Multi-objective evolutionary algorithms (MOEAs) have been proved to be competitive in recommender systems. As the application scenarios of recommender systems become increasingly complex, the number of objectives to b...
详细信息
Multi-objective evolutionary algorithms (MOEAs) have been proved to be competitive in recommender systems. As the application scenarios of recommender systems become increasingly complex, the number of objectives to be considered in the recommender systems increases. However, most existing multi-objective recommendation algorithms lead to increased environmental selection pressure as the number of objectives increases. To tackle the issue, in this paper, we propose a multi-population based evolutionary algorithm named MP-MORS for many-objective recommendations, where two subpopulations and one major population are used to evolve and interact to find high-quality solutions. Specifically, the objectives are firstly divided into those evaluated on individual users (defined as IndObjectives) and those evaluated on all users (defined as as AllObjectives). Then two subpopulations are suggested to optimize the two types of objectives respectively, with which the potential good solutions can be easily found. In addition, the major population considers the balance of all objectives and refines these potential good solutions. Finally, a set of high-quality solutions can be obtained by the proposed adaptive population interaction strategy. Experiments on the datasets Movielens and Douban show that the proposed MP-MORS outperforms the state-of-the-art algorithms for many-objective recommendations.
The Graph Query Language (GraphQL) is a powerful language for application programming interface (API) manipulation in web services. It has been recently introduced as an alternative solution for addressing the limitat...
详细信息
The Graph Query Language (GraphQL) is a powerful language for application programming interface (API) manipulation in web services. It has been recently introduced as an alternative solution for addressing the limitations of RESTful APIs. This article introduces an automated solution for GraphQL API testing. We present a full framework for automated API testing, from the schema extraction to test case generation. In addition, we consider two kinds of testing: white-box and black-box testing. The white-box testing is performed when the source code of the GraphQL API is available. Our approach is based on evolutionary search. Test cases are evolved to intelligently explore the solution space while maximizing code coverage and fault-finding criteria. The black-box testing does not require access to the source code of the GraphQL API. It is therefore of more general applicability, albeit it has worse performance. In this context, we use a random search to generate GraphQL data. The proposed framework is implemented and integrated into the open source EvoMaster tool. With enabled white-box heuristics (i.e., white-box mode), experiments on 7 open source GraphQL APIs and three search algorithms show statistically significant improvement of the evolutionary approach compared to the baseline random search. In addition, experiments on 31 online GraphQL APIs reveal the ability of the black-box mode to detect real faults.
Manually designed convolutional neural networks have demonstrated excellent performance in various domains, but designing neural networks suitable for specific tasks poses significant challenges, and the emergence of ...
详细信息
Manually designed convolutional neural networks have demonstrated excellent performance in various domains, but designing neural networks suitable for specific tasks poses significant challenges, and the emergence of Neural Structure Search (NAS) provides a new solution to this problem. However, existing algorithms either focus solely on network lightweight, resulting in subpar network performance, or excessively emphasize performance, leading to substantial network redundancy. With consideration for both network parameters and performance, this paper designs a hybrid search space based on residual modules and RepVGG modules using genetic algorithm, and stacks them together to form a more efficient network. To achieve this, we propose an efficient variable-length encoding strategy, utilizing units as the fundamental encoding space to encode variable-length individuals;we design evolutionary operations encompassing single-point crossover and three types of mutation operators to ensure population diversity;during training, a random forest-based performance predictor is employed to significantly shorten the network search time. To demonstrate the effectiveness of the proposed algorithm, we introduce the concept of transfer learning, which involves decoding the globally optimal solution, fine-tuning it, and then transferring it to three categories of real-world application datasets. Through comparisons with various algorithms, our approach consistently achieved leading performance.
Multi -objective optimization research has mostly focused on continuous -variable problems. However, real -world optimization problems often involve multiple types of variables (continuous, integer, and discrete) and ...
