Santiago Muelas

A MOS-based dynamic memetic differential evolution algorithm for continuous optimization: a scalability test

Continuous optimization is one of the areas with more activity in the field of heuristic optimization. Many algorithms have been proposed and compared on several benchmarks of functions, with different performance depending on the problems. For this reason, the combination of different search strategies seems desirable to obtain the best performance of each of these approaches. This contribution explores the use of a hybrid memetic algorithm based on the multiple offspring framework. The proposed algorithm combines the explorative/exploitative strength of two heuristic search methods that separately obtain very competitive results. This algorithm has been tested with the benchmark problems and conditions defined for the special issue of the Soft Computing Journal on Scalability of Evolutionary Algorithms and other Metaheuristics for Large Scale Continuous Optimization Problems. The proposed algorithm obtained the best results compared with both its composing algorithms and a set of reference algorithms that were proposed for the special issue.

Large scale global optimization: Experimental results with MOS-based hybrid algorithms

Continuous optimization is one of the most active research Iines in evolutionary and metaheuristic algorithms. Through CEC 2005 to CEC 2013 competitions, many different algorithms have been proposed to solve continuous problems. The advances on this type of problems are of capital importance as many real-world problems from very different domains (biology, engineering, data mining, etc.) can be formulated as the optimization of a continuous function. In this paper we describe the whole process of creating a competitive hybrid algorithm, from the experimental design to the final statistical validation of the resuIts. We prove that a good experimental design is able to find a combination of algorithms that outperforms any of its composing algorithms by automatically selecting the most appropriate heuristic for each function and search phase. We also show that the proposed algorithm obtains statistically better results than the reference algorithm DECC-G.

A comprehensive comparison of large scale global optimizers

Large Scale Global Optimization is one of the most active research lines in evolutionary and metaheuristic algorithms. In the last five years, several conference sessions and journal special issues have been conducted, and many algorithmic alternatives and hybrid methods, more and more sophisticated, have been proposed. However, most of the proposed algorithms are only evaluated on a particular benchmark of functions and thus its performance in other benchmarks presenting different characteristics remains unknown. In this paper, it is our aim to fill in this gap by evaluating and comparing 10 of the most recently proposed algorithms, in particular, those reporting the best performance in the last major competitions. This paper proposes an evaluation consisting of a broader testbed that considers all the functions of three well-known benchmarks, including a comparative statistical study of the results and the identification of algorithm profiles for those with an equivalent performance. As a part of the comparative analysis this paper also includes three different studies: (1) first, on the complexity of the compared algorithms; (2) then, on the relevance of the comparative statistical tests; and (3) finally, on direct/indirect measures of the exploration/exploitation capabilities of the most representative algorithms in the overall comparison. In addition, this work introduces an open-access web service to perform future analysis and keep trace of new algorithm performances offered to the community of researchers in the field.

A Memetic Differential Evolution Algorithm for Continuous Optimization

Continuous optimization is one of the most active research lines in evolutionary and metaheuristic algorithms. Since CEC 2005 and CEC 2008 competitions, many different algorithms have been proposed to solve continuous problems. Despite there exist very good algorithms reporting high quality results for a given dimension, the scalability of the search methods is still an open issue. Finding an algorithm with competitive results in the range of 50 to 500 dimensions is a difficult achievement. This contribution explores the use of a hybrid memetic algorithm based on the differential evolution algorithm, named MDE-DC. The proposed algorithm combines the explorative/exploitative strength of two heuristic search methods, that separately obtain very competitive results in either low or high dimensional problems. This paper uses the benchmark problems and conditions required for the workshop on “evolutionary algorithms and other metaheuristics for Continuous Optimization Problems – A Scalability Test” chaired by Francisco Herrera and Manuel Lozano.

Segmentation of neuronal nuclei based on clump splitting and a two-step binarization of images

In this paper we present an algorithm to segment the nuclei of neuronal cells in confocal microscopy images, a key technical problem in many experimental studies in the field of neuroscience. We describe the whole procedure, from the original images to the segmented individual nuclei, paying particular attention to the binarization of the images, which is not straightforward due to the technical difficulties related to the visualization of nuclei as individual objects and incomplete and irregular staining. We have focused on the division of clusters of nuclei that appear frequently in these images. Thus we have developed a clump-splitting algorithm to separate touching or overlapping nuclei allowing us to accurate account for both the number and size of the nuclei. The results presented in the paper show that the proposed algorithm performs well on different sets of images from different layers of the cerebral cortex.

