Mutation Operator Research Articles

Research on the Digital Transformation of the Enterprise Marketing Ecosystem Based on Genetic Algorithms

This paper examines the current state of the Enterprise Marketing Ecosystem (EME), especially in the context of Digital Transformation (DT), highlighting both the challenges and opportunities it presents. A significant focus is placed on the potential of genetic algorithms to address the complex optimization problems inherent in this transformation. The core issues faced by enterprises in their digital marketing evolution are elaborated, setting the stage for the exploration of genetic algorithms in this domain. The research delves into the specific steps of the proposed method, utilizing the multi-objective genetic algorithm NSGA-II to optimize various aspects of market strategies. The methodology includes the generation of an initial population, evaluation of fitness according to the tailored needs of the marketing ecosystem, selection, crossover, and mutation operations, followed by the formation of a new generation of solutions. The termination criteria of the algorithm are also discussed. Experimental results are presented, demonstrating the effectiveness of the genetic algorithm in enhancing the DT of the EME. Comparisons with traditional approaches reveal significant improvements in various key performance metrics, underscoring the algorithm’s capability in navigating the complexities of digital marketing optimization.

Sequential Task Allocation of More Scalable Artificial Dragonfly Swarms Considering Dubins Trajectory

With the rapid advancement of UAV technology and the increasing complexity of tasks, multi-UAV systems face growing challenges in task execution. Traditional task allocation algorithms often perform poorly when dealing with issues such as local optima, slow convergence speed, and low convergence accuracy, making it difficult to meet the demands for efficiency and practicality in real-world applications. To address these problems, this paper focuses on collaborative task allocation technology for multi-UAV. It proposes a collaborative task allocation strategy for multi-UAV in a multi-target environment, which comprehensively considers various complex constraints in practical application scenarios. The strategy utilizes Dubins curves for trajectory planning and constructs a multi-UAV collaborative task allocation model, with targets including the shortest total distance index, the minimum time index, and the trajectory coordination index. Each UAV is set as an artificial dragonfly by modifying the traditional dragonfly algorithm, incorporating differential evolution algorithms and their crossover, mutation, and selection operations to bring UAV swarms closer to the characteristics of biological dragonflies. The modifications can enhance the global scalability of artificial dragonfly swarms (ADSs), including wider search capacity, wider speed range, and more diverse search accuracy. Meanwhile, potential solutions with global convergence properties are stored to better support real-time adjustments to task allocation. The simulation results show that the proposed strategy can generate a conflict-free task execution scheme and plan the trajectory, which has advantages in changing the data scale of the UAV and the target and improves the reliability of the system to a certain extent.

U-AEFA: Online and offline learning-based unified artificial electric field algorithm for real parameter optimization

Optimization problems in real-world scenarios require algorithms that effectively balance exploration and exploitation to avoid local optima and achieve global solutions. To address this, we propose a unified artificial electric field algorithm (U-AEFA) that integrates both offline learning (inspired by high-performing metaheuristic algorithms) and online learning (through historical data generated during evolution). U-AEFA introduces a unique three-layer population structure to enhance search efficiency, consisting of the best agent in the first layer, the top-performing agents in the second layer, and the remaining agents in the third layer. Key features of U-AEFA include (i) an effective Coulomb’s constant for improved exploration, (ii) a non-uniform mutation operator to mitigate premature convergence, and (iii) two acceleration coefficients for enhanced performance. These three factors constitute offline learning and have been implemented to improve different design elements of the algorithm. As part of online learning, it employs a difference vector reuse (DVR) strategy to evolve the first-layer agents. The algorithm is evaluated using the CEC 2017 test suite across multiple dimensions (10, 30, 50, 100), where it consistently outperforms seven state-of-the-art algorithms, demonstrating superior accuracy and convergence speed. Moreover, U-AEFA’s robustness is validated on 12 high-dimensional feature selection problems, further highlighting its effectiveness in solving complex optimization tasks. The source code of U-AEFA is available at

