Juan Seijas

Importance Sampling and Mean-Square Error in Neural Detector Training

This letter deals with the use of Importance Sampling (IS) techniques and the Mean-Square (MS) error in neural network training, for applications to detection in communication systems. Topics such as modifications of the MS objective function, optimal and suboptimal IS probability density functions, and adaptive importance sampling are presented. A genetic algorithm was used for the neural network training, having considered adaptive IS techniques for improving MS error estimations in each iteration of the training. Also, some experimental results of the training process are shown in this letter. Finally, we point out that the mean-square error (estimated by importance sampling) attains quasi-optimum training in the sense of minimum error probability (or minimum misclassification error).

Swarm architectures for Ground-based Air Defense Systems of Systems

New generations of Ground based Air Defense Systems of Systems used in modernized Armed Forces maintain the architectures used in their previous versions, constrained by hierarchical radio communications and centralized Command and Control restrictions. Current state-of-the-art technology in these areas allows a refocusing of these architectures towards a swarm approach. Advantages in terms of effectiveness, scalability and survivability are identified, and a preliminary set of swarm algorithms are proposed and analyzed. Upgrades of these preliminary algorithms are proposed as future research areas.

Edges Detection of Clusters of Microcalcifications with SOM and Coordinate Logic Filters

Breast cancer is one of the leading causes to women mortality in the world. Clusters of Microcalcifications (MCCs) in mammograms can be an important early sign of breast cancer, the detection is important to prevent and treat the disease. Coordinate Logic Filters (CLF), are very efficient in digital signal processing applications, such as noise removal, magnification, opening, closing, skeletonization, and coding, as well as in edge detection, feature extraction, and fractal modelling. This paper presents an edge detector of MCCs in Regions of Interest (ROI) from mammograms using a novel combination. The edge detector consist in the combination of image enhancement by histogram adaptive technique, a Self Organizing Map (SOM) Neural Network and CLF. The experiment results show that the proposed method can locate MCCs edges. Moreover, the proposed method is quantitatively evaluated by Pratts figure of merit together with two widely used edge detectors and visually compared, achieving the best results.

Wavelet Time Shift Properties Integration with Support Vector Machines

This paper presents a short evaluation about the integration of information derived from wavelet non-linear-time-invariant (non-LTI) projection properties using Support Vector Machines (SVM). These properties may give additional information for a classifier trying to detect known patterns hidden by noise. In the experiments we present a simple electromagnetic pulsed signal recognition scheme, where some improvement is achieved with respect to previous work. SVMs are used as a tool for information integration, exploiting some unique properties not easily found in neural networks.

Improving the Efficiency of Multiple Sequence Alignment by Genetic Algorithms

Multiple alignments of biological nucleic acid sequences are one of the most commonly used techniques in sequence analysis. These techniques demand a big computational load. We present a Genetic Algorithms (GA) that optimizes an objective function that is a measure of alignment quality (distance). Each individual in the population represents (in an efficient way) some underlying operations on the sequences and they evolve, by means of natural selection, to better populations where they obtain better alignment of the sequences. The improvement of the effectiveness is obtained by an elitism operator specially designed and by initial bias given to the population by the background knowledge of the user. Our GA presents some characteristics as robustness, convergence to solution, extraordinary capability of generalization and a easiness of being coded for parallel processing architectures, that make our GA very suitable for multiple molecular biology sequences analysis.

Combining edge detection algorithms in intensity SAR images

Edge-detection algorithms are applied to SAR images in order to generate different edge strength maps. The results obtained can be combined to improve the final output in order to get a more defined edge strength map. In this paper different combinations of methods are implemented and evaluated using a criterion based on the quality of the detected edges over the original image. The aspects considered are the false alarm rate, edge thickness and discontinuities inside the detected edges. The results obtained with the edge enhancement techniques are compared and conclusions about the best combination strategies are extracted.

Cooperation and competition based on free market in swarm system architectures for air defense

Algorithms to perform distributed Command and Control functions in Ground based Air Defense Systems of Systems are presented and analyzed. In previous research we presented the benefits of using swarm architectures for this type of application in terms of no single point of failure, lower detectability and higher scalability. Here we propose some Free Market distributed algorithms based on the principles of negotiation and competition among the swarm of agents performing air defense missions. In our research, parameter optimization as well as experimental results in tactical scenarios are obtained through simulation.

Neural Networks Historical Review

This chapter starts with a historical summary of the evolution of Neural Networks from the first models which are very limited in application capabilities to the present ones that make possible to think in applying automatic process to tasks that formerly had been reserved to the human intelligence. After the historical review, Neural Networks are dealt from a computational point of view. This perspective helps to compare Neural Systems with classical Computing Systems and leads to a formal and common presentation that will be used throughout the book

Robustness of artificial metaplasticity learning algorithm

Abstract Artificial metaplasticity learning algorithm is inspired by the biological metaplasticity property of neurons and Shannon׳s information theory. It is based on the bio-inspired hypothesis that neurons do not learn in the same amount (metaplasticity of biological learning) from unfrequent patterns than from common ones, as the former are expected to contain more information than the latter (information entropy concept). On MLPs, the artificial metaplasticity can be formulated as an improvement in regular backpropagation algorithm by using a variable learning rate affecting all the weights in each iteration step and so resembling heterosynaptic plasticity of biological neurons. The variable rate involves statistical inference on the training set and it is common to successfully assume Gaussian distribution for the training patterns. Nevertheless, Gaussian assumption may diverge from the real one and using statistical information extracted from the training patterns may be necessary. In this research, robustness to significative variations on Gaussian assumption is evaluated using input sets generated with different probability distributions. For the cases where Gaussian assumption shows to degrade learning, a general algorithm is applied. This algorithm takes advantage of the inherent statistical inference performed by the MLP through the a posteriori probabilities of input patterns estimation provided by its outputs. The generality of this last algorithm for any input distribution is then demonstrated.

Discrete wavelet cycle integration for pulse detection

Current approaches in electromagnetic pulse detection (radar, communications, etc.) use domain transformations so as to concentrate in frequency related information that can be distinguishable from noise. Wavelet transforms are usually described in two main sets: continuous wavelet transforms and discrete wavelet transforms. Although they share mathematical motivation, both transformations have different algorithms and properties. The property we focus on is the time invariance (also called translation invariance). Continuous wavelets are time invariant, but they are also very expensive to calculate. Real time systems find it hard to process all that information with enough accuracy. On the other hand, uniformly sampling the translation parameter as input to the discrete wavelet process destroys this time invariance. In this paper we introduce experimental results that show an alternative way to generate a time invariant representation of a signal using discrete wavelet transforms. This algorithm upgrades the cost-efficiency of pulse detection capability with respect to the basic approach.