Beatriz A. Garro

Design of artificial neural networks using a modified Particle Swarm Optimization algorithm

In the last years, bio-inspired algorithms have shown their power in different non-linear optimization problems. Due to the efficiency and adaptability of bio-inspired algorithms, in this paper we explore a new way to design an artificial neural network (ANN). For this task, a modified PSO algorithm was used. We do not only study the problem of finding the optimal synaptic weights of an ANN but also its topology and transfer functions. In other words, given a set of inputs and desired patterns, with the proposal we are able to find the best topology, the number of neurons, the transfer function for each neuron, as well as the synaptic weights. This allows to designing an ANN to be used to solve a given problem. The proposal is tested using several non-linear problems.

Artificial neural network synthesis by means of artificial bee colony (ABC) algorithm

Artificial bee colony (ABC) algorithm has been used in several optimization problems, including the optimization of synaptic weights from an Artificial Neural Network (ANN). However, this is not enough to generate a robust ANN. For that reason, some authors have proposed methodologies based on so-called metaheuristics that automatically allow designing an ANN, taking into account not only the optimization of the synaptic weights as well as the ANNs architecture, and the transfer function of each neuron. However, those methodologies do not generate a reduced design (synthesis) of the ANN. In this paper, we present an ABC based methodology, that maximizes its accuracy and minimizes the number of connections of an ANN by evolving at the same time the synaptic weights, the ANNs architecture and the transfer functions of each neuron. The methodology is tested with several pattern recognition problems.

Evolving ant colony system for optimizing path planning in mobile robots

Path planning is one of the problems in robotics. It consists on automatically determine a path from an initial position of the robot to its final position. In this paper we propose a variant of the ant colony system (ACO) applied to optimize the path that a robot can follow to reach its target destination. We also propose to evolve some parameters of the ACO algorithm by using a genetic algorithm (ACO-GA) to optimize the search of the shortest path. We compare the accuracy of ACO against ACO-GA using real environments.

Design of artificial neural networks using differential evolution algorithm

The design of an Artificial Neural Network (ANN) is a difficult task for it depends on the human experience. Moreover it needs a process of testing and error to select which kind of a transfer function and which algorithm should be used to adjusting the synaptic weights in order to solve a specific problem. In the last years, bio-inspired algorithms have shown their power in different nonlinear optimization problems. Due to their efficiency and adaptability, in this paper we explore a new methodology to automatically design an ANN based on the Differential Evolution (DE) algorithm. The proposed method is capable to find the topology, the synaptic weights and the transfer functions to solve a given pattern classification problems.

A new bi-directional associative memory

Hebbian hetero-associative learning is inherently asymmetric. Storing a forward association from pattern A to pattern B enables the recalling of pattern B given pattern A. This, in general, does not allow the recalling of pattern A given pattern B. The forward association between A and B will tend to be stronger than the backward association between B and A. In this paper it is described how the dynamical associative model proposed in [10] can be extended to create a bi-directional associative memory where forward association between A and B is equal to backward association between B and A. This implies that storing a forward association, from pattern A to pattern B, would enable the recalling of pattern B given pattern A and the recalling of pattern A given pattern B. We give some formal results that support the functioning of the proposal, and provide some examples were the proposal finds application.

Evolving neural networks: a comparison between differential evolution and particle swarm optimization

Due to their efficiency and adaptability, bio-inspired algorithms have shown their usefulness in a wide range of different non-linear optimization problems. In this paper, we compare two ways of training an artificial neural network (ANN): Particle Swarm Optimization (PSO) and Differential Evolution (DE) algorithms. The main contribution of this paper is to show which of these two algorithms provides the best accuracy during the learning phase of an ANN. First of all, we explain how the ANN training phase could be seen as an optimization problem. Then, we explain how PSO and DE could be applied to find the best synaptic weights of the ANN. Finally, we perform a comparison between PSO and DE approaches when used to train an ANN applied to different non-linear problems.

