M.C. Pegalajar

A real-coded genetic algorithm for training recurrent neural networks

The use of Recurrent Neural Networks is not as extensive as Feedforward Neural Networks. Training algorithms for Recurrent Neural Networks, based on the error gradient, are very unstable in their search for a minimum and require much computational time when the number of neurons is high. The problems surrounding the application of these methods have driven us to develop new training tools. In this paper, we present a Real-Coded Genetic Algorithm that uses the appropriate operators for this encoding type to train Recurrent Neural Networks. We describe the algorithm and we also experimentally compare our Genetic Algorithm with the Real-Time Recurrent Learning algorithm to perform the fuzzy grammatical inference.

A genetic algorithm to obtain the optimal recurrent neural network

Abstract Selecting the optimal topology of a neural network for a particular application is a difficult task. In the case of recurrent neural networks, most methods only induce topologies in which their neurons are fully connected. In this paper, we present a genetic algorithm capable of obtaining not only the optimal topology of a recurrent neural network but also the least number of connections necessary. Finally, this genetic algorithm is applied to a problem of grammatical inference using neural networks, with very good results.

Multiobjective Hybrid Optimization and Training of Recurrent Neural Networks

The application of neural networks to solve a problem involves tasks with a high computational cost until a suitable network is found, and these tasks mainly involve the selection of the network topology and the training step. We usually select the network structure by means of a trial-and-error procedure, and we then train the network. In the case of recurrent neural networks (RNNs), the lack of suitable training algorithms sometimes hampers these procedures due to vanishing gradient problems. This paper addresses the simultaneous training and topology optimization of RNNs using multiobjective hybrid procedures. The proposal is based on the SPEA2 and NSGA2 algorithms for making hybrid methods using the Baldwinian hybridization strategy. We also study the effects of the selection of the objectives, crossover, and mutation in the diversity during evolution. The proposals are tested in the experimental section to train and optimize the networks in the competition on artificial time-series (CATS) benchmark.

Full-range optical pH sensor based on imaging techniques.

A new colour-based disposable sensor array for a full pH range (0-14) is described. The pH sensing elements are a set of different pH indicators immobilized in plasticized polymeric membranes working by ion-exchange or co-extraction. The colour changes of the 11 elements of the optical array are obtained from a commercial scanner using the hue or H component of the hue, saturation, value (HSV) colour space, which provides a robust and precise parameter, as the analytical parameter. Three different approaches for pH prediction from the hue H of the array of sensing elements previously equilibrated with an unknown solution were studied: Linear model, Sigmoid competition model and Sigmoid surface model providing mean square errors (MSE) of 0.1115, 0.0751 and 0.2663, respectively, in the full-range studied (0-14). The performance of the optical disposable sensor was tested for pH measurement, validating the results against a potentiometric reference procedure. The proposed method is quick, inexpensive, selective and sensitive and produces results similar to other more complex optical approaches for broad pH sensing.

A multiobjective genetic algorithm for obtaining the optimal size of a recurrent neural network for grammatical inference

Grammatical inference has been extensively studied in recent years as a result of its wide field of application, and in turn, recurrent neural networks have proved themselves to be a good tool for grammatical inference. The learning algorithms for these neural networks, however, have been far less studied than those for feed-forward neural networks. Classical training methods for recurrent neural networks suffer from being trapped in local minimal and having a high computational time. In addition, selecting the optimal size of a neural network for a particular application is a difficult task. This suggests that the problems of developing methods to determine optimal topologies and new training algorithms should be studied. In this paper, we present a multi-objective evolutionary algorithm which is able to determine the optimal size of recurrent neural networks in any particular application. This is specially analyzed in the case of grammatical inference: in particular, we study how to establish the optimal size of a recurrent neural network in order to learn positive and negative examples in a certain language, and how to determine the corresponding automaton using a self-organizing map once the training has been completed.

