Christina Brester

Multi-objective heuristic feature selection for speech-based multilingual emotion recognition

Abstract If conventional feature selection methods do not show sufficient effectiveness, alternative algorithmic schemes might be used. In this paper we propose an evolutionary feature selection technique based on the two-criterion optimization model. To diminish the drawbacks of genetic algorithms, which are applied as optimizers, we design a parallel multicriteria heuristic procedure based on an island model. The performance of the proposed approach was investigated on the Speech-based Emotion Recognition Problem, which reflects one of the most essential points in the sphere of human-machine communications. A number of multilingual corpora (German, English and Japanese) were involved in the experiments. According to the results obtained, a high level of emotion recognition was achieved (up to a 12.97% relative improvement compared with the best F-score value on the full set of attributes).

Cooperative Multi-objective Genetic Algorithm with Parallel Implementation

In this paper we introduce the multi-agent heuristic procedure to solve multi-objective optimization problems. To diminish the drawbacks of the evolutionary search, an island model is used to involve various genetic algorithms which are based on different concepts NSGA-II, SPEA2, and PICEA-g. The main benefit of our proposal is that it does not require additional experiments to expose the most appropriate algorithm for the problem considered. For most of the test problems the effectiveness of the developed algorithmic scheme is comparable with or even better than the performance of its component which provides the best results separately. Owing to the parallel work of island model components we have managed to decrease computational time significantly approximately by a factor of 2.7.

Acoustic emotion recognition two ways of features selection based on self-adaptive multi-objective genetic algorithm

In this paper the efficiency of feature selection techniques based on the evolutionary multi-objective optimization algorithm is investigated on the set of speech-based emotion recognition problems (English, German languages). Benefits of developed algorithmic schemes are demonstrated compared with Principal Component Analysis for the involved databases. Presented approaches allow not only to reduce the amount of features used by a classifier but also to improve its performance. According to the obtained results, the usage of proposed techniques might lead to increasing the emotion recognition accuracy by up to 29.37% relative improvement and reducing the number of features from 384 to 64.8 for some of the corpora.

Comparison of Two-Criterion Evolutionary Filtering Techniques in Cardiovascular Predictive Modelling

In this paper we compare a number of two-criterion filtering techniques for feature selection in cardiovascular predictive modelling. We design two-objective schemes based on different combinations of four criteria describing the quality of reduced feature sets. To find attribute subsystems meeting the introduced criteria in an optimal way, we suggest applying a cooperative multi-objective genetic algorithm. It includes various search strategies working in a parallel way, which allows additional experiments to be avoided when choosing the most effective heuristic for the problem considered. The performance of filtering techniques was investigated in combination with the SVM model on a population-based epidemiological database called KIHD (Kuopio Ischemic Heart Disease Risk Factor Study). The dataset consists of a large number of variables on various characteristics of the study participants. These baseline measures were collected at the beginning of the study. In addition, all major cardiovascular events that had occurred among the participants over an average of 27 years of follow-up were collected from the national health registries. As a result, we found that the usage of the filtering technique including intra- and inter-class distances led to a significant reduction of the feature set (up to 11 times, from 433 to 38 features) without detriment to the predictive ability of the SVM model. This implies that there is a possibility to cut down on the clinical tests needed to collect the data, which is relevant to the prediction of cardiovascular diseases.

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.

Speaker state recognition with neural network-based classification and self-adaptive heuristic feature selection

While the implementation of existing feature sets and methods for automatic speaker state analysis has already achieved reasonable results, there is still much to be done for further improvement. In our research, we tried to carry out speech analysis with the self-adaptive multi-objective genetic algorithm as a feature selection technique and with a neural network as a classifier. The proposed approach was evaluated using a number of multi-language speech databases (English, German and Japanese). According to the obtained results, the developed technique allows an increase in emotion recognition performance by up to 6.2% relative improvement in average F-measure, up to 112.0% for the speaker identification task and up to 6.4% for the speech-based gender recognition, having approximately half as many features.

Multi-objective Dynamical System Parameters and Initial Value Identification Approach in Chemical Disintegration Reaction Modelling.

