Olga Semenkina

Comprehensive evolutionary approach for neural network ensemble automatic design

Neural network ensemble is an approach based on cooperative usage of many neural networks for problem solving. Often this approach enables to solve problem more efficiently than approach where only one network is used. The two major stages of the neural network ensemble construction are: design and training component networks, combining of the component networks predictions to produce the ensemble output. In this paper, a probability-based method is proposed to accomplish the first stage. Although this method is based on the genetic algorithm, it requires fewer parameters to be tuned. A method based on genetic programming is proposed for combining the predictions of component networks. This method allows us to build nonlinear combinations of component networks predictions providing more flexible and adaptive solutions. To demonstrate robustness of the proposed approach, its results are compared with the results obtained using other methods.

Self-Configuring Hybrid Evolutionary Algorithm for Fuzzy Imbalanced Classification with Adaptive Instance Selection

Abstract A novel approach for instance selection in classification problems is presented. This adaptive instance selection is designed to simultaneously decrease the amount of computation resources required and increase the classification quality achieved. The approach generates new training samples during the evolutionary process and changes the training set for the algorithm. The instance selection is guided by means of changing probabilities, so that the algorithm concentrates on problematic examples which are difficult to classify. The hybrid fuzzy classification algorithm with a self-configuration procedure is used as a problem solver. The classification quality is tested upon 9 problem data sets from the KEEL repository. A special balancing strategy is used in the instance selection approach to improve the classification quality on imbalanced datasets. The results prove the usefulness of the proposed approach as compared with other classification methods.

Multicriteria Neural Network Design in the Speech-based Emotion Recognition Problem

In this paper we introduce the two-criterion optimization model to design multilayer perceptrons taking into account two objectives, which are the classification accuracy and computational complexity. Using this technique, it is possible to simplify the structure of neural network classifiers and at the same time to keep high classification accuracy. The main benefits of the approach proposed are related to the automatic choice of activation functions, the possibility of generating the ensemble of classifiers, and the embedded feature selection procedure. The cooperative multi-objective genetic algorithm is used as an optimizer to determine the Pareto set approximation in the two-criterion problem. The effectiveness of this approach is investigated on the speech-based emotion recognition problem. According to the results obtained, the usage of the proposed technique might lead to the generation of classifiers comprised by fewer neurons in the input and hidden layers, in contrast to conventional models, and to an increase in the emotion recognition accuracy by up to a 4.25% relative improvement due to the application of the ensemble of classifiers.

Instance Selection Approach for Self-Configuring Hybrid Fuzzy Evolutionary Algorithm for Imbalanced Datasets

We propose an instance selection technique with subsample balancing for an evolutionary classification algorithm. The technique creates subsamples of the training sample in a way to guide the learning process towards problematic areas of the search space. For unbalanced datasets, the number of instances of different classes is artificially balanced to get better classification results. We apply this technique to a self-configured hybrid evolutionary fuzzy classification algorithm. We performed tests on 4 datasets to evaluate the accuracy as well as other classification quality measures for different parameters of the active instance selection procedure. The results shown by our algorithm are comparable or even better than other algorithms on the same classification problems.

Why don’t you use Evolutionary Algorithms in Big Data?

In this paper we raise the question of using evolutionary algorithms in the area of Big Data processing. We show that evolutionary algorithms provide evident advantages due to their high scalability and flexibility, their ability to solve global optimization problems and optimize several criteria at the same time for feature selection, instance selection and other data reduction problems. In particular, we consider the usage of evolutionary algorithms with all kinds of machine learning tools, such as neural networks and fuzzy systems. All our examples prove that Evolutionary Machine Learning is becoming more and more important in data analysis and we expect to see the further development of this field especially in respect to Big Data.

Instance Selection Approach for Self-Configuring Evolutionary Fuzzy Rule Based Classification Systems

We propose an instance selection technique for a hybrid self-configuring fuzzy evolutionary algorithm to decrease the computation time and increase the accuracy. The classification algorithm used implements both Pitts burg and Michigan approaches to generate fuzzy rules. The instance selection method changes the training sample every fixed number of generations, or adaptation periods. The instances of the training sample are assigned probabilities depending on how are they used and how successfully are they classified. The change in probabilities guides the learning algorithm towards problematic areas of the feature space to generate rule bases, that may appear to be more accurate on the whole dataset. The best solution for the whole training sample is kept independently and always included into the population. We demonstrate that this approach decreases the computation time depending on the size of the selected sub sample. We also test the algorithm for different length of the adaptation period. Results of numerical experiments show the usefulness of the approach developed.

Self-configuring hybrid evolutionary algorithm for fuzzy classification with active learning

A novel approach for active training example selection in classification problems is presented. This active selection of training examples is designed to decrease the amount of computation resources required and increase the classification quality achieved. The approach changes the training sample during the evolutionary process so that the algorithm concentrates on problematic instances that are hard to classify. A fuzzy classifier designed with a self-configuring modification of a hybrid evolutionary algorithm is applied as a classification problem solver. The benchmark containing 9 data sets from KEEL is used to prove the usefulness of the approach proposed.

Comparative Analysis of Two Distribution Building Optimization Algorithms

This paper proposes the modification of genetic algorithm, which uses genetic operators, effecting not on particular solutions, but on the probabilities distribution of solution vector’s components. This paper also compares reliability and efficiency of basic algorithm and proposed modification using the set of benchmark functions and real-world problem of dynamic scheduling of truck painting.

Optimization Tools for Support of Decision Making in Design of Spacecrafts' Systems

For the decision support in design of spacecrafts’ systems, the specific optimization tools reflecting properties of arising problems are necessary. Such tools including local and adaptive search algorithms, user support for decomposition and unification of multi-scale variables, solving multicriterial optimization problems and user support for choice of appropriate algorithm, that are under development are discussed.