Jing-Jing Li

Distributed evolutionary algorithms and their models

Graphical abstractDisplay Omitted HighlightsProvide an updated and systematic review of distributed evolutionary algorithms.Classify the models into population and dimension-distributed groups semantically.Analyze the parallelism, search behaviors, communication costs, scalability, etc.Highlight recent research hotspots in this field.Discuss challenges and potential research directions in this field. The increasing complexity of real-world optimization problems raises new challenges to evolutionary computation. Responding to these challenges, distributed evolutionary computation has received considerable attention over the past decade. This article provides a comprehensive survey of the state-of-the-art distributed evolutionary algorithms and models, which have been classified into two groups according to their task division mechanism. Population-distributed models are presented with master-slave, island, cellular, hierarchical, and pool architectures, which parallelize an evolution task at population, individual, or operation levels. Dimension-distributed models include coevolution and multi-agent models, which focus on dimension reduction. Insights into the models, such as synchronization, homogeneity, communication, topology, speedup, advantages and disadvantages are also presented and discussed. The study of these models helps guide future development of different and/or improved algorithms. Also highlighted are recent hotspots in this area, including the cloud and MapReduce-based implementations, GPU and CUDA-based implementations, distributed evolutionary multiobjective optimization, and real-world applications. Further, a number of future research directions have been discussed, with a conclusion that the development of distributed evolutionary computation will continue to flourish.

Genetic Learning Particle Swarm Optimization

Social learning in particle swarm optimization (PSO) helps collective efficiency, whereas individual reproduction in genetic algorithm (GA) facilitates global effectiveness. This observation recently leads to hybridizing PSO with GA for performance enhancement. However, existing work uses a mechanistic parallel superposition and research has shown that construction of superior exemplars in PSO is more effective. Hence, this paper first develops a new framework so as to organically hybridize PSO with another optimization technique for “learning.” This leads to a generalized “learning PSO” paradigm, the *L-PSO. The paradigm is composed of two cascading layers, the first for exemplar generation and the second for particle updates as per a normal PSO algorithm. Using genetic evolution to breed promising exemplars for PSO, a specific novel *L-PSO algorithm is proposed in the paper, termed genetic learning PSO (GL-PSO). In particular, genetic operators are used to generate exemplars from which particles learn and, in turn, historical search information of particles provides guidance to the evolution of the exemplars. By performing crossover, mutation, and selection on the historical information of particles, the constructed exemplars are not only well diversified, but also high qualified. Under such guidance, the global search ability and search efficiency of PSO are both enhanced. The proposed GL-PSO is tested on 42 benchmark functions widely adopted in the literature. Experimental results verify the effectiveness, efficiency, robustness, and scalability of the GL-PSO.

Renumber strategy enhanced particle swarm optimization for cloud computing resource scheduling

Cloud computing offers unprecedented capacity to execute large-scale workflows in the “era of big data”. In 2014, a cost-minimization and deadline-constrained workflow scheduling (CMDCWS) model is firstly proposed by Rodriguez and Buyya, which is applicable for the business need of cloud computing that a workflow task should be finished by minimizing the execute cost within a deadline constraint. As scheduling cloud computing resources for workflow is an NP-hard problem, Rodriguez and Buyya proposed to use particle swarm optimization (PSO) to solve the CMDCWS problem. In traditional PSO for CMDCWS, each dimension in the particle position stands for each task and the value of the corresponding dimension stands for the index of the cloud resource that executes this task. However, this may have drawback because the value of each dimension does not relate to the resource characteristic but is only a meaningless index number. Therefore the learning behaviors among the particles do not make sense because learning from index number may not lead to better position. In this paper, we present a resource renumber strategy to encode the particle position and design a renumber PSO (RNPSO) for CMDCWS. In RNPSO, all the resources are re-ordered and re-numbered according to their computational ability, i.e., the cost per unit time. By this, the values of particle position can make sense and the positions difference between the well-performed and poorly-performed particles can guide poorly-performed particle to promising region. We conduct experiments on test cases with small, middle, and large scales to compare the performance of PSO and RNPSO. The results show that the resource renumber strategy is promising for enhancing the PSO performance.

