Eric Ronco and Peter Gawthrop. Modular neural networks: a state of the art. Technical report, Centre for System and Control. University of Glasgow, Glasgow, UK, 1995. Home/Search Document Details and Download
Summary Related Articles Check |
This paper is cited in the following contexts:
Applying Simple Combining Techniques with Artificial.. - Sancho.. (Correct)
....of [2] we do not use any kind of preprocessing, because preprocessing is mainly dependent on the nature of the problem, and generally creates ad hoc solutions. 3 Combining Classifiers The idea of combining several modules to improve the results of the each one considered by it self is not new [6, 16]. Already Laplace in 1818 proposed the combination of several estimators to get more reliable results. At the end, the idea is to build systems not to much complicated that probably cannot offer a good accurate solution, but combining them those imperfections can be compensated and produce better ....
Eric Ronco and Peter Gawthrop. Modular neural networks: a state of the art. Technical report, Centre for System and Control. University of Glasgow, Glasgow, UK, 1995.
Neural Networks for Visual Servoing in Robotics - Urban, al. (1997) (Correct)
....ocular saccades, Berthoz et al. 22] have identified several independent modules. Besides, these don t form a serial chain between perception and motorcommand, but operates in multiple sensori motor loops. Neural techniques appropriate to modular decomposition are still barely developed (see e.g. [23] for a state of the art) Learning rules are especially difficult to derive when several interrelated modules collaborate. Neural network applications in the field of robotics are no exception to this present limitation. Modular decompositions are nevertheless easy: numerous tasks can be managed ....
Ronco, E. and Gawthrop, P. (1995), Modular Neural Networks : A State of the Art, Technical Report CSC95026, Center of System and Control Un. Glasgow.
Characterization Of Time-Series By Dynamical Clustering - Reine, Zöller (Correct)
....so lution to this problem is the segmentation of the time series to be modeled into sections that were generated by the hidden sub processes [5] Each segment is then approximated by a different model. Approaches that follow this principle are Mixtures of Experts [3,7] Modular Neural Networks [9] and Local Models [4] The elementary approach uses a two step algorithm, first finding the segment boundaries using a conventional segmentation algorithm and then applying modeling techniques to the segments. The optimality in terms of withinsegment prediction error is not guaranteed. Other ....
E. Ronco and P. Gawthrop. Modular Neural Networks: a state of the art. Tech. Rep. CSC-95026, Univ. of Glasgow, 1995
Neural Simulation Of Acquired And Progressive Language Disorders - Wright, Ahmad (Correct)
....form the input to another. We choose the term multi network architecture in preference to other commonly used terms such as hybrid neural architecture [8] or modular network . These and other terms have been used in the neural network literature, yet there is little consensus on their meaning [9]. Haykin [10] for example, specifies that a neural network is modular if the computation performed. can be decomposed into two or more modules (subsystems) that operate on distinct inputs without communicating with each other . Similar architectures to Haykin s modular network have been ....
Ronco E, Gawthrop P. Modular neural networks: a state of the art. Technical report CSC-95026, Centre for System and Control, University of Glasgow, 1995
Gated Modular Neural Networks for Control Oriented Modelling - Ronco, Gawthrop, Hill (1998) (1 citation) (Correct)
....computation of the information: each input vector is computed by only one part of the whole architecture. This is believed to be the most characteristic feature of a Modular Neural Network (MNN) Embedding modularity into NN gives many advantages over the use of single NN (see for further details (Ronco and Gawthrop, 1995)) Modularity is a natural way to ease the learning of complex behaviours. Mountcastle (Mountcastle, 1978) and many others (Karmiloff Smith, 1994; Houk and Wise, 1995; Carpenter, 1984) argue that modularity is a brain organising principle. Feldman, 1989) and (Simon, 1981) highlight the fact that ....
.... Hence, a modular architecture has the advantage do be easily modifiable (Jacobs and Jordan, 1993; Bottou and Galliinari, 1991) It is also possible to reuse the processing of modules for different tasks instead of having to learn each time some tasks common parts (Fogelman Soulie, 1993; Szilas and Ronco, 1995). The main issues to develop a MNN are the modular organisation in the network architecture and the decomposition of a task (problem, environment) into sub tasks (Ronco and Gawthrop, 1995) A sensible way to decompose a task is to do so according to the computation capability of the modules. ....
[Article contains additional citation context not shown here]
Ronco, Eric and Peter J. Gawthrop (1995). Modular neural networks: a state of the art. Technical Report CSC-95026. Centre for System and Control. Faculty of mechanical Engineering, University of Glasgow, Uk. Available at http://www.ee.usyd.edu.au/~ericr/pub.
Modular Neural Networks and Self-Decomposition - Ronco, Gollee, Gawthrop (1997) (2 citations) Self-citation (Ronco) (Correct)
No context found.
Ronco, E. and Peter J. Gawthrop (1995a). Modular neural networks: a state of the art. Technical Report CSC-95026. Centre for System and Control.
Neural Networks for Modelling and Control - Ronco, Gawthrop (1997) (1 citation) Self-citation (Gawthrop) (Correct)
.... the MLP can approximate any computable function (Cybenko, 1988; Lapedes and Farber, 1987) However, the usefulness of this algorithm, especially for modelling and control purposes, appeared clearly highly constrained by the three following features (see for details (Ronco, 1994; Szilas and Ronco, 1995; Ronco and Gawthrop, 1995) ffl There is no systematic way to determine the required structure to model a given system. This makes the above demonstrations almost irrelevant. ffl The back propagation learning algorithm essentially based on the gradient descent does not ensure any convergence ....
.... can approximate any computable function (Cybenko, 1988; Lapedes and Farber, 1987) However, the usefulness of this algorithm, especially for modelling and control purposes, appeared clearly highly constrained by the three following features (see for details (Ronco, 1994; Szilas and Ronco, 1995; Ronco and Gawthrop, 1995)) ffl There is no systematic way to determine the required structure to model a given system. This makes the above demonstrations almost irrelevant. ffl The back propagation learning algorithm essentially based on the gradient descent does not ensure any convergence toward a solution (even ....
[Article contains additional citation context not shown here]
Ronco, Eric and Peter J. Gawthrop (1995). Modular neural networks: a state of the art. Technical Report CSC-95026. Centre for System and Control. Faculty of mechanical Engineering, University of Glasgow, Uk. Available at www.mech.gla.ac.uk/~ericr/pub/surveyMNN.ps.
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