A novel memory usage guided fuzzing based on particle swarm optimization

Fuzzing has become the focus of research in the field of software testing due to its advantages such as simple testing process, easy deployment, and easy reproduction of crashes. However, fuzzing also suffers from the disadvantages of poor test purpose and invalid generated seeds. To solve the above problems, researchers proposed the Memory Usage Guided Fuzzing (MUGF). To further optimize the performance of MUGF, this paper proposes a fuzzing method named Memory Usage Guided Fuzzing based on Particle Swarm Optimization (MUGF-PSO). MUGF-PSO will guide the selection of mutation operators for subsequent testing by learning the effectiveness of each mutation operators during previous testing. Specifically, MUGF-PSO regards each mutation operator as a particle in the particle swarm optimization algorithm, and the coverage change is regarded as an important factor to evaluate the seed’s local and global best position. In addition, the efficient selection probability distribution of mutation operator is constantly iteratively searched to make the MUGF tends to select the mutation operator that is more able to trigger new coverage. Furthermore, the MUGF-PSO is integrated into MemLock developed by MUGF, that is, MemLock-PSO is developed. We conduct a comparison experiment with 5 fuzzers (including AFL, MemLock, EcoFuzz, HavocMAB and Darwin), and the results show that MemLock-PSO is able to find more paths and crashes in 12 widely used program with different functions from 10 tools, with a significant gap of around 12 h. Meanwhile, the MemLock-PSO have a significant difference in the number of paths and crashes compared to other fuzzers with a good stability. Our work also proves the remarkable significance of applying swarm intelligence optimization algorithms in fuzzing in order to solve the problem of selection of mutation operators.

Evolutionary stochastic configuration networks for industrial data analytics

Stochastic configuration network (SCN) with compact architecture is expected for data modeling. However, the hidden-node parameters (HNPs) randomly configured may result in a slow learning process due to the redundant nodes embedded in the model. To resolve this problem, an evolutionary SCN based on an improved differential evolution (DE) algorithm is proposed in this paper. Specifically, the improved DE reuses the assignment information of last hidden node to find an appropriate search scope for the current one; employs a space reduction method to seed a promising population in the scope; and develops a performance-aware scheme to adjust the scale factor of mutation operators. The proposed evolutionary SCNs are compared with other methods on six datasets and then applied for two real-world applications. Experimental results demonstrate that the proposed method obtains superior performance in terms of compactness and accuracy, with great potential for real-world data analysis.

The Application of an Improved LESS Dung Beetle Optimization in the Intelligent Topological Reconfiguration of ShipPower Systems

To address the shortcomings of the Dung Beetle Optimization (DBO) algorithm in ship power-system fault reconfiguration, such as low population diversity and an imbalance between global exploration and local exploitation, the authors of this paper propose an improved Dung Beetle Optimization (LESSDBO) algorithm. The improvements include optimizing the initial population using Latin hypercube sampling and an elite population strategy, optimizing parameters with an improved sigmoid activation function, introducing the sine–cosine algorithm (SCA) for position update optimization, and performing multi-population mutation operations based on individual quality. The LESSDBO algorithm was applied to simulate the fault reconfiguration of a ship power system, and it was compared with the traditional DBO, Genetic Algorithm (GA), and Modified Particle Swarm Optimization (MSCPSO) methods. The simulation results showed that LESSDBO outperformed the other algorithms in terms of convergence accuracy, convergence speed, and global search capability. Specifically, in the reconfiguration under Fault 1, LESSDBO achieved optimal convergence in seven iterations, reducing convergence iterations by more than 30% compared with the other algorithms. In the reconfiguration under Fault 2, LESSDBO achieved optimal convergence in eight iterations, reducing convergence iterations by more than 23% compared with the other algorithms. Additionally, in the reconfiguration under Fault Condition 1, LESSDBO achieved a minimum of four switch actions, which is 33% fewer than the other algorithms, on average. In the reconfiguration under Fault Condition 2, LESSDBO achieved a minimum of eight switch actions, which is a 5.9% reduction compared with the other algorithms. Furthermore, LESSDBO obtained the optimal reconfiguration solution in all 50 trials for both Faults 1 and 2, demonstrating a 100% optimal convergence probability and significantly enhancing the reliability and stability of the algorithm. The proposed method effectively overcomes the limitations of the traditional DBO in fault reconfiguration, providing an efficient and stable solution for the intelligent topology reconfiguration of ship power systems.

