Fritzke, B. [1994], "Making hard problems linearly separable -- incremental radial basis function approaches," (submitted to ICANN'94: International Conference on Artificial Neural Networks), Sorrento, Italy. @ NUS Home/Search Document Not in Database Summary Related Articles Check |
This paper is cited in the following contexts:
Function Approximation With Continuous-Valued.. - Martin Eldracher, .. (1994) (Correct)
....of the encoding. Missing to model the cross correlations between the coordinates directly in the encoding, the concentration of granule cells also arises in areas, where this is not needed in order to solve the function approximation problem. However, compared to other approaches, e.g. Fritzke (1994), our approach is very fast, since it is based on the CMAC approach. Additionally Fritzke (1994) does not change the shape of the receptive fields according to the cross correlations of the coordinate errors, but uses fixed shaped receptive fields. Using CMAC for speed reasons, a direct encoding ....
....in the encoding, the concentration of granule cells also arises in areas, where this is not needed in order to solve the function approximation problem. However, compared to other approaches, e.g. Fritzke (1994) our approach is very fast, since it is based on the CMAC approach. Additionally Fritzke (1994) does not change the shape of the receptive fields according to the cross correlations of the coordinate errors, but uses fixed shaped receptive fields. Using CMAC for speed reasons, a direct encoding of a concept of moving the granule cells according to the 4 At each intersection point of two ....
Fritzke, B. (1994). Making hard problems linearly separable -- incremental radial basis function approaches. In Proceedings of the International Conference on Artificial Neural Networks (ICANN'94), pp. 455--458.
Supervised Learning with Growing Cell Structures - Fritzke (1994) (9 citations) Self-citation (Fritzke) (Correct)
....oe. One extreme approach is to use one unit per data points and to position the units directly at the data points. If one chooses the width of the Gaussians sufficiently small it is possible to construct a network which correctly classifies the training data, no matter how complicated the task is (Fritzke, 1994). However, the network size is very large and might even be infinite in the case of a continuous stream of non repeating stochastic input data. Moreover, such a network can be expected to generalize poorly. Moody Darken (1989) in contrast, propose to use a fixed number of local units (which is ....
Fritzke, B. [1994], "Making hard problems linearly separable -- incremental radial basis function approaches," (submitted to ICANN'94: International Conference on Artificial Neural Networks), Sorrento, Italy.
Transforming Hard Problems into Linearly Separable ones with.. - Fritzke (1996) Self-citation (Fritzke) (Correct)
....of the input space with a limited overlap between neighboring units. ffl The centers of the Gaussians are fine tuned by competitive learning. This strategy seems to lead to small networks with a very strong generalization ability. We have reported results of the growing cell structures in [3]. Here we like to show an example simulation for the successor model growing neural gas [2, 4] which at least for supervised learning replaces the growing cell structures. Both models differ only in the constraints imposed on their topology. The growing cell structures have a topology consisting ....
B. Fritzke. Making hard problems linearly separable -- incremental radial basis function approaches. In M. Marinaro and P. G. Morasso, editors, ICANN'94: International Conference on Artificial Neural Networks, pages 455--458, Sorrento, Italy, 1994. Springer.
Online articles have much greater impact More about CiteSeer.IST at NUS Add search form to your site Submit documents Feedback
CiteSeer.IST at NUS - Copyright Penn State and NEC. Hosted by the School of Computing, National University of Singapore.