Jorge Dantas de Melo

A Neural Network Multiagent Architecture Applied to Industrial Networks for Dynamic Allocation of Control Strategies Using Standard Function Blocks

This paper presents a multiagent architecture applied to factory automation. These agents detect faults in automated processes and allocate intelligent algorithms in field device function blocks (FBs) to solve these faults. We also present a dynamic FB parameter exchange strategy that allows agent fieldbus allocation. This architecture is a foundation for intelligent physical agents standard-based agent platform developed using Foundation Fieldbus technology. The aim is to enable problem detection activities, independent of user intervention. The use of artificial neural network (ANN)-based algorithms enables the agents to learn about fault patterns and adapt an algorithm that can be used in fault situations. Thus, we intend to reduce supervisor intervention in selecting and implementing an appropriate structure for FB algorithms. Furthermore, these algorithms, when implemented in device FBs, provide a solution at the fieldbus level, reducing data traffic between gateway and device, and speeding up the process of problem resolution. We also show some examples of our approach. The first is a neural network architecture change that allocates different types of neural networks in field devices without interrupting the fieldbus network operation. The second shows a multiagent architecture that implements the neural network change in a laboratory test process, where fault scenarios have been simulated.

Creating an ensemble of diverse support vector machines using Adaboost

Support vector machines are one of the most employed methods of pattern classification, and the Adaboost algorithm is an effective way of improving the performance of the weak learners that compose the ensemble. In this article, we propose to create an Adaboost-based ensemble of SVM, by altering the Gaussian width parameter of the RBF-SVM. Using data sets from the UCI repository, we made tests to evaluate the algorithm.

Reactive Search strategies using Reinforcement Learning, local search algorithms and Variable Neighborhood Search

Abstract Optimization techniques known as metaheuristics have been applied successfully to solve different problems, in which their development is characterized by the appropriate selection of parameters (values) for its execution. Where the adjustment of a parameter is required, this parameter will be tested until viable results are obtained. Normally, such adjustments are made by the developer deploying the metaheuristic. The quality of the results of a test instance [The term instance is used to refer to the assignment of values to the input variables of a problem.] will not be transferred to the instances that were not tested yet and its feedback may require a slow process of “trial and error” where the algorithm has to be adjusted for a specific application. Within this context of metaheuristics the Reactive Search emerged defending the integration of machine learning within heuristic searches for solving complex optimization problems. Based in the integration that the Reactive Search proposes between machine learning and metaheuristics, emerged the idea of putting Reinforcement Learning, more specifically the Q-learning algorithm with a reactive behavior, to select which local search is the most appropriate in a given time of a search, to succeed another local search that can not improve the current solution in the VNS metaheuristic. In this work we propose a reactive implementation using Reinforcement Learning for the self-tuning of the implemented algorithm, applied to the Symmetric Travelling Salesman Problem.

Classification of cardiac arrhythmias using competitive networks

Information generated by sensors that collect a patients vital signals are continuous and unlimited data sequences. Traditionally, this information requires special equipment and programs to monitor them. These programs process and react to the continuous entry of data from different origins. Thus, the purpose of this study is to analyze the data produced by these biomedical devices, in this case the electrocardiogram (ECG). Processing uses a neural classifier, Kohonen competitive neural networks, detecting if the ECG shows any cardiac arrhythmia. In fact, it is possible to classify an ECG signal and thereby detect if it is exhibiting or not any alteration, according to normality.

FPGA based implementation of a Fuzzy Neural Network modular architecture for embedded systems

This paper presents a FPGA based approach for a modular architecture of Fuzzy Neural Networks (FNN) to embed with easily different topologies set up. The project is based on a Takagi - Hayashi (T-H) method for the construction and tuning of fuzzy rules, this is commonly referred as neural network driven fuzzy reasoning. The proposed architecture approach consists of two main configurable modules: a Multilayer Perceptron - MLP with sigmoidal activation function that composes the first module to determine a Fuzzy membership function; the second employs an MLP with pure linear activation function to define the consequents. The DSPBuilder® software along the Simulink® is used to connect, set and synthesize the Fuzzy Neural Network desired. Other hardware components employed in the architecture proposed cooperate to the system modularity. The system was tested and validated through a control problem and an interpolation problem. Several papers proposed different hardware architecture to implement hybrid systems by using Fuzzy logic and Neural Network. However, there is no approach with this specific neural network driven fuzzy reasoning by T-H method and the aim to be embedded. The Self-Organizing Map (SOM) and Levenberg-Marquardt backpropagation were used to train the FNN proposed off-line.

An genetic approach to Support Vector Machines in classification problems

There are a lot of different methods in pattern classification, in which one of the most popular is the Support Vector Machine. Lots of tools have been developed to improve SVM classification, mainly the development of new classifying methods and the employment of SVM ensembles. Meanwhile, evolutionary algorithms are recognized tools to solve optimization problems, and have in the genetic algorithm its most popular metaheuristic. So, in this paper, our proposal is to unite both techniques, applying a genetic algorithm to optimize the classification of a set of SVM, testing with some benchmark data sets.

A parallel hybrid implementation using genetic algorithm, GRASP and reinforcement learning

In the process of searching for better solutions, a metaheuristic can be guided to regions of promising solutions using the acquisition of information on the problem under study. In this work this is done through the use of reinforcement learning. The performance of a metaheuristic can also be improved using multiple search trajectories, which act competitively and/or cooperatively. This can be accomplished using parallel processing. Thus, in this paper we propose a hybrid parallel implementation for the GRASP metaheuristics and the genetic algorithm, using reinforcement learning, applied to the symmetric traveling salesman problem.

Implementation and Evaluation of Modular Neural Networks in a Multiple Processor System on Chip to Classify Electric Disturbance

This paper shows the effectiveness of modular neural networks composed of multilayers experts trained with a hybrid algorithm implemented in a multiprocessor system on chip. The network is applied on the classification of electric disturbances. The objective is to show that, even a FPGA with hardware restrictions, could be used to implement a complex problem, when parallel processing is used. To improve the system performance was used four soft processors with a shared memory

Random subspace method and genetic algorithm applied to a LS-SVM ensemble

The Least Squares formulation of SVM (LS-SVM) finds the solution by solving a set of linear equations instead of quadratic programming implemented in SVM. The LS-SVMs provide some free parameters that have to be correctly chosen in order that the performance. Lots of tools have been developed to improve their performance, mainly the development of new classifying methods and the employment of ensembles. So, in this paper, our proposal is to use both the theory of ensembles and a genetic algorithm to enhance the LS-SVM classification. First, we randomly divide the problem into subspaces to generate diversity among the classifiers of the ensemble. So, we apply a genetic algorithm to find the values of the LS-SVM parameters and also to find the weights of the linear combination of the ensemble members, used to take the final decision.

Comparative study on dimension reduction techniques for cluster analysis of microarray data

This paper proposes a study on the impact of the use of dimension reduction techniques (DRTs) in the quality of partitions produced by cluster analysis of microarray datasets. We tested seven DRTs applied to four microarray cancer datasets and ran four clustering algorithms using the original and reduced datasets. Overall results showed that using DRTs provides a improvement in performance of all algorithms tested, specially in the hierarchical class. We could see that, despite Principal Component Analysis (PCA) being the most widely used DRT, its was overcome by other nonlinear methods and it did not provide a substantial performance increase in the clustering algorithms. On the other hand, t-distributed Stochastic Embedding (t-SNE) and Laplacian Eigenmaps (LE) achieved good results for all datasets.