A New Hybrid Improved Arithmetic Optimization Algorithm for Solving Global and Engineering Optimization Problems

The Arithmetic Optimization Algorithm (AOA) is a novel metaheuristic inspired by mathematical arithmetic operators. Due to its simple structure and flexible parameter adjustment, the AOA has been applied to solve various engineering problems. However, the AOA still faces challenges such as poor exploitation ability and a tendency to fall into local optima, especially in complex, high-dimensional problems. In this paper, we propose a Hybrid Improved Arithmetic Optimization Algorithm (HIAOA) to address the issues of susceptibility to local optima in AOAs. First, grey wolf optimization is incorporated into the AOAs, where the group hunting behavior of GWO allows multiple individuals to perform local searches at the same time, enabling the solution to be more finely tuned and avoiding over-concentration in a particular region, which can improve the exploitation capability of the AOA. Second, at the end of each AOA run, the follower mechanism and the Cauchy mutation operation of the Sparrow Search Algorithm are selected with the same probability and perturbed to enhance the ability of the AOA to escape from the local optimum. The overall performance of the improved algorithm is assessed by selecting 23 benchmark functions and using the Wilcoxon rank-sum test. The results of the HIAOA are compared with other intelligent optimization algorithms. Furthermore, the HIAOA can also solve three engineering design problems successfully, demonstrating its competitiveness. According to the experimental results, the HIAOA has better test results than the comparator.

Model Optimization and Application of Straw Mulch Quantity Using Remote Sensing

Straw mulch quantity is an important indicator in the detection of straw returned to the field in conservation tillage, but there is a lack of large-scale automated measurement methods. In this study, we estimated global straw mulch quantity and completed the detection of straw returned to the field. We used an unmanned aerial vehicle (UAV) carrying a multispectral camera to acquire remote sensing images of straw in the field. First, the spectral index was selected using the Elastic-net (ENET) algorithm. Then, we used the Genetic Algorithm Hybrid Particle Swarm Optimization (GA-HPSO) algorithm, which embeds crossover and mutation operators from the Genetic Algorithm (GA) into the improved Particle Swarm Optimization (PSO) algorithm to solve the problem of machine learning model prediction performance being greatly affected by parameters. Finally, we used the Monte Carlo method to achieve a global estimation of straw mulch quantity and complete the rapid detection of field plots. The results indicate that the inversion model optimized using the GA-HPSO algorithm performed the best, with the coefficient of determination (R2) reaching 0.75 and the root mean square error (RMSE) only being 0.044. At the same time, the Monte Carlo estimation method achieved an average accuracy of 88.69% for the estimation of global straw mulch quantity, which was effective and applicable in the detection of global mulch quantity. This study provides a scientific reference for the detection of straw mulch quantity in conservation tillage and also provides a reliable model inversion estimation method for the estimation of straw mulch quantity in other crops.