详细信息
Multi -objective optimization research has mostly focused on continuous -variable problems. However, real -world optimization problems often involve multiple types of variables (continuous, integer, and discrete) and multiple conflicting optimization objectives, called mixed -variable multi -objective optimization problems (MVMOPs). Discrete variables make the decision space of the problem discrete. In contrast, while different types of variables need different treatments by the evolutionary algorithm, which poses a challenge to the efficient search of the evolutionary algorithm. Therefore, we propose an evolutionary algorithm based on a fully connected weight network (FCWNEA). The fully connected network structure characterizes the entire decision space, the node access count records the frequency of visits to the node, and the weights of connections and the activity of variables estimate the distribution of the decision space. This information assists in generating offspring solutions. To evaluate the performance of the proposed algorithm, we conduct empirical experiments on different types of problems. The results show that the proposed algorithm has a significant advantage in mixed -variable multi -objective problems. Moreover, the proposed algorithm is also quite competitive in continuous problems and can better handle the correlation between variables in optimization problems.
Solving a multiple-criteria optimization problem with severe constraints remains a significant issue in multi-objective evolutionary algorithms. The problem primarily stems from the need for a suitable constraint hand...
详细信息
Solving a multiple-criteria optimization problem with severe constraints remains a significant issue in multi-objective evolutionary algorithms. The problem primarily stems from the need for a suitable constraint handling technique. One potential approach is balancing the search in feasible and infeasible regions to find the Pareto front efficiently. The justification for such a strategy is that the infeasible region also provides valuable information, especially in problems with a small percentage of feasibility areas. To that end, this paper investigates the potential of the infeasibility-driven principle based on multiple constraint ranking-based techniques to solve a multi-objective problem with a small feasibility ratio. By analyzing the results from intensive experiments on a set of test problems, including the realistic multi-objective car structure design and actuator design problem, it is shown that there is a significant improvement gained in terms of convergence by utilizing the generalized version of the multiple constraint ranking techniques.
In evolutionary constrained multi-objective optimization, the use of auxiliary optimization is gradually attracting attention. It is noted that different forms of auxiliary optimization have different advantages. Comb...
详细信息
In evolutionary constrained multi-objective optimization, the use of auxiliary optimization is gradually attracting attention. It is noted that different forms of auxiliary optimization have different advantages. Combining these advantages in an appropriate manner can further improve the algorithm's performance. Motivated by this inspiration, we propose a double auxiliary optimization constrained multi-objective evolutionary algorithm, namely DAO. In DAO, two auxiliary optimizations, i.e., the unconstrained optimization and the (M+1)-objective optimization, are applied in a proposed tri-population co-evolution strategy. In this strategy, three populations are used to optimize the core optimization and the two auxiliary optimizations in an interactive evolution form. Furthermore, DAO develops a (M+1)-objective environmental selection strategy to deeply explore the boundary between the feasible and infeasible regions. In experimental studies, the performance of DAO and four other state -of -the -art algorithms is evaluated on three distinct benchmark test suites and two intricate real -world problem scenarios. The outcome of the comprehensive evaluation shows the competitive of DAO solving constrained multi-objective optimization problems.
作者：
Wu, YingYang, NaChen, LongTian, YeTang, ZhenzhouWenzhou Univ
Wenzhou Key Lab Intelligent Networking Wenzhou 325035 Peoples R China Zhejiang Normal Univ
Coll Teacher Educ Key Lab Intelligent Educ Technol & Applicat Zhejia Jinhua 321000 Peoples R China Anhui Univ
Inst Phys Sci & Informat Technol Key Lab Intelligent Comp & Signal Proc Minist Educ Hefei 230601 Peoples R China
For large-scale multi-objective evolutionary algorithms (LSMOEAs), obtaining efficient evolutionary directions in an ultrahigh-dimensional decision space to produce high-quality offspring is a major challenge. This st...
详细信息
For large-scale multi-objective evolutionary algorithms (LSMOEAs), obtaining efficient evolutionary directions in an ultrahigh-dimensional decision space to produce high-quality offspring is a major challenge. This study proposes a novel framework called the directed quick search-guided large-scale evolutionary framework (QSLMOF) to address multi-objective optimization problems on a large scale. The framework contains two innovative strategies: the bidirectional vector-based sampling strategy (BDVS) and the quick search-guided directed reproduction strategy (QS-DRS). BDVS is introduced as an approach to swiftly discern promising solutions that can steer the exploration in the large-scale decision space. This is achieved by formulating two distinct types of sampling directions to rapidly reduce the search space and strike a delicate balance between convergence and diversity. In QS-DRS, we introduced the concept of the potential convergence gradient (PCG), incorporating directional information from historical searches and convergence directions indicated byelite solutions in the current population. With this property, inferior solutions can obtain excellent search directions to explore the decision space, which can improve the convergence rate and prevent the search from falling into a local optimum. The proposed large-scale evolutionary framework incorporates an existing environmental selection mechanism. Comprehensive experiments show that the two novel strategies improve the search efficiency and evolutionary quality of LSMOEAs in ultrahigh-dimensional decision spaces. Moreover, the proposed framework outperformed seven state-of-the-art LSMOEAs for nine large-scale multi-objective benchmark problems LSMOP1-LSMOP9 with up to three objectives and 10000 decision variables.