Parallel Particle Swarm Optimization Using Message Passing Interface

Parallel computation is an efficient way to combine the advantages of different computation paradigms to obtain promising solution. In order to analyze the performance of parallel computation techniques to the particle swarm optimization (PSO) algorithm, a parallel particle swarm optimization (PPSO) is proposed in this paper. Since the theorem of “no free lunch” exists, there is not an optimization algorithm that can perfectly tackle all problems. The PPSO provides a paradigm to combine different variants of PSO algorithms by using the Message Passing Interface (MPI) so that the advantages of diverse PSO algorithms can be utilized. The PPSO divides the whole evolution process into several stages. At the interval between two successive stages, each PSO algorithm exchanges the achievement of their evolution and then continues with the next stage of evolution. By merging the global model PSO (GPSO), the local model PSO (LPSO), the bare bone PSO (BPSO), and the comprehensive learning PSO (CLPSO), the PPSO achieves higher solution quality than the serial version of these four PSO algorithms, according to the simulation results on benchmark functions.

Adaptive particle swarm optimization with variable relocation for dynamic optimization problems

This paper proposes to solve the dynamic optimization problem (DOP) by using an adaptive particle swarm optimization (APSO) algorithm with an variable relocation strategy (VRS). The VRS based APSO algorithm (APSO/VRS) has the following two advantages when solving DOP. Firstly, by using the APSO optimizing framework, the algorithm benefits from the fast optimization speed due to the adaptive parameter control. More importantly, the adaptive parameter and operator in APSO make the algorithm fast respond to the environment changes of DOP. Secondly, VRS was reported in the literature to help dynamic evolutionary algorithm (DEA) to relocate the individual position in promising region when environment changes. Therefore, the modified VRS used in APSO can collect historical information in the stability stage and use such information to guide the particle variable relocation in the change stage. We evaluated both APSO and APSO/VRS on several dynamic benchmark problems and compared with two state-of-the-art DEAs and DEA that also used the VRS. The results show that both APSO and APSO/VRS can obtain very competitive results on these problems, and APSO/VRS outperforms others on most of the test cases.

Tournament Selection Based Artificial Bee Colony Algorithm with Elitist Strategy

Artificial bee colony (ABC) algorithm is a novel heuristic algorithm inspired from the intelligent behavior of honey bee swarm. ABC algorithm has a good performance on solving optimization problems of multivariable functions and has been applied in many fields. However, traditional ABC algorithm chooses solutions on the onlooker stage with roulette wheel selection (RWS) strategy which has several disadvantages. Firstly, RWS is suitable for maximization optimization problem. The fitness value has to be converted when solving minimization optimization problem. This makes RWS difficult to be generally used in real-world applications. Secondly, RWS has no any parameter that can control the selection pressure. Therefore, RWS is not easy to adapt to various optimization problems. This paper proposes a tournament selection based ABC (TSABC) algorithm to avoid these disadvantages of RWS based ABC. Moreover, this paper proposes an elitist strategy that can be applied to traditional ABC, TSABC, and any other ABC variants, so as to avoid the phenomenon that ABC algorithm may abandon the globally best solution in the scout stage. We compare the performance of traditional ABC and TSABC on a set of benchmark functions. The experiment results show that TSABC is more flexible and can be efficiently adapted to solve various optimization problems by controlling the selection pressure.

