D. Rumelhart, B. Widrow, and M. Lehr. The basic ideas in neural networks. Communications of the ACM, 37(3), 1994.  Home/Search   Document Not in Database   Summary   ACM   TOC   Related Articles   Check

....important issue with the present dairy problem. NNs have the ability to approximate non linear relationships between sets of inputs and their corresponding sets of outputs [13] The dairy data could be expected to display some degree of non linearity. The ability of NNs to generalise well [18] and to learn concepts involving real valued features [5] are potential advantages with this project, since the predicted daughter responses are continuous variables. However an attempt has been made to categorise this data into discrete classes and analyse it using symbolic learning paradigms ....

D.E. Rumelhart, B. Widrow, and M.A. Lehr. The basic ideas in neural networks. Communications of the ACM, 37:87--92, 1994.

....kind of a learning algorithm can be used in IML. They invoke the fundamental question of which machine learning algorithm fits all of these criteria. We discuss several options and the reason why they are not viable before we settle on our algorithm of choice: decision trees (DT) Neural Networks [12] are a powerful and often used machine learning algorithm. They can provably approximate any function in two layers. Their strength lies in their abilities to intelligently integrate a variety of features. Neural networks also produce relatively small and efficient classifiers, however, there are ....

Rumelhart, D., Widrow, B., and Lehr, M. "The Basic Ideas in Neural Networks." Communications of the ACM, 37(3), (1994), pp 87-92.

....design of ANN architectures [3, 4] has been found to be a valuable alternative and usually improves the overall performance of an ANN. Many researchers also observed that networks of low complexity, i.e. few hidden neurons and or sparse connectivity, improve the generalization ability of ANNs [5]. This property is of potential benefit to RE algorithms, specifically, for RE techniques analyzing the ANN structure (decompositional approach) as the extracted rule bases become smaller. The more compact set of rules then possibly comprises more general concepts with increased comprehensibility ....

David E. Rumelhart, Bernard Widrow, and Michael A. Lehr. The Basic Ideas in Neural Networks. Communications of the ACM, 37(3):87--92, March 1994.

....(r) is equal to the dimensionality of the reduced feature space. In the output layer, the number of output units (m) is equal to the number of pre defined categories in the particular text categorization task. The number of hidden units in the neural network affects the generalization performance [25, 48]. The choice depends on the size of the training set and the complexity of the classification task the network is trying to learn, and can be found empirically based on the categorization performance (see section 6.3.4) For classification of the documents, reduced feature vectors representing ....

D. E. Rumelhart, B. Widrow, and M. A. Lehr, "The basic ideas in neural networks," Communications of the ACM, vol. 37, no. 3, pp. 87--92, 1994.

....learned model. A model with higher generalization accuracy is normally preferred. There are several machine learning approaches for computing classification models, for example, decision tree learning [ Quinlan, 1986 ] rule induction [ Clark and Niblett, 1989; Cohen, 1995 ] neural networks [ Rumelhart et al. 1994 ] Bayesian learning [ Cheeseman and Stutz, 1996 ] etc. Each approach uses a di#erent model representation, e.g. a decision tree or a set of rules, etc. and a di#erent search strategy and heuristic for traversing the space of possible models. Search Heuristic Information gain is the most ....

D. Rumelhart, B. Widrow, and M. Lehr. The basic ideas in neural networks. Communications of the ACM, 37(3):87--92, 1994.

....output has to be o hj 2 [0; 1] We use f(a) 1 1 exp( Gammaa) Such a feedforward multilayer perceptron can approximate any function after a suitable amount of training. Therefor known discrete values of this function are presented to the net. The net is expected to learn the function rule [2]. The behaviour of the net is changed by modification of the weights and bias values. The back propagation learning algorithms we use to optimize these values is described later together with its parallelizations. 2.2 Preprocessing the input data An efficient preprocessing of the data is ....

D.E. Rumelhart, B. Widrow, M.A. Lehr. The Basic Ideas in Neural Networks. Communications of the ACM Vol.37, No.3, pp. 87-92, March 1994.

....learning of the training data set, will result in a loss of generalization capabilities, a problem known as overfitting. It has been found that ANNs with lower complexity, i.e. small number of neurons, low connectivity, show a better generalization performance than more complex networks [RWL94] The space of possible network topologies is enormous, e.g. there are 2 n(n Gamma1) 2 ways to connect n neurons (all partial connections included) which increases even more with various learning method parameters. Specifically, as we restrict our ANN topologies to Feed Forward Networks, we ....

David E. Rumelhart, Bernard Widrow, and Michael A. Lehr. The basic ideas in neural networks. Communications of the ACM, 37(3):87--92, March 1994.