Enhancing network robustness with structural prior and evolutionary techniques

Robustness optimization in complex networks is a critical research area due to its implications for the reliability and stability of various systems. However, existing algorithms encounter two key challenges: the lack of integration of prior network knowledge, leading to suboptimal solutions, and high computational costs, which hinder their practical application. To address these challenges, this paper introduces Eff-R-Net, an efficient evolutionary algorithm framework aimed at enhancing the robustness of complex networks through accelerated evolution. Eff-R-Net leverages global and local network information, featuring a novel three-part composite crossover operator. Prior network knowledge is incorporated in mutation and local search operators to expedite the construction of networks with superior robustness. Additionally, a simplified method for calculating robustness enhances efficiency, while adaptive hyper-parameters dynamically adjust operators execution probabilities for optimal evolution. Extensive evaluations on both synthetic (Scale-Free, Erdös-Rényi, and Small-World) and three infrastructure real-world networks demonstrate the superiority of Eff-R-Net. The algorithm improves robustness by 12.8% and reduces computational time by 25.4% compared to state-of-the-art algorithm in real-world network experiments. These findings underscore Eff-R-Net's versatility and potential in enhancing network robustness across different domains.

Multi-UAV Cooperative Pursuit of a Fast-Moving Target UAV Based on the GM-TD3 Algorithm

Recently, developing multi-UAVs to cooperatively pursue a fast-moving target has become a research hotspot in the current world. Although deep reinforcement learning (DRL) has made a lot of achievements in the UAV pursuit game, there are still some problems such as high-dimensional parameter space, the ease of falling into local optimization, the long training time, and the low task success rate. To solve the above-mentioned issues, we propose an improved twin delayed deep deterministic policy gradient algorithm combining the genetic algorithm and maximum mean discrepancy method (GM-TD3) for multi-UAV cooperative pursuit of high-speed targets. Firstly, this paper combines GA-based evolutionary strategies with TD3 to generate action networks. Then, in order to avoid local optimization in the algorithm training process, the maximum mean difference (MMD) method is used to increase the diversity of the policy population in the updating process of the population parameters. Finally, by setting the sensitivity weights of the genetic memory buffer of UAV individuals, the mutation operator is improved to enhance the stability of the algorithm. In addition, this paper designs a hybrid reward function to accelerate the convergence speed of training. Through simulation experiments, we have verified that the training efficiency of the improved algorithm has been greatly improved, which can achieve faster convergence; the successful rate of the task has reached 95%, and further validated UAVs can better cooperate to complete the pursuit game task.

Optimisation of underwater acoustic sensor networks

The oceans are the scene of intense economical and industrial activities. It is becoming even more competitive with recent military activities and the upcoming development of marine renewable energies. All of these activities have a significant impact on marine ecosystems and must be carefully evaluated for any maritime planning. The marine environment is also a very effective medium for the propagation of acoustic waves over long distances, depending on the frequency, the source level, the static (bathymetry) and the dynamic (hydrology) oceanographic conditions. Such conditions make acoustic sensors primary candidates for underwater investigations and the distribution of mixed sensors with various sensitivities, frequency bandwidths or positioning can be optimized for detection, localization or monitoring purposes. This contribution aims to present a methodology based on evolutionary algorithms which considers the overall physical state of the marine environment and determines the optimal configurations of mixed underwater acoustic networks. The relevance of various objectives, or metrics, can be discussed, as well as specific search operators for crossover and mutation based on preliminary results.

Filter transfer learning algorithm for nonlinear systems modeling with heterogeneous features

Transfer learning can handle the domain adaptation of different feature spaces in nonlinear systems. Most existing studies only focus on common features between heterogeneous scenes rather than specific features that account for latent similarities between them. To deal with this problem, a filter transfer learning algorithm for nonlinear systems modeling with heterogeneous features is proposed. First, nonlinear mapping is constructed to learn the potential relationships between different feature attributes, including common features and specific features. Then, source knowledge from different domains can be obtained in the form of process parameters according to the mapping relationship. Second, a hierarchical filter framework is presented to reconstruct source knowledge in different transfer phases. In the pre-transfer phase, a knowledge filter is designed to increase the diversity of knowledge through selection, crossover, and mutation operations. In the post-transfer phase, a guided filter is established to achieve a coupling balance between source knowledge and target domain by using target samples as guidance information. Third, a dynamic parameter learning strategy is given to promote the learning performance of heterogeneous tasks. Finally, the effectiveness of innovations and the superiority of this proposed algorithm are verified by the experimental results in nonlinear systems with heterogeneous features.