Physics-informed neural networks formulation allows the neural network to be trained by both the training data and prior domain knowledge about the physical system that models the data. In particular, it has a loss fu...
详细信息
Physics-informed neural networks formulation allows the neural network to be trained by both the training data and prior domain knowledge about the physical system that models the data. In particular, it has a loss function for the data and the physics, where the latter is the deviation from a partial differential equation describing the system. Conventionally, both loss functions are combined by a weighted sum, whose weights are usually chosen manually. It is known that balancing between different loss terms can make the training process more efficient. In addition, it is necessary to find the optimal architecture of the neural network in order to find a hypothesis set in which is easier to train the PINN. In our work, we propose a multi-objective optimization approach to find the optimal value for the loss function weighting, as well as the optimal activation function, number of layers, and number of neurons for each layer. We validate our results on the Poisson, Burgers, and advection-diffusion equations and show that we are able to find accurate approximations of the solutions using optimal hyperparameters.
Responding quickly and accurately to environmental changes is a challenge in addressing dynamic multi-objective optimization problems (DMOPs). Although many dynamic multi- objective evolutionary algorithms (DMOEAs) ha...
详细信息
Responding quickly and accurately to environmental changes is a challenge in addressing dynamic multi-objective optimization problems (DMOPs). Although many dynamic multi- objective evolutionary algorithms (DMOEAs) have demonstrated impressive performance, there is still room for improvement in accurately predicting population behavior. To address this limitation, this paper proposes a prediction strategy based on Mahalanobis distance and intracluster individual correlation rectification (MCIR) to deal with DMOPs. First, a manifold clustering method is used to partition the Pareto set of the population into subpopulations, in which individuals with similar movement trends are grouped into clusters. Second, the Mahalanobis distance is introduced to systematically measure the relationships between clusters in adjacent environments. Time series models are established for each cluster center to predict their positions in the neighboring environment. On this basis, the movement characteristics of individuals within clusters are further rectified by calculating the correlation between intra-cluster individuals and the cluster center, facilitating more accurate tracking of the changing Pareto set/Pareto front. Finally, Gaussian noise is introduced to ensure the diversity of new individuals. The effectiveness of the MCIR algorithm is demonstrated by comparing it with four DMOEAs using 18 test instances. Experimental results confirm that MCIR holds great promise in addressing DMOPs.
In the evolutionary algorithm, an intuitive phenomenon is that the eliminated bad particles are also beneficial to convergence in evolutionary algorithms by preventing the generated particles from being close to those...
详细信息
In the evolutionary algorithm, an intuitive phenomenon is that the eliminated bad particles are also beneficial to convergence in evolutionary algorithms by preventing the generated particles from being close to those eliminated bad particles. Most existing algorithms do not take full advantage of the historical information of these particles or use surrogate models without guaranteeing approximation accuracy. In this study, we propose a multi-granularity general framework to divide the feasible region into different granularities by utilizing completely random trees and computing the spatial distribution of individuals. Secondly, through the sampling and migration strategy, make full use of the sparsity of the calculated individual space distribution and the locality of the best individual in history to replace the poor individual in the current population to speed up the local convergence speed of the algorithm. The time complexity of the algorithm using this framework is equal to the maximum between the time complexity of the evolutionary algorithm using this framework and O(tMlogM), where M denotes the number of points and historical particles generated in an iteration and t denotes the number of iterations. Therefore, the additional computational cost incurred by this framework is very low. Experiments on 12 classical functions, including high-dimensional functions, show that the proposed framework can improve four respective evolutionary algorithms and achieve significantly better results in terms of convergence performance and optimization accuracy.
暂无评论