Enhancing distributed differential evolution with a space-driven topology

Differential evolution (DE) is a simple and efficient evolutionary algorithm for global optimization. In distributed differential evolution (DDE), the population is divided into several sub-populations and each sub-population evolves independently for enhancing population diversity as well as algorithmic performance. Sub-populations in DDE share their elite individuals with neighborhood through a predefined migration topology. However, the construction of traditional migration topologies does not consider the position information of sub-populations in the search space. The position information is helpful in controlling the degree of diversity between the sub-populations and their migrated individuals. A proper degree of diversity could promote the balance between exploration and exploitation for DDE algorithms. To achieve this target, a dynamic space-driven migration topology is proposed in this paper. The proposed topology is constructed and updated according to the distances between sub-populations. Based on this proposed topology, some sub-populations receive diverse individuals from neighborhood far away while others communicate with neighborhood nearby. Numerical experiments have been performed on 13 diverse test functions. Results verify the advantage of DDE with the proposed migration topology compared to those with classic topologies.

A splicing-driven memetic algorithm for reconstructing cross-cut shredded text documents

Graphical abstractDisplay Omitted HighlightsDevelop a splicing-driven memetic algorithm to reconstruct cross-cut shredded text documents.Design a comprehensive cost function to evaluate solutions and to guide individual search.Design novel reproduction operators to effectively utilize the adjacency information of shreds.Propose an elitism-based local search strategy to further enhance efficiency.Obtain good reconstruction performance in terms of the solution quality and convergence speed. Reconstruction of cross-cut shredded text documents (RCCSTD) plays a crucial role in many fields such as forensic and archeology. To handle and reconstruct the shreds, in addition to some image processing procedures, a well-designed optimization algorithm is required. Existing works adopt some general methods in these two aspects, which may not be very efficient since they ignore the specific structure or characteristics of RCCSTD. In this paper, we develop a splicing-driven memetic algorithm (SD-MA) specifically for tackling the problem. As the name indicates, the algorithm is designed from a splicing-centered perspective, in which the operators and fitness evaluation are developed for the purpose of splicing the shreds. We design novel crossover and mutation operators that utilize the adjacency information in the shreds to breed high-quality offsprings. Then, a local search strategy based on shreds is performed, which further improves the evolution efficiency of the population in complex search space. To extract valid information from shreds and improve the accuracy of splicing costs, we propose a comprehensive objective function that considers both edge and empty row-based splicing errors. Experiments are carried out on 30 RCCSTD scenarios and comparisons are made against previous best-known algorithms. Experimental results show that the proposed SD-MA displays a significantly improved performance in terms of solution accuracy and convergence speed.

Generating Software Test Data by Particle Swarm Optimization

Search-based method using meta-heuristic algorithms is a hot topic in automatic test data generation. In this paper, we develop an automatic test data generating tool named particle swarm optimization data generation tool PSODGT. The PSODGT is characterized by the following two features. First, the PSODGT adopts the condition-decision coverage C/DC as the criterion of software testing, aiming to build an efficient test data set that covers all conditions. Second, the PSODGT uses a particle swarm optimization PSO approach to generate test data set. In addition, a new position initialization technique is developed for PSO. Instead of initializing the test data randomly, the proposed technique uses the previously-found test data that can reach the target condition as the initial positions so that the search speed of PSODGT can be further accelerated. The PSODGT is tested on four practical programs. Experimental results show that the proposed PSO approach is promising.

Fast multiple human detection with neighborhood-based speciation differential evolution

Human detection plays a crucial role in a number of real world applications. Because of the popularity of smart car, Virtual Reality (VR) and other applications, strong demand of real-time detecting rises. The efficiency of a human detection algorithm becomes more crucial than ever before. In this work, a novel human detection framework combining the Histograms of Oriented Gradients (HOG) feature, Support Vector Machine and Neighborhood-based Speciation Differential Evolution (NSDE), is proposed in consideration of fast and accurate detection. Instead of inefficiently traversing and grouping all of the detecting windows as the conventional method, HOG-SVM-NSDE framework searches the whole image in a heuristic way with the unique niching strategy. Experiment results show that the HOG-SVM-NSDE framework achieves a favorable efficiency while still maintains a practical accuracy.