Ship Lock Chamber Arrangement in Xiangjiang River Based on Artificial Bee Colony and Adaptive Genetic Algorithm

Abstract To plan the ship lock chamber arrangement in Xiangjiang river, an artificial bee colony-adaptive genetic algorithm was proposed for the scheduling of the lock chamber, considering that the performance of the basic genetic algorithm relies on the quality of the initial population, the selection of genetic operators, crossover, and mutation operations. Firstly, the artificial bee colony algorithm was used to initialize the population to enhance population diversity. The utilization rate of the lock chamber and the waiting time for the lock were used as evaluation indicators for the fitness function. Then, adaptive strategies were designed to adjust the crossover and mutation operators to improve the convergence speed of the algorithm. Experimental results showed that, in a Xiangjiang river ship lock chamber with a size of 280 m×34 m, the average number of ships selected by the proposed artificial bee colony-adaptive genetic algorithm was 2 more than that of manual scheduling and 1.5 more than that of the genetic algorithm. The utilization rate of the lock chamber scheduling was increased by 13.01% compared to manual scheduling and by 5.4% compared to the genetic algorithm. The average waiting time for the lock was reduced by 1.3 hours compared to manual scheduling and by 0.8 hours compared to the genetic algorithm. It effectively improves the efficiency of ship traffic and optimizes the utilization of the lock chamber, making it practical for real-world applications.

SR-Forest: A Genetic Programming-Based Heterogeneous Ensemble Learning Method

Ensemble learning methods have been widely used in machine learning in recent years due to their high predictive performance. With the development of genetic programming-based symbolic regression methods, many papers begin to choose a popular ensemble learning method, random forests, as the baseline competitor. Instead of considering them as competitors, an alternative idea might be to consider symbolic regression as an enhancement technique for random forest. Genetic programming-based symbolic regression methods which fit a smooth function are complementary to the piecewise nature of decision trees, as the smooth variation is common in regression problems. In this article, we propose to form an ensemble model with symbolic regression-based decision trees to address this issue. Furthermore, we design a guided mutation operator to speed up the search on high-dimensional problems, a multi-fidelity evaluation strategy to reduce the computational cost and an ensemble selection mechanism to improve predictive performance. Finally, experimental results on a regression benchmark with 120 datasets show that the proposed ensemble model outperforms 25 existing symbolic regression and ensemble learning methods. Moreover, the proposed method can provide notable insights on an XGBoost hyperparameter performance prediction task, which is an important application area of ensemble learning methods.

Divide-and-conquer initialization and mutation operators for the large-scale mixed Capacitated Arc Routing Problem

As cities continue to grow, thus will the size of the routing problems necessary to the functioning of all cities. Applications such as waste collection, road maintenance, or winter gritting span over an entire city, therefore algorithms must be designed to solve large-scale problems. The Capacitated Arc Routing Problem (CARP) is an important combinatorial optimization problem that is typically used to model these applications. Classical algorithms for CARP struggle to find quality solutions for large-scale instances with thousands of services within a reasonable computational budget. To address the issue of scalability, several divide-and-conquer heuristics have recently been proposed. In this paper, we propose to integrate divide-and-conquer heuristics into a memetic algorithm by adapting these as an initialization method and as a mutation operator. The resulting algorithm, which we call Memetic Algorithm with Divide-and-Conquer Mutation (MADCoM), outperforms state-of-the-art algorithms on large-scale instances and new best solutions are found for 17 instances of MCARP, 2 of which are optimal solutions, and for 23 large-scale CARP instances. These results demonstrate the potential of the integration of divide-and-conquer heuristics into metaheuristics as a strategy to efficiently solve large-scale problems.