Design and implementation of intelligent systems with LEGO Mindstorms for undergraduate computer engineers

We provide a set of projects to put in practice artificial intelligence techniques using LEGO Mindstorms in an undergraduate computer degree, covering reactive and deliberative agents, rule‐based systems, graph search algorithms, and planning methods. The projects have been applied for teaching in a third‐year undergraduate subject of a computer engineering degree at the University of Granada (Spain). After the contextualization and development of the projects, we discuss the results, advantages, and drawbacks of our experience.

Improving learning management through semantic web and social networks in e-learning environments

Internet social networks have arisen in the last years as powerful tools where people exchange knowledge and multimedia content. They help to share interests between groups of people with common features. Undoubtedly, there is an inherent social network in any e-learning system, where the main actors are teachers, learners and learning resources. Most e-learning software are mainly focused in content dissemination and group work, but the possibilities that Internet LMSs could offer go further. Recently, there has been research work focused on Web Communities for learning and their formulation as Social Networks. Thus, social network analysis may be applied to infer group structures and to make intelligent recommendation systems and data mining. This paper proposes a method for the formulation and interpretation of learning management platforms as social networks. In order to achieve a major generalization, we develop an ontology to integrate the information from different Learning Management Systems. After that, a personalized social network is extracted from the ontology. This change in the point of view of a LMS could be a challenge to make further studies about learners, teachers and learning resources to obtain a better understanding of their social structure, and therefore to make or improve decisions about the learning process.

Feasibility of the use of disposable optical tongue based on neural networks for heavy metal identification and determination.

This study presents the development and characterization of a disposable optical tongue for the simultaneous identification and determination of the heavy metals Zn(II), Cu(II) and Ni(II). The immobilization of two chromogenic reagents, 1-(2-pyridylazo)-2-naphthol and Zincon, and their arrangement forms an array of membranes that work by complexation through a co-extraction equilibrium, producing distinct changes in color in the presence of heavy metals. The color is measured from the image of the tongue acquired by a scanner working in transmission mode using the H parameter (hue) of the HSV color space, which affords robust and precise measurements. The use of artificial neural networks (ANNs) in a two-stage approach based on color parameters, the H feature of the array, makes it possible to identify and determine the analytes. In the first stage, the metals present above a threshold of 10(-7) M are identified with 96% success, regardless of the number of metals present, using the H feature of the two membranes. The second stage reuses the H features in combination with the results of the classification procedure to estimate the concentration of each analyte in the solution with acceptable error. Statistical tests were applied to validate the model over real data, showing a high correlation between the reference and predicted heavy metal ion concentration.

Memetic evolutionary training for recurrent neural networks: an application to time‐series prediction

: Artificial neural networks are bio-inspired mathematical models that have been widely used to solve complex problems. The training of a neural network is an important issue to deal with, since traditional gradient-based algorithms become easily trapped in local optimal solutions, therefore increasing the time taken in the experimental step. This problem is greater in recurrent neural networks, where the gradient propagation across the recurrence makes the training difficult for long-term dependences. On the other hand, evolutionary algorithms are search and optimization techniques which have been proved to solve many problems effectively. In the case of recurrent neural networks, the training using evolutionary algorithms has provided promising results. In this work, we propose two hybrid evolutionary algorithms as an alternative to improve the training of dynamic recurrent neural networks. The experimental section makes a comparative study of the algorithms proposed, to train Elman recurrent neural networks in time-series prediction problems.

A common framework for information sharing in e-learning management systems

Internet Learning Management Systems (LMSs) are powerful tools that help us in our daily teaching and learning activities. Most users and software are mainly focused in content dissemination and group works, but the possibilities that Internet LMSs could offer go further. Some recent approaches use semantic web to improve the capabilities and user experiences in e-learning by mean of artificial intelligence and knowledge management techniques. In this work, we develop a procedure to achieve the integration of different e-learning systems, and to give semantics to entities and relations in the database of LMSs by mean of ontologies. This integration could ease the dissemination of learning resources and knowledge from the databases of the Learning Management Systems. Moreover, the semantic interpretation of database schemes would allow to find precise information quickly.