Low frequency response and random feature selection applied to face recognition

A novel method for face recognition based on some biological aspects of infant vision is proposed in this paper. The biological hypotheses of this method are based on the role of the response to low frequencies at early stages, and some conjectures concerning how an infant detects subtle features (stimulating points) from a face. In order to recognize a face from different images of it we make use of a bank of dynamic associative memories (DAM). As the infant vision responds to low frequencies of the signal, a low-filter is first used to remove high frequency components from the image. We then detect subtle features in the image by means of a random feature selection detector. At last, the network of DAMs is fed with this information for training and recognition. To test the accuracy of the proposal a benchmark of faces is used.

Training spiking neurons by means of particle swarm optimization

Meta-heuristic algorithms inspired by nature have been used in a wide range of optimization problems. These types of algorithms have gained popularity in the field of artificial neural networks (ANN). On the other hand, spiking neural networks are a new type of ANN that simulates the behaviour of a biological neural network in a more realistic manner. Furthermore, these neural models have been applied to solve some pattern recognition problems. In this paper, it is proposed the use of the particle swarm optimization (PSO) algorithm to adjust the synaptic weights of a spiking neuron when it is applied to solve a pattern classification task. Given a set of input patterns belonging to K classes, each input pattern is transformed into an input signal. Then, the spiking neuron is stimulated during T ms and the firing rate is computed. After adjusting the synaptic weights of the neural model using the PSO algorithm, input patterns belonging to the same class will generate similar firing rates. On the contrary, input patterns belonging to other classes will generate firing rates different enough to discriminate among the classes. At last, a comparison between the PSO algorithm and a differential evolution algorithm is presented when the spiking neural model is applied to solve non-linear and real object recognition problems.

3D object recognition based on low frequency response and random feature selection

In this paper we propose a view-based method for 3D object recognition based on some biological aspects of infant vision. The biological hypotheses of this method are based on the role of the response to low frequencies at early stages, and some conjectures concerning how an infant detects subtle features (stimulating points) from an object. In order to recognize an object from different images of it (different orientations from 0° to 100deg;) we make use of a dynamic associative memory (DAM). As the infant vision responds to low frequencies of the signal, a low-filter is first used to remove high frequency components from the image. Then we detect subtle features in the image by means of a random feature selection detector. At last, the DAM is fed with this information for training and recognition. To test the accuracy of the proposal we use the Columbia Object Image Library (COIL 100) database.

Diseño Automático de Redes Neuronales Artificiales mediante el uso del Algoritmo de Evolución Diferencial (ED)

En el area de la Inteligencia Artificial, las Redes Neuronales Artificiales (RNA) han sido aplicadas para la solucion de multiples tareas. A pesar de su declive y del resurgimiento de su desarrollo y aplicacion, su diseno se ha caracterizado por un mecanismo de prueba y error, el cual puede originar un desempeno bajo. Por otro lado, los algoritmos de aprendizaje que se utilizan como el algoritmo de retropropagacion y otros basados en el gradiente descenciente, presentan una desventaja: no pueden resolver problemas no continuos ni problemas multimodales. Por esta razon surge la idea de aplicar algoritmos evolutivos para disenar de manera automatica una RNA. En esta investigacion, el algoritmo de Evolucion Diferencial (ED) encuentra los mejores elementos principales de una RNA: la arquitectura, los pesos sinapticos y las funciones de transferencia. Por otro lado, dos funciones de aptitud son propuestas: el error cuadraatico medio (MSE por sus siglas en ingles) y el error de clasificacion (CER) las cuales, involucran la etapa de validacion para garantizar un buen desempeno de la RNA. Primero se realizo un estudio de las diferentes configuraciones del algoritmo de ED, y al determinar cual fue la mejor configuracion se realizo una experimentacion exhaustiva para medir el desempeno de la metodologia propuesta al resolver problemas de clasificacion de patrones. Tambien, se presenta una comparativa contra dos algoritmos clasicos de entrenamiento: Gradiente descendiente y Levenberg–Marquardt