A survey of genetic algorithms for clustering: Taxonomy and empirical analysis

Clustering, an unsupervised learning technique, aims to group patterns into clusters where similar patterns are grouped together, while dissimilar ones are placed in different clusters. This task can present itself as a complex optimization problem due to the extensive search space generated by all potential data partitions. Genetic Algorithms (GAs) have emerged as efficient tools for addressing this task. Consequently, significant advancements and numerous proposals have been developed in this field.This work offers a comprehensive and critical review of state-of-the-art mono-objective Genetic Algorithms (GAs) for partitional clustering. From a more theoretical standpoint, it examines 22 well-known proposals in detail, covering their encoding strategies, objective functions, genetic operators, local search methods, and parent selection strategies. Based on this information, a specific taxonomy is proposed. In addition, from a more practical standpoint, a detailed experimental study is conducted to discern the advantages and disadvantages of approaches. Specifically, 22 different cluster validation indices are considered to compare the performance of clustering techniques. This evaluation is performed across 94 datasets encompassing diverse configurations, including the number of classes, separation between classes, and pattern dimensionality. Results reveal interesting findings, such as the key role of local search in optimizing results and reducing search space. Additionally, representations based on centroids and labels demonstrate greater efficiency and crossover and mutation operators do not prove to be as relevant. Ultimately, while the results are satisfactory, real-world clustering problems introduce additional complexity, especially for algorithms aiming to determine the number of clusters, resulting in diminished performance and the need for new approaches to be explored. Code, datasets and instructions to run algorithms in the LEAL library are available in an associated repository, in order to facilitate future experiments in this environment.

An energy-saving distributed flexible job shop scheduling with machine breakdowns

Due to the change of consumption pattern, distributed flexible production is now becoming a major manufacturing mode. The machine breakdown seriously affects the production in workshops. How to efficiently adjust the scheduling scheme after machine breakdowns in distributed flexible job shop is an important concern for enterprises. Meanwhile, energy consumption is attracting more and more attention as environmental issues become increasingly serious. Thus, the energy-saving scheduling on distributed flexible job shop considering machine breakdowns is studied, and a mixed-integer programming model is established to optimize makespan, total energy consumption and machine load difference, simultaneously. Based on the problem characteristics, an improved memetic algorithm is designed. To improve the quality of initial solutions, an initialization strategy incorporating multiple rules is designed. An active decoding method with an adjustment strategy is adopted to fully utilize machine idleness. To expand the search range, the idea of dual-population collaborative optimization is used. Multi-level intelligent mutation operation strategies are introduced to further enhance the quality of solutions. In addition, an adaptive parameter rule is designed to adjust to appropriate search capabilities at different evolutionary stages. To verify the effectiveness of the proposed strategies and algorithm, the memetic algorithm, the artificial bee colony algorithm, the differential evolution algorithm and the estimation of distribution algorithm are chosen as comparison algorithms. The same iteration number and same running time are used as the termination criteria of all algorithms, and 60 instances of different scales are solved. The experimental results demonstrate that the effectiveness of the strategies and algorithm designed in this paper. Notably, our algorithm consistently achieved good results even when the same running time is used as the termination criterion.

An Enhanced Symmetric Sand Cat Swarm Optimization with Multiple Strategies for Adaptive Infinite Impulse Response System Identification