A variable neighborhood search algorithm for the optimization of a dial-a-ride problem in a large city

On-demand transportation is becoming a new necessary service for modern (public and private) mobility and logistics providers. Large cities are demanding more and more share transportation services with flexible routes, resulting from user dynamic demands. In this study a new algorithm is proposed for solving the problem of computing the best routes that a public transportation company could offer to satisfy a number of customer requests. In this problem, known in the literature as the dial-a-ride problem, a number of passengers has to be transported between pickup and delivery locations trying to minimize the routing costs while respecting a set of pre-specified constraints (maximum pickup time, maximum ride duration and maximum load per vehicle). For optimizing this problem, a new variable neighborhood search has been developed and tested on a set of 24 different scenarios of a large-scale dial-a-ride problem in the city of San Francisco. The results have been compared against two state-of-the-art algorithms of the literature and validated by means of statistical procedures proving that the new algorithm has obtained the best overall results.

Multiple Offspring Sampling in Large Scale Global Optimization

Continuous optimization is one of the most active research lines in evolutionary and metaheuristic algorithms. Through CEC 2005 to CEC 2011 competitions, many different algorithms have been proposed to solve continuous problems. The advances on this type of problems are of capital importance as many real-world problems from very different domains (biology, engineering, data mining, etc.) can be formulated as the optimization of a continuous function. In this paper we analyze the behavior of a hybrid algorithm combining two heuristics that have been successfully applied to solving continuous optimization problems in the past. We show that the combination of both algorithms obtains competitive results on the proposed benchmark by automatically selecting the most appropriate heuristic for each function and search phase.

Benchmarking a hybrid DE-RHC algorithm on real world problems

Continuous optimization is one of the most active research lines in evolutionary and metaheuristic algorithms. Through CEC 2005 to CEC 2010 competitions, many different algorithms have been proposed to solve continuous problems. The advances on this type of problems are of capital importance as many real-world problems from very different domains (biology, engineering, data mining, etc.) can be formulated as the optimization of a continuous function. For this reason, we have proposed a hybrid DE-RHC algorithm that combines the search strength of Differential Evolution with the explorative ability of a Random Hill Climber, which can help the Differential Evolution algorithm to reach new promising areas in difficult fitness landscapes, such as those than can be found on real-world problems. To evaluate this approach, the benchmark problems proposed in the “Testing Evolutionary Algorithms on Real-world Numerical Optimization Problems” CEC 2011 special session have been considered.

3D segmentations of neuronal nuclei from confocal microscope image stacks

In this paper, we present an algorithm to create 3D segmentations of neuronal cells from stacks of previously segmented 2D images. The idea behind this proposal is to provide a general method to reconstruct 3D structures from 2D stacks, regardless of how these 2D stacks have been obtained. The algorithm not only reuses the information obtained in the 2D segmentation, but also attempts to correct some typical mistakes made by the 2D segmentation algorithms (for example, under segmentation of tightly-coupled clusters of cells). We have tested our algorithm in a real scenario—the segmentation of the neuronal nuclei in different layers of the rat cerebral cortex. Several representative images from different layers of the cerebral cortex have been considered and several 2D segmentation algorithms have been compared. Furthermore, the algorithm has also been compared with the traditional 3D Watershed algorithm and the results obtained here show better performance in terms of correctly identified neuronal nuclei.

Machine Learning to Analyze Migration Parameters in Parallel Genetic Algorithms

Parallel genetic algorithms (PGAs) are a powerful tool to deal with complex optimization problems. Nevertheless, the task of selecting its parameters accurately is an optimization problem by itself. Any additional help or hints to adjust the configuration parameters will lead both towards a more efficient PGA application and to a better comprehension on how these parameters affect optimization behavior and performance. This contribution offers an analysis on certain PGA parameters such as migration frequency, topology, connectivity and number of islands. The study has been carried out on an intensive set of experiments that collect PGA performance on several representative problems. The results have been analyzed using machine learning methods to identify behavioral patterns that are labeled as “good” PGA configurations. This study is a first step to generalize relevant patterns from the problems analyzed that identify better configurations in PGAs.