Real-coded Evolutionary Algorithm Research Articles

Deep architectures for the segmentation of frontal sinuses in X-ray images: Towards an automatic forensic identification system in comparative radiography

Comparative radiography (CR) is the forensic anthropology technique in which ante-mortem (AM) and post-mortem (PM) radiographic materials (e.g., X-ray images or CTs) are compared in order to determine the identity of a human being (either by providing a positive identification or by reducing the number of candidates). One of the most commonly used anatomical structures in CR are the frontal sinuses, which are used in forensic human identification tasks due to their singularity and high identification power. In order to automate the comparison of frontal sinuses in AM and PM materials, we need to perform their segmentation and registration. However, these tasks are time-consuming, subjective, and complex. This paper presents the first CR-based identification system for the comparison of frontal sinuses and supporting the forensic expert’s decision. The proposed system comprises two building blocks. First, a frontal sinus segmentation method in X-ray images using deep convolutional neural networks. Different convolutional architectures are compared in solving the segmentation problem, obtaining an average Dice Similarity Coefficient of the frontal sinuses of 0.823 on a dataset composed of 234 skull radiographs. Second, an evolutionary-based 2D-3D IR method, that searches for the best alignment of segmented AM and PM images using a real-coded evolutionary algorithm. The proposed system is evaluated on a real multiple-comparison identification scenario including 10 X-ray images and 10 CTs, where manual and automatic segmentation approaches are compared. The global results shows that the proposed system is able to filter 50% of the sample. These preliminary results suggest that our system can reliably keep the true positive identity in the first half of the sample, allowing for a significant reduction of forensic experts’ workload and shortening identification times.

Performance analysis of real-coded evolutionary algorithms under a computationally expensive optimization scenario: 3D–2D Comparative Radiography

Real-coded evolutionary algorithms have solved numerous real-world optimization problems. In this work, we aim to analyze the behavior and robustness of several real-coded evolutionary algorithms from the state of the art in a challenging real world optimization problem. This optimization problem consists on the superimposition of 3D and 2D images of skeletal structures (i.e. bones and cavities) based on their silhouette. This task is required for the automation of a forensic identification technique known as comparative radiography, via the generation of the best projection of the 3D image with respect to the 2D image. This superimposition problem was tackled in a recent proposal using an evolutionary 3D–2D image registration method based on differential evolution. However, the results obtained were insufficient for its use in real scenarios, due to: (1) the complexity and multi-modality of search space, despite the reduced number of parameters to be optimized (7 in its simple version and 9 in a more complex one, proposed in this work); and (2) the high computational cost of generating and evaluating a superimposition. Particularly, we have performed a rigorous comparative study of six state-of-the-art real-coded evolutionary algorithms (DE, L-SHADE, CMA-ES, BIPOP-CMA-ES, CRO-SL, and MVMO-SH) with synthetic images of three forensic anatomical structures (frontal sinuses, clavicles, and patellae), showing that the best results are always obtained by MVMO-SH in terms of precision, robustness and computational cost. Furthermore, we have validated the quality of the superimpositions obtained by the evolutionary image registration method using MVMO-SH with real images of frontal sinuses. We have performed the comparison of 50 head radiographs and 50 3D images, resulting in 2,500 cross-comparisons (50 positive and 2,450 negatives). The obtained results are promising since the superimpositions obtained allowed us to filter out 88% of the possible candidates with 0 error rate in a fully automatic manner, showing the high quality of the superimposition obtained.

Optimization of Steiner trees search in the flow Steiner tree problem

For the flow Steiner tree problem on a transport network, a two-level composite evolutionary algorithm was developed. At the upper level a tree topology is searched; at the low level optimal Steiner points’ coordinates are searched for the three with a given topology. In this paper for solving the latter problem, a real-coded evolutionary algorithm is suggested. The algorithm uses an ontogenesis model and a special maturation operator. The operator is based on a random modification of the coordinate-wise-decent method adapted for the flow problem.