An infinite impulse response (IIR) system might comprise a multimodal error surface and accurately discovering the appropriate filter parameters for system modeling remains complicated. The swarm intelligence algorithms facilitate the IIR filter’s parameters by exploring parameter domains and exploiting acceptable filter sets. This paper presents an enhanced symmetric sand cat swarm optimization with multiple strategies (MSSCSO) to achieve adaptive IIR system identification. The principal objective is to recognize the most appropriate regulating coefficients and to minimize the mean square error (MSE) between an unidentified system’s input and the IIR filter’s output. The MSSCSO with symmetric cooperative swarms integrates the ranking-based mutation operator, elite opposition-based learning strategy, and simplex method to capture supplementary advantages, disrupt regional extreme solutions, and identify the finest potential solutions. The MSSCSO not only receives extensive exploration and exploitation to refrain from precocious convergence and foster computational efficiency; it also endures robustness and reliability to facilitate demographic variability and elevate estimation precision. The experimental results manifest that the practicality and feasibility of the MSSCSO are superior to those of other methods in terms of convergence speed, calculation precision, detection efficiency, regulating coefficients, and MSE fitness value.

Distributed enhanced multi-objective evolutionary algorithm based on decomposition for cluster analysis in wireless sensor network

Conventional clustering algorithms do not recognize patterns and structures with contradicting objectives in large, distributed datasets. Distributed clustering leverages rapid processing capabilities to allow multiple nodes to work together. This paper proposes a Distributed clustering based on Multiobjective Evolutionary Algorithm by Decomposition (D-MOEA/d) to solve various multiobjective optimization problems in wireless sensor networks (WSNs). In MOEA/d, a multiobjective optimization problem decomposes into several scalar optimization subproblems, each focusing on a distinct objective. Each subproblem is expressed as a clustering problem that uses local data to perform distributed clustering. The proposed method has been extended to achieve improved accuracy in less time by using a smaller feature subset with less redundancy. The Distributed Enhanced MOEA/d (DE-MOEA/d) avoids local optima by achieving diversity in the population using fuzzy-based nearest neighbor selection, sparse population initialization, and evolved mutation operator. This integration improves the accuracy of the clustering process at WSN nodes, ensuring the attainment of well-balanced solutions across multiple optimization criteria in the distributed environment. Average Euclidean and total symmetrical deviations are the two cost functions used to minimize while clustering on the MOEA/d framework. Six real-life WSN datasets are used to assess the performance of the proposed technique: (1) the Delhi air pollution dataset, (2) the Canada weather station dataset, (3) the Thames River water quality dataset, (4) the Narragansett Bay water quality dataset, (5) the Cook Agricultural land dataset and 6) Gordon Soil dataset. The simulation results of both proposed algorithms are compared with Multiobjective distributed particle swarm optimization (DMOPSO) and Distributed K-means (DK-Means). The proposed algorithm DE-MOEA/d performs better in terms of the Silhouette index (SI), Dunn index (DI), Davies–Bouldin index (DBI), and Kruskal–Wallis (KW) statistical test.

Multi-branch evolutionary generative adversarial networks based on covariance crossover operators

Generative Adversarial Networks (GANs) have brought surprises in image generation and processing, particularly excelling in alleviating the problem of sparse samples in few-shot learning. Evolutionary Generative Adversarial Networks (EGANs) aim to convert the generation task into an optimization problem by integrating multiple loss functions to minimize the gap between the generated distribution and the data distribution. However, EGANs’ heavy reliance on mutation operations introduces excessive randomness, resulting in unstable generator updates and affecting the diversity of generated samples. This leads to challenges such as model collapse and gradient vanishing. In this paper, inspired by human grafting mechanism, we propose a novel Multi-branch Evolutionary Generative Adversarial Network (ME-GAN) based covariance crossover operator, which includes two distinct branches – Evolutionary GAN (E-GAN) and Conditional Evolutionary GAN (CE-GAN). Specifically, the E-GAN branch involves a mutation process to introduce randomness to the generated samples, while the CE-GAN branch aims to provide more realistic and diversify generated samples through conditional enhancement and relevant mutation processes. We propose a crossover operation that utilizes the covariance similarity metric to transfer different feature attributes between offspring of different generations, thereby generating a diverse sample. Extensive experiments on CIFAR-10, STL-10, and CelebA datasets demonstrate the effectiveness and superiority of the proposed framework.