Alan R. R. de Freitas

A Non-parametric Harmony-Based Objective Reduction Method for Many-Objective Optimization

Multiobjective optimization has been applied successfully to real-world optimization problems with few objectives. However, the performance of current algorithms for multiobjective optimization reduces exponentially as the number of objectives grows. In this paper we present a non-parametric harmony-based approach for objective reduction in order to deal with this issue in many-objective optimization problems. The proposed approach has many advantages such as the independence of the relationship between the objectives as long as they are harmonious and the possibility to visualize conflict and trade-off in the reductions performed.

PSA Based Multi Objective Evolutionary Algorithms

It has generally been acknowledged that both proximity to the Pareto front and a certain diversity along the front, should be targeted when using evolutionary multiobjective optimization. Recently, a new partitioning mechanism, the Part and Select Algorithm (PSA), has been introduced. It was shown that this partitioning allows for the selection of a well-diversified set out of an arbitrary given set, while maintaining low computational cost. When embedded into an evolutionary search (NSGA-II), the PSA has significantly enhanced the exploitation of diversity. In this paper, the ability of the PSA to enhance evolutionary multiobjective algorithms (EMOAs) is further investigated. Two research directions are explored here. The first one deals with the integration of the PSA within an EMOA with a novel strategy. Contrary to most EMOAs, that give a higher priority to proximity over diversity, this new strategy promotes the balance between the two. The suggested algorithm allows some dominated solutions to survive, if they contribute to diversity. It is shown that such an approach substantially reduces the risk of the algorithm to fail in finding the Pareto front. The second research direction explores the use of the PSA as an archiving selection mechanism, to improve the averaged Hausdorff distance obtained by existing EMOAs. It is shown that the integration of the PSA into NSGA-II-I and Δ p -EMOA as an archiving mechanism leads to algorithms that are superior to base EMOAS on problems with disconnected Pareto fronts.

Differential evolution and perceptron decision trees for classification tasks

Due to its predictive capacity and applicability in different fields, classification has been one of the most important tasks in data mining. In this task, the Perceptron Decision Trees (PDT) have been used with good results. Thus, this paper presents a Differential Evolution algorithm that evolves PDTs. Furthermore, we also present the concept of legitimacy which is used to reduce the costs of solution evaluation, a time consuming part of the algorithm. The experiments comparing our method with other seven well known classifiers, show that the proposed approach is competitive and has potential to build very accurate models. The best solutions found by it were the best ones in the majority of the tested databases.

A study on the configuration of migratory flows in island model differential evolution

The Island Model (IM) is a well known multi-population approach for Evolutionary Algorithms (EAs). One of the critical parameters for defining a suitable IM is the migration topology. Basically it determines the Migratory Flows (MF) between the islands of the model which are able to improve the rate and pace of convergence observed in the EAs coupled with IMs. Although, it is possible to find a wide number of approaches for the configuration of MFs, there still is a lack of knowledge about the real performance of these approaches in the IM. In order to fill this gap, this paper presents a thorough experimental analysis of the approaches coupled with the state-of-the-art EA Differential Evolution. The experiments on well known benchmark functions show that there is a trade-off between convergence speed and convergence rate among the different approaches. With respect to the computational times, the results indicate that the increase in implementation complexity does not necessarily represent an increase in the overall execution time.

Differential Evolution and Perceptron Decision Trees for Fault Detection in Power Transformers

Classifying data is a key process for extracting relevant information out of a database. A relevant classification problem is classifying the condition of a transformer based on its chromatography data. It is a useful problem formulation as its solution makes it possible to repair the transformer with less expenditure given that a correct classification of the equipment status is available. In this paper, we propose a Differential Evolution algorithm that evolves Perceptron Decision Trees to classify transformers from their chromatography data. Our approach shows that it is possible to evolve classifiers to identify failure in power transformers with results comparable to the ones available in the literature.

Optimizing two-level reverse distribution networks with hybrid memetic algorithms

In a Two-Level Reverse Distribution Network, products are returned from customers to manufacturers through collection and refurbishing sites. The costs of the reverse chain often overtake the costs of the forward chain by many times. With some known algorithms for the problem as reference, we propose a hybrid memetic algorithm that uses linear programming and a heuristic for defining routes. Moreover, we describe heuristics for deciding locations, algorithms to define routes for the products, and problem-specific genetic operators. Memetic algorithms have returned the best results for all instances.

A study on self-configuration in the differential evolution algorithm

The great development in the area of evolutionary algorithms in recent decades has increased the range of applications of these tools and improved its performance in different fronts. In particular, the Differential Evolution (DE) algorithm has proven to be a simple and efficient optimizer in several contexts. Despite of its success, its performance is closely related to the choice of variation operators and the parameters which control these operators. To increase the robustness of the method and the ease of use for the average user, the pursuit for methods of self-configuration has been increasing as well. There are several methods in the literature for setting parameters and operators. In order to understand the effects of these approaches on the performance of DE, this paper presents a thorough experimental analysis of the main existing paradigms. The results show that simple approaches are able to bring significant improvements to the performance of DE.

Genetic Algorithms Applied to Reverse Distribution Networks

Reverse Distribution Networks are designed to plan the distribution of products from customers to manufacturers. In this paper, we study the problem with two-levels,with products transported from origination points to collection sites before being sent to a refurbishing site. The optimization of reverse distribution networks can reduce the costs of this reverse chain and help companies become more environmentally efficient. In this paper we describe heuristics for deciding locations, algorithms for defining routes, and problem-specific genetic operators. The results of a comparative analysis of 11 algorithms over 25 problem instances suggest that genetic algorithms hybridized with simplex routing algorithms were significantly better than the other approaches tested.

Automatic evaluation methods in evolutionary music: an example with bossa melodies

Evolutionary algorithms need measures of how appropriate a solution is in order to make decisions. This is always a problem for evolving art as codifying aesthetics is a complex task. In this paper we consider the problem of evaluating melodies. The evaluation of melodies in evolutionary music is an open problem that has been tackled by many authors with interactive evaluation, fitness-free genetic algorithms and even neural networks. However, all approaches based on formal analysis of databases or formal music theory have been partial, which is something to be expected for such a complex problem. Thus, we present many metrics that can be used for evaluating melodies and their practical results when applied to a Bossa Nova database of melodies coded by the authors. Although the paper is meant to extend the cycle of possible ideas for evolutionary composers, we argue that there is still much to be developed in this field and each genre of music will always need specific measures of quality.

A New Perspective on Channel Allocation in WLAN: Considering the Total Marginal Utility of the Connections for the Users

The channel allocation problem consists in defining the frequency used by Access Points (APs) in Wireless Local Area Networks (WLAN). An overlap of channels in a WLAN is the major factor of performance reduction for the users in a network. For this reason, we propose a new model for channel allocation that aims to maximize the total quality of the connection of the user by considering their marginal utility. The results show that an allocation model that does not take into account the total utility of each connection tends to prioritize the quality of connection of a few users and lead to a large unbalance in the distribution of connection speed between users. Thus, the new model can handle the importance of degradation caused by the levels of interference in the user connection separately.