Showing papers by "Federico Morán published in 1997"
TL;DR: This work simulates the development of neurons selective to orientation and size organized in a map, underscoring the importance of anti-Hebbian learning for normal neural visual system development.
7 citations
04 Jun 1997
TL;DR: A neural network classification algorithm has been used to perform the classification of electron microscopy samples in two classes using two labeled sets as a refference, and the parameters and architecture of the classifier were optimized using a genetic algorithm.
Abstract: Automatic classification of electron-microscopy images is an important step in the complex task of determination of the structure of biologial macromolecules. The process of 3D reconstruction from the images implies its previous classification in different classes corresponding to the main different views. In this paper a neural network classification algorithm has been used to perform the classification of electron microscopy samples in two classes. Using two labeled sets as a refference, the parameters and architecture of the classifier were optimized using a genetic algorithm. The global automatic process of training and optimization is implemented using the previously described g-lvq algorithm, and compared to a non-optimized version of the algorithm, Kohonen's LVQ. Using a part of the sample as training set, the results presented here show an efficient (90%) classification of unknown samples in two classes. The implication of this kind of automatic classification algorithms in determination of three dimensional structure of biological particles is finaly discused.
5 citations
04 Jun 1997
TL;DR: A model for ontogenetic development of receptive fields in the visual nervous system using a semistochastic approach where random uncorrelated activity is generated in the input layer and propagated through the network.
Abstract: A model for ontogenetic development of receptive fields in the visual nervous system is presented. The model uses a semistochastic approach where random uncorrelated activity is generated in the input layer and propagated through the network. The evolution of the synaptic connections between two neurons are assumed to be a function of their activity, with two interpretations of the Hebb's rule: (a) the synaptic weight is modified proportional to the product of the activity of the two connected neurons; and (b) proportional to the statistical correlation of their activity. Both models explain the origin of either on-off and off-on receptive fields with symetric and non symetric forms. These results agree with previous models based on deterministic equations. The approach presented here has two main advantages. Firstly the lower computer time that allows the study of more complex architectures. And secondly, the possibility of the extension of this model to cover more complex behavior, for instance, the inclusion of time delay in the transmition of the activity between layers.