J. J. Merelo

G-Prop: Global optimization of multilayer perceptrons using GAs

Abstract A general problem in model selection is to obtain the right parameters that make a model fit observed data. For a multilayer perceptron (MLP) trained with back-propagation (BP), this means finding appropiate layer size and initial weights. This paper proposes a method (G-Prop, genetic backpropagation) that attempts to solve that problem by combining a genetic algorithm (GA) and BP to train MLPs with a single hidden layer. The GA selects the initial weights and changes the number of neurons in the hidden layer through the application of specific genetic operators. G-Prop combines the advantages of the global search performed by the GA over the MLP parameter space and the local search of the BP algorithm. The application of the G-Prop algorithm to several real-world and benchmark problems shows that MLPs evolved using G-Prop are smaller and achieve a higher level of generalization than other perceptron training algorithms, such as Quick-Propagation or RPROP, and other evolutive algorithms, such as G-LVQ.

Evolving Multilayer Perceptrons

This paper proposes a new version of a method (G-Prop, genetic backpropagation) that attempts to solve the problem of finding appropriate initial weights and learning parameters for a single hidden layer Multilayer Perceptron (MLP) by combining an evolutionary algorithm (EA) and backpropagation (BP). The EA selects the MLP initial weights, the learning rate and changes the number of neurons in the hidden layer through the application of specific genetic operators, one of which is BP training. The EA works on the initial weights and structure of the MLP, which is then trained using QuickProp; thus G-Prop combines the advantages of the global search performed by the EA over the MLP parameter space and the local search of the BP algorithm. The application of the G-Prop algorithm to several real-world and benchmark problems shows that MLPs evolved using G-Prop are smaller and achieve a higher level of generalization than other perceptron training algorithms, such as QuickPropagation or RPROP, and other evolutive algorithms. It also shows some improvement over previous versions of the algorithm.

g-lvq, a combination of genetic algorithms and lvq

One of the ultimate goal in neural network research is the complete optimization of a neural network: topology, learning algorithm, initial weights and number of neurons. Up to now, only partial solutions have been found. This optimization should look at two conditions if the task assigned to the NN is going to be classification: accuracy, and obtention of a good representation of the sample. lvq neural nets are a supervised classification algorithms created by Kohonen. In this work, lvq NNs will be optimized using a GA according to three criteria: accuracy, net size and distortion. These three criteria are considered a vector fitness with three components that must be optimized separately. In order to carry this out, variable-length genomes are used to represent the neural network; each neuron is codified together with its label. Results in synthetic and real-world problems show that g-lvq is able to find an optimal size of the network, as well as combinations of weights that maximize classification accuracy.

Mapping and fuzzy classification of macromolecular images using self-organizing neural networks.

In this work the effectiveness of the fuzzy kohonen clustering network (FKCN) in the unsupervised classification of electron microscopic images of biological macromolecules is studied. The algorithm combines Kohonens self-organizing feature maps (SOFM) and Fuzzy c-means (FCM) in order to obtain a powerful clustering technique with the best properties inherited from both. Exploratory data analysis using SOFM is also presented as a step previous to final clustering. Two different data sets obtained from the G40P helicase from B. Subtilis bacteriophage SPP1 have been used for testing the proposed method, one composed of 2458 rotational power spectra of individual images and the other composed by 338 images from the same macromolecule. Results of FKCN are compared with self-organizing feature maps (SOFM) and manual classification. Experimental results prove that this new technique is suitable for working with large, high-dimensional and noisy data sets and, thus, it is proposed to be used as a classification tool in electron microscopy.

G-Prop-II: global optimization of multilayer perceptrons using GAs

A general problem in model selection is to obtain the right parameters that make a model fit observed data. For a multilayer perceptron (MLP) trained with backpropagation (BP), this means finding the right hidden layer size, appropriate initial weights and learning parameters. The paper proposes a method (G-Prop-II) that attempts to solve that problem by combining a genetic algorithm (GA) and BP to train MLPs with a single hidden layer. The GA selects the initial weights and the learning rate of the network, and changes the number of neurons in the hidden layer through the application of specific genetic operators. G-Prop-II combines the advantages of the global search performed by the GA over the MLP parameter space and the local search of the BP algorithm. The application of the G-Prop-II algorithm to several real world and benchmark problems shows that MLPs evolved using G-Prop-II are smaller and achieve a higher level of generalization than other perceptron training algorithms, such as QuickPropagation or RPROP, and other evolutive algorithms, such as G-LVQ. It also shows some improvement over previous versions of the algorithm.

Automatic Classification of Biological Particles fromElectron-microscopy Images Using Conventional and Genetic-algorithm Optimized Learning Vector Quantization

Automatic classification of transmission electron-microscopy images is an important step in the complex task of determining the structure of biologial macromolecules. The process of 3D reconstruction from a set of such images implies their previous classification into homogeneous image classes. In general, different classes may represent either distinct biochemical specimens or specimens from different directions of an otherwise homogenous specimen. In this paper, a neural network classification algorithm has been applied to a real-data case in which it was known a priori the existence of two differentiated views of the same specimen. Using two labeled sets as a reference, 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 (genetic learning vector quantization) [10] algorithm, and compared to a non-optimized version of the algorithm, Kohonens lvq (learning vector quantization) [7]. Using a part of the sample as training set, the results presented here show an efficient (approximately 90%) average classification rate of unknown samples in two classes. Finally, the implication of this kind of automatic classification of algorithms in the determination of three dimensional structure of biological particles is discused. This paper extends the results already presented in [11], and also improves them.

A new approach to fuzzy controller designing and coding via genetic algorithms

Examines the applicability of genetic algorithms to fuzzy controller optimization and presents a methodology to simultaneously design membership functions and rule sets for them. We propose a new approach to fuzzy controller coding via genetic algorithms which achieves a reduction in the number of parameters that define the membership functions, and a better understanding of the resulting system with no loss of efficiency.

Modeling Malaria with Multi-Agent Systems

Malaria is a vector-borne disease that greatly affects social and economic development. We adopt the complex system paradigm in our analysis of the problem. Our aim is to assess the impact of education on malaria healthcare. Multi-agent systems are employed to model the spread of malaria in Haiti, where we introduce malaria education as a possible way of regulating deaths due to the parasite. We launch three experiments, each with environment modifications: three hospitals; three hospitals and 20 schools; and five hospitals and 20 schools. The results of running 10 simulations for each experiment show that there is a reduction in malaria deaths not only when including schools, but in combination with increasing the number of hospitals.

Evolutionary algorithm for speech segmentation

Speech segmentation is one of the problems in the speech processing area. The main techniques that attempt to solve it are manual segmentation and hidden Markov model alignment. In this work a new technique based on an evolutionary algorithm that permits to segment the speech without a previous training process is presented.

Application of the Fuzzy Kohonen Clustering Network to biological macromolecules images classification

In this work we study the effectiveness of the Fuzzy Kohonen Clustering Network (FKCN) in the unsupervised classification of electron microscopic images of biological macromolecules. The algorithm combines Kohonens Self-Organizing Feature Maps (SOM) and Fuzzy c-means clustering technique (FCM) in order to obtain a powerful clustering technique that inherits their best properties. Two different data sets obtained from the G40P helicase from B. Subtilis bacteriophage SPP1 have been used for testing the proposed method, one composed of 2458 rotational power spectra of individual images and the other composed by 338 images from the same macromolecule. Results of FKCN are compared with Self-Organizing Maps (SOM) and manual classification. Experimental results have proved that this new technique is suitable for working with large, high dimensional and noisy data sets. This method is proposed to be used as a classification tool in Electron Microscopy.