....ability. ANN architecture or complexity is determined by the number of neurons, number of connections, and their connectivity. It has been found that ANNs with lower complexity, i.e. small number of neurons, low connectivity, show a better generalization performance than more complex networks [RWL94] which tend to lose generalization ability, i.e. overfitting. Various methods have been suggested for the automatic construction of ANN topologies. Among these are Network Growing (e.g. Cascade Correlation [FL90] Network Pruning (e.g. OBS [HSW93] OBD [CDS90] and Evolutionary Design [Yao93] ....

David E. Rumelhart, Bernard Widrow, and Michael A. Lehr. The basic ideas in neural networks. Communications of the ACM, 37(3):87--92, March 1994.

....bias value, and x j is the output value of unit j. The output value of a unit is normally given by a non linear function (the activation function) of its activation value. A typical choice of such a non linear function, also called a squashing function, is the sigmoid. The reader is referred to [Rume94] for an introduction. Specific units of a neural network are used as global inputs and as global outputs. Input units receive their input values from the outside world with no weight associated and output units deliver their output values to the outside world. As a whole, a neural network may be ....

Rumelhart, D., et al.: The Basic Ideas in Neural Networks. Communications of the ACM, Vol. 37, No. 3, March 1994.

....suit neural network learning. 4 Neural Network Approaches To make our comparison more complete, we will now describe lifelong approaches that rely exclusively on artificial neural network representations. Neural networks have been applied successfully to a variety of real world learning problems [47, 43, 49]. 4.1 Back Propagation Probably the most common way to learn a function f n : d Gamma f0; 1g with an artificial neural network is to approximate it using the Back Propagation algorithm (or a variation thereof) The network that approximates f n might have d input units, one for each of ....

Rumelhart, D. E., Widrow, B., and Lehr, M. A. The basic Ideas in Neural Networks. Communications of the ACM, vol. 37 (1994), pp. 87--92.

....is a hybrid learning mechanism, which integrates inductive and analytical learning. Before explaining EBNN, let us briefly consider its components: inductive) neural network learning and (analytical) explanation based learning. 2. 1 Neural Network Backpropagation Artificial neural networks (see [10, 30, 40] for an introduction) consist of a set of simple, densely interconnected processing units. These units transform signals in a non linear S. Thrun An Approach to Learning Mobile Robot Navigation 4 (a) Training examples is has made of upward open flat is light handle Styrofoam concave color ....

David E. Rumelhart, Bernard Widrow, and Michael A. Lehr. The basic ideas in neural networks. Communications of the ACM, 37(3):87--92, March 1994.

....elements each computing a simple mathematical function, but highly inter connected one to another, can exhibit a high computational power. To the processing units is attributed the name of neurons and to the whole structure of interconnected elements is given the name of neural networks [27, 29, 33, 44, 66, 68]. Neural networks have been used with success in many diverse areas of scientific and technical disciplines including computer science, engineering, physics, medicine, cognitive science, neurophysiology and human perception. Many models have been investigated from many different points of view ....

Rumelhart, D., Widrow, B., Lehr, M. (1994). The basic ideas in neural networks. Communications of the ACM, March 1994, Vol.37, No.3.

....(r) is equal to the dimensionality of the reduced feature space. In the output layer, the number of output units (m) is equal to the number of pre defined categories in the particular text categorization task. The number of hidden units in the neural network affects the generalization performance [4, 9]. The choice depends on the size of the training set and the complexity of the classification task the network is trying to learn, and can be found empirically based on the categorization performance. For classification of the documents, reduced feature vectors representing the documents are fed ....

D. E. Rumelhart, B. Widrow, and M. A. Lehr. The basic ideas in neural networks. Communications of the ACM, 37(3):87--92, 1994.

....neural network means the setting of the parameters of Goos et al.: Utilization of Artificial Intelligent Techniques in Making Offers. 103 the processing elements according to some learning procedure like backpropagation. Fundamental description about the neural networks is in following references [3, 5, 20]. The neural networks have been applied in various areas and articles have been published in journals, for example [13] and [14] Genetic algorithms are a category of random search techniques. The search process is based on the natural evolution process. The method consists of a population of ....

David. E. Rumelhart et. al. The basic ideas in neural networks. Communications of the ACM, 37(3):87--92, 1994.

No context found.

D. Rumelhart, B. Widrow, and M. Lehr. The basic ideas in neural networks. Communications of the ACM, 37(3), 1994.

No context found.

D.E.Rumelhart, B.Widrow and M.A.Lehr, The basic ideas in neural networks, Communications of ACM, 37(1994) 87-92.

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