning paradigm employed for the training of the neural ....

....and decision trees. Neural tree architectures are decision trees with a neural network in each node. The existing neural tree architectures can be grouped according to the learning paradigm employed for the training of the neural networks. Most architectures use supervised learning [2] 5] [6]. The nodes usually contain a feed forward network trained using gradient descent. Other approaches use unsupervised learning [1] 4] The nodes of such trees contain competitive networks or feature maps. Another criterion to classify neural tree architectures is the type of decision that is made ....

[Article contains additional citation context not shown here]

A. Sankar, R. J. Mammone, "Growing and Pruning Neural Tree Networks," IEEE Transactions on Computers, vol. 42, no. 3, pp. 291--299, 1993.

.... of the on line learning methods deal with a fixed structure of the NN [1, 14] other exploit dynamically changing structure of NN through either a structural growth (constructivism) 8, 6, 12] or a structural pruning (selectivism) 45, 13, 16, 21, 36, 37, 42, 53] or both growing and shrinking [2, 24, 49, 39]. The paper further develops and explores spatial temporal adaptation in evolving fuzzy neural networks (EFuNNs) for the task of on line adaptive speech recognition. This task is explained and illustrated in section 2. The major principles of spatial temporal EFuNNs are presented in section 3. ....

A Sankar and R.J. Mammone. Growing and pruning neural tree networks. IEEE Trans Comput., 42(3):291--299, 1993.

....recently considered hybrid structures between decision trees and neural nets. Though these techniques were developed as neural networks whose structure could be automatically determined, 19 their outcome can be interpreted as decision trees with nonlinear splits. Examples of this work include [173, 448, 46, 87, 207, 425, 102]. Techniques very similar to those used in tree construction, such as information theoretic splitting criteria and pruning, can be found in neural tree construction also. In addition to these methods, there exist other hybrid techniques between decision trees and neural networks. Sethi [435] ....

Anant Sankar and Richard J. Mammone. Growing and pruning neural tree networks. IEEE Transactions on Computers, 42(3):291--299, March 1993.

....the logistic effectively implements a threshold activation. Contexts in which this simplification is used include network visualization [ Munro, 1992 ] skeletonization [ Ramachandran and Pratt, 1992 ] unsupervised learning [ Schraudolph and Sejnowski, 1993 ] hybrid learning architectures [ Sankar and Mammone, 1993 ] and transfer between networks for related tasks [ Pratt, 1993, Sharkey and Sharkey, 1993 ] In this simplification, a threshold is considered to occur at the point where the logistic crosses 0:5. This simplification is often justified by noting that network weights tend to grow with ....

A. Sankar and R.J. Mammone. Growing and Pruning Neural Tree Networks. IEEE Transactions on Computers, 42(3):291--299, March 1993.

....logistic effectively implements a threshold activation. Contexts in which this simplification has been used include network visualization [ Munro, 1992 ] skeletonization [ Ramachandran and Pratt, 1992 ] unsupervised learning [ Schraudolph and Sejnowski, 1993 ] hybrid learning architectures [ Sankar and Mammone, 1993 ] and transfer, as described here. Justifications for this simplification were discussed in Section 3.1.3. We call the assumption that input to hidden (IH) layer hyperplane positions largely follow the decision boundary the hyperplane assumption. This assumption involves assuming (1) that the ....

A. Sankar and R.J. Mammone. Growing and Pruning Neural Tree Networks. IEEE Transactions on Computers, 42(3):291--299, March 1993.

....neural network community have recently considered hybrid structures between decision trees and neural networks. Although these techniques were developed as neural networks whose structure could be automatically determined, their outcome can be interpreted as decision trees with nonlinear splits [10, 21, 22, 25, 29, 43, 49, 54]. In most of these hybrid structures, a small neural network at each node of the tree classifier is used to implement nonlinear and multivariate splits instead of a hyperplane in decision tree approaches. Accordingly, the neural tree is automatically generated by training the neural network at ....

....instead of searching for the best hyperplane with some distortion metric in decision tree approaches. Basically, such neural trees follow the principle underlying decision trees, i.e. non overlapping ( hard ) split in each nonterminal node and only one class label associated with each leaf node [22, 43, 49, 54]. For generalization, a pruning procedure is also needed in these neural trees as same as in decision tree approaches such as Classification And Regression Tree (CART) 3] The other hybrid techniques between decision trees and neural networks use elaborate methods for converting a decision tree ....

A. Sankar and R. J. Mammone. Growing and pruning neural tree networks. IEEE Transactions on Computer, 42(3):291--299, 1993.

....automatically determined, their outcome can be interpreted as decision trees with nonlinear splits. Techniques very similar to those used in tree construction, such as information theoretic splitting criteria and pruning, can be found in neural tree construction also. Examples of this work include [127, 342, 32, 59, 150, 324, 72]. Sethi [331] described a method for converting a univariate decision tree into a neural net and then retraining it, resulting in tree structured entropy nets with sigmoidal splits. An extension of entropy nets, that converts linear decision trees into neural nets was described in [288] Decision ....

Anant Sankar and Richard J. Mammone. Growing and pruning neural tree networks. IEEE Trans. on Comp., 42(3):291--299, March 1993. 46 SREERAMA K. MURTHY

....Perceptron Trees [13] etc. We will call them Perceptron Decision Trees (PDTs) as they can be regarded as binary trees having a simple perceptron associated to each decision node. Different algorithms for Top Down induction of PDTs from data have been proposed, based on different principles, [10], 5] 8] Experimental study of learning by means of PDTs indicates that their performances are sometimes better than those of traditional decision trees in terms of generalization error, and usually much better in terms of tree size [8] 6] but on some data set PDTs can be outperformed by ....

Sankar A., Mammone R.J., Growing and Pruning Neural Tree Networks, IEEE Transactions on Computers, 42:291-299, 1993.

No context found.

#Sankar, A., and R.J. Mammone, Growing and Pruning Neural Tree Networks, IEEE Trans. Comput. 42(3) 291-299 (1993).

No context found.

Sankar, A., and R.J. Mammone, "Growing and Pruning Neural Tree Networks", IEEE Trans. Comput. 42(3) 291-299 (1993).

No context found.

A. Sankar and R. J. Mammone, "Growing and pruning neural tree networks, " IEEE Trans. Neural Networks, vol. 42, no. 3, pp. 291--299, 1993.

No context found.

Anant Sankar and Richard J. Mammone. Growing and pruning neural tree networks. IEEE Trans. on Comp., 42#3#:291#299, March 1993. 46 SREERAMA K. MURTHY

No context found.

A. Sankar, R. J. Mammone, "Growing and Pruning Neural Tree Networks," IEEE Transactions on Computers, vol. 42, no. 3, pp. 291--299, 1993. This article was processed using the L A T E X macro package with LLNCS style

No context found.

Sankar, A., & Mammone, R.J. (1993) Growing and Pruning Neural Tree Networks, IEEE Transactions on Computers, 42:291-299.

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