A multi-criteria multi-output dynamical system identification problem is considered. The inverse mathematical problem of estimating the parameters of a system of differential equations and its initial point using the measured data is provided for the hexadecane disintegration reaction. The aim of modelling is to approximate the dynamical behaviour of hexadecane and the concentrations of its products, which according to chemical kinetics are determined by a differential equation. Since the dynamical model adequacy is based on the error between its output and the sample data and the output itself depends on the initial point values, the inverse mathematical modelling problem is the simultaneous estimation of the model parameters and the initial point. At the same time, the initial point is unknown and the sample data is noisy, and for this reason, the inverse mathematical modelling problem is reduced to a two-objective optimization problem. The reduced problem is a sample of black-box optimization problems; it is complex, multimodal and requires a reliable technique to solve it. This is why a specific heterogeneous multi-objective genetic algorithm with the island meta-heuristic is used and its efficiency in solving this problem is proved by the investigation results.

Multi-objective optimization algorithms with the island metaheuristic for effective project management problem solving

Abstract Background and Purpose: In every organization, project management raises many different decision-making problems, a large proportion of which can be efficiently solved using specific decision-making support systems. Yet such kinds of problems are always a challenge since there is no time-efficient or computationally efficient algorithm to solve them as a result of their complexity. In this study, we consider the problem of optimal financial investment. In our solution, we take into account the following organizational resource and project characteristics: profits, costs and risks. Design/Methodology/Approach: The decision-making problem is reduced to a multi-criteria 0-1 knapsack problem. This implies that we need to find a non-dominated set of alternative solutions, which are a trade-off between maximizing incomes and minimizing risks. At the same time, alternatives must satisfy constraints. This leads to a constrained two-criterion optimization problem in the Boolean space. To cope with the peculiarities and high complexity of the problem, evolution-based algorithms with an island meta-heuristic are applied as an alternative to conventional techniques. Results: The problem in hand was reduced to a two-criterion unconstrained extreme problem and solved with different evolution-based multi-objective optimization heuristics. Next, we applied a proposed meta-heuristic combining the particular algorithms and causing their interaction in a cooperative and collaborative way. The obtained results showed that the island heuristic outperformed the original ones based on the values of a specific metric, thus showing the representativeness of Pareto front approximations. Having more representative approximations, decision-makers have more alternative project portfolios corresponding to different risk and profit estimations. Since these criteria are conflicting, when choosing an alternative with an estimated high profit, decision-makers follow a strategy with an estimated high risk and vice versa. Conclusion: In the present paper, the project portfolio decision-making problem was reduced to a 0-1 knapsack constrained multi-objective optimization problem. The algorithm investigation confirms that the use of the island meta-heuristic significantly improves the performance of genetic algorithms, thereby providing an efficient tool for Financial Responsibility Centres Management.

Evolutionary feature selection for emotion recognition in multilingual speech analysis

In the case when conventional feature selection methods do not demonstrate sufficient performance, alternative algorithmic schemes might be applied. In this paper we propose an evolutionary feature selection technique based on the two-criteria optimization model. To diminish the drawbacks of genetic algorithms, which are used as optimizers, we design a parallel multi-criteria heuristic procedure based on an island model. The effectiveness of the proposed approach was investigated on the Speech-based Emotion Recognition Problem, which reflects one of the crucial aspects in the sphere of human-machine communications. A number of multilingual corpora (German, English and Japanese) were engaged in the experiments. According to the results obtained, a high level of emotion recognition was achieved (up to a 11.15% relative improvement compared with the best F-score value on the full set of attributes).

On Island Model Performance for Cooperative Real-Valued Multi-objective Genetic Algorithms.

Solving a multi-objective optimization problem results in a Pareto front approximation, and it differs from single-objective optimization, requiring specific search strategies. These strategies, mostly fitness assignment, are designed to find a set of non-dominated solutions, but different approaches use various schemes to achieve this goal. In many cases, cooperative algorithms such as island model-based algorithms outperform each particular algorithm included in this cooperation. However, we should note that there are some control parameters of the islands’ interaction and, in this paper, we investigate how they affect the performance of the cooperative algorithm. We consider the influence of a migration set size and its interval, the number of islands and two types of cooperation: homogeneous or heterogeneous. In this study, we use the real-valued evolutionary algorithms SPEA2, NSGA-II, and PICEA-g as islands in the cooperation. The performance of the presented algorithms is compared with the performance of other approaches on a set of benchmark multi-objective optimization problems.