Quantum seeded evolutionary computational technique for constrained optimization in engineering design and manufacturing

In this paper an attempt is made to develop a new Quantum Seeded Hybrid Evolutionary Computational Technique (QSHECT) that is general, flexible and efficient in solving single objective constrained optimization problems. It generates initial parents using quantum seeds. It is here that QSHECT incorporates ideas from the principles of quantum computation and integrates them in the current framework of Real Coded Evolutionary Algorithm (RCEA). It also incorporates Simulated Annealing (SA) in the selection process of Evolutionary Algorithm (EA) for child generation. The proposed algorithm has been tested on standard test problems and engineering design problems taken from the literature. In order to test this algorithm on domain-specific manufacturing problems, Neuro-Fuzzy (NF) modeling of hot extrusion is attempted and the NF model is incorporated as a fitness evaluator inside the QSHECT to form a new variant of this technique, i.e. Quantum Seeded Neuro Fuzzy Hybrid Evolutionary Computational Technique (QSNFHECT) and is effectively applied for process optimization of hot extrusion process. The neuro-fuzzy model (NF) is also compared with statistical regression analysis (RA) model for evaluating the extrusion load. The NF model was found to be much superior. The optimal process parameters obtained by Quantum Seeded Neuro Fuzzy Hybrid Evolutionary Technique (QSNFHECT) are validated by the finite element model. The proposed methodology using QSNFHECT is a step towards meeting the challenges posed in intelligent manufacturing systems and opens new avenues for parameter estimation and optimization and can be easily incorporated in existing manufacturing setup.

K-Bit-Swap: a new operator for real-coded evolutionary algorithms

There have been a variety of crossover operators proposed for real-coded genetic algorithms (RCGAs). Such operators recombine values from pairs of strings to generate new solutions. In this article, we present a recombination operator for RCGAs that selects the string locations for change separately randomly in the parent and offspring, enabling solution parts to move within a string, and compare it to mainstream crossover operators in a set of experiments on a range of standard multidimensional optimization problems and a real-world clustering problem. We present two variants of the operator, either selecting bits uniformly at random in both strings or sampling the second bit from a normal distribution centered at the selected location in the first string. While the operator is biased toward exploitation of fitness space, the random selection of the second bit for swapping reduces this bias slightly. Statistical analysis of the experimental results using a nonparametric test shows the advantage of the new recombination operators on our test optimization functions.

Partition based real-valued encoding scheme for evolutionary algorithms

Encoding feasible solutions is one of the most important aspects to be taken into account in the field of evolutionary computation in order to solve search or optimization problems. This paper proposes a new encoding scheme for real-coded evolutionary algorithms. It is called partition based encoding scheme, and satisfies two restrictions. Firstly, each of the components of a decoded vector that conforms a candidate solution to a problem at hand belongs to a predefined interval. Secondly, the sum of the components of each of these decoded vectors is always equal to a predefined constant. The proposed encoding scheme inherently guarantees these constraints for all the individuals that are generated within the evolution process as a consequence of applying the genetic operators. Partition based encoding scheme is successfully applied to learning conditional probability tables for a given discrete Bayesian network topology, where each row of the tables must exactly add up to one, and the components of each row belong to the interval [0,1] as they are probability values. The results given by the proposed encoding system for this learning problem is compared to a deterministic algorithm and another evolutionary approach. Better results are shown in terms of accuracy with respect to the former one, and accuracy and convergence speed with respect to the later one.

Some modifications of low-dimensional simplex evolution and their convergence

Low-dimensional simplex evolution (LDSE) is a real-coded evolutionary algorithm for global optimization. In this paper, we introduce three techniques to improve its performance: low-dimensional reproduction (LDR), normal struggle (NS) and variable dimension (VD). LDR tries to preserve the elite by keeping some of its (randomly chosen) components. LDR can also help the offspring individuals to escape from the hyperplane determined by their parents. NS tries to enhance its local search capability by allowing unlucky individual search around the best vertex of m-simplex. VD tries to draw lessons from recent failure by making further exploitation on its most promising sub-facet. Numerical results show that these techniques can improve the efficiency and reliability of LDSE considerably. The convergence properties are then analysed by finite Markov chains. It shows that the original LDSE might fail to converge, but modified LDSE with the above three techniques will converge for any initial population. To evaluate the convergence speed of modified LDSE, an estimation of its first passage time (of reaching the global minimum) is provided.

Modeling dynamics of a real-coded CHC algorithm in terms of dynamical probability distributions

Some theoretical models have been proposed in the literature to predict dynamics of real-coded evolutionary algorithms. These models are often applied to study very simplified algorithms, simple real-coded functions or sometimes these make difficult to obtain quantitative measures related to algorithm performance. This paper, trying to reduce these simplifications to obtain a more useful model, proposes a model that describes the behavior of a slightly simplified version of the popular real-coded CHC in multi-peaked landscape functions. Our approach is based on predicting the shape of the search pattern by modeling the dynamics of clusters, which are formed by individuals of the population. This is performed in terms of dynamical probability distributions as a basis to estimate its averaged behavior. Within reasonable time, numerical experiments show that is possible to achieve accurate quantitative predictions in functions of up to 5D about performance measures such as average fitness, the best fitness reached or number of fitness function evaluations.

A GA–DE hybrid evolutionary algorithm for path synthesis of four-bar linkage

Abstract A new real-coded evolutionary algorithm is proposed for application to path synthesis of a four-bar linkage. This new evolutionary algorithm is obtained by combining differential evolution (DE) with the real-valued genetic algorithm (RGA). We term this the “GA–DE hybrid algorithm.” The only difference between the proposed algorithm and RGA is in the content of the crossover. The crossover operation in the RGA is replaced by differential vector perturbation, with the best individual or some excellent individuals as the base vectors. Thus, both the main perturbation of differential vectors and the minor perturbation of mutation are used as genetic operators in the GA–DE hybrid algorithm. The efficiency and accuracy of the proposed method are tested using four cases. Findings show that much more accurate solutions for three cases are obtained with this method than those obtained using other evolutionary methods as discussed in the literature. A moveable stick diagram of the synthesized mechanisms can be obtained using the 2D sketch feature of SolidWorks®. This can be used to check whether the synthesized mechanisms encounter circuit defects or are incapable of motion.

An experimental study on the applicability of evolutionary algorithms to craniofacial superimposition in forensic identification

Photographic supra-projection is a forensic process that aims to identify a missing person from a photograph and a skull found. One of the crucial tasks throughout all this process is the craniofacial superimposition which tries to find a good fit between a 3D model of the skull and the 2D photo of the face. This photographic supra-projection stage is usually carried out manually by forensic anthropologists. It is thus very time consuming and presents several difficulties. In this paper, we aim to demonstrate that real-coded evolutionary algorithms are suitable approaches to tackle craniofacial superimposition. To do so, we first formulate this complex task in forensic identification as a numerical optimization problem. Then, we adapt three different evolutionary algorithms to solve it: two variants of a real-coded genetic algorithm and the state of the art evolution strategy CMA-ES. We also consider an existing binary-coded genetic algorithm as a baseline. Results on several superimposition problems of real-world identification cases solved by the Physical Anthropology lab at the University of Granada (Spain) are considered to test our proposals.

Rethinking the design of real-coded evolutionary algorithms: Making discrete choices in continuous search domains

Although real-coded evolutionary algorithms (EAs) have been applied to optimization problems for over thirty years, the convergence properties of these methods remain poorly understood. We discuss the use of discrete random variables to perform search in real-valued EAs. Although most real-valued EAs perform mutation with continuous random variables, we argue that EAs using discrete random variables for mutation can be much easier to analyze. In particular, we present and analyze two simple EAs that make discrete choices of mutation steps.

A robust real-coded evolutionary algorithm with toroidal search space conversion

This paper presents a robust Real-coded evolutionary algorithm. Real-coded evolutionary algorithms (RCEAs), such as real-coded genetic algorithms and evolution strategies, are known as effective methods for function optimization. However, they often fail to find the optimum in case the objective function is multimodal, discrete or high-dimensional. It is also reported that most crossover (or recombination) operators for RCEAs has sampling bias that prevents to find the optimum near the boundary of search space. They like to search the center of search space much more than the other. Therefore, they will not work on functions that have their optima near the boundary of search space. In this paper, we apply two methods, genetic algorithm with search area adaptation (GSA) and toroidal search space conversion (TSC), to the function optimization for improving the robustness of RCEAs. The former method searches adaptively and the latter one removes the sampling bias. Through several experiments, we have confirmed that GSA works adaptively and it shows higher performance, and RCEAs with TSC show effectiveness to find the optima near the boundary of search space.

Real-Parameter Evolutionary Monte Carlo With Applications to Bayesian Mixture Models

We propose an evolutionary Monte Carlo algorithm to sample from a target distribution with real-valued parameters. The attractive features of the algorithm include the ability to learn from the samples obtained in previous steps and the ability to improve the mixing of a system by sampling along a temperature ladder. The effectiveness of the algorithm is examined through three multimodal examples and Bayesian neural networks. The numerical results confirm that the real-coded evolutionary algorithm is a promising general approach for simulation and optimization.