Computational creativity is an emerging topic in artificial intelligence area. Past studies have shown the effectiveness of deep learning methods for image generation. Contrast to simple image generation, it is more practical to generate images according to specific description, known as text-to-image generation. However, text-to-image generation is much more difficult than simple image generation as it may involve natural language processing for semantic retrieval, learning mapping between images to semantics, and learning image generator. This study proposes an evolutionary framework for language-free text-to-image generation. There are three main features in the proposed framework. First, the framework bases on language-free model for image generation, which purifies the text-to-image generation problem as an optimization problem over a learnt image generator. Second, the use of language-free model simplifies the requirements of data preparation and learning process. Third, in the proposed framework an objective function is designed for improving the quality of the generated images in terms of both photo realisticness and semantic consistency. Experimental results validate the effectiveness of the proposed framework in text-to-image generation on different scenarios, in comparison to state-of-the-art methods. This study further provides and examines the possibility for the proposed framework on image manipulation as a potential application. This study also discusses the designed operators and hyper-parameters and provides guidelines for the settings.

Reinforcement learning has gathered lots of attention as many real-world applications such as complex control and optimization problems can be formulated as reinforcement learning problem. Reinforcement learning aims to find a policy capable of pursuing high return through policy optimization. Evolutionary multitasking is an emerging technique in evolutionary computation field, and it has been proved to be effective in handling multiple complicated optimization problems. This study proposes a novel policy search method: the symbiosis in biocoenosis optimization for reinforcement learning (SBO-RL). The SBO-RL is composed of three main features. First, the SBO-RL copes with multiple RL problems concurrently. Second, the SBO-RL integrates evolutionary-based policy optimization with policy gradient into symbiosis in biocoenosis optimization. Third, the SBO-RL transfers actors across domains by a simple domain adaptation strategy. The present study conducts experiments to examine the effectiveness of the proposed method. Experimental results show the SBO-RL can improve the quality and convergence of advanced policy-gradient and policy-search based deep reinforcement learning methods. Results also discover that the SBO-RL is capable of saving 50.06% and 31.15% computational cost for advanced deep reinforcement learning methods.

Evolutionary computation has served as a blooming research area for decades, in which evolutionary algorithms are inspired from mechanisms of evolution as well as cognitive and social behaviors of creatures in nature as searching processes. An emerging branch of evolutionary computation proposed in recent year is the evolutionary multitasking, which is to manipulate evolutionary algorithms for tackling multitask optimization problems. Recent studies of evolutionary multitasking have drawn much attention on designing effective knowledge transfer mechanisms. This study proposes a novel evolutionary multi-tasking method named summit-assisted evolutionary multitasking (SaEMT) by integrating summit-based knowledge transfer into the general multi-population evolutionary multitasking method. There are two main features in the proposed method, including the summit-based recombination, and the dynamic control of transfer rate. Empirical results show that the proposed method can outperform classical and advanced evolutionary multitasking methods in terms of solution quality and convergence speed. Experimental results also discover that the SaEMT is able to complete in acceptable running time.

Association mining is a famous data mining technology because its form is explainable by human beings. Innovating fuzzy set theory to associations mining provides a solution to quantitative database, where membership function plays an important role in mining fuzzy associations. Genetic algorithm (GA) has been successfully applied to the optimization of membership functions. Based on the spirit of divide-and-conquer, GA optimizes the membership functions for each item separately. Nevertheless, the cooperation among different items in the course of evolution was never considered. Evolutionary multitasking optimization (EMO) is an emerging searching paradigm which dedicates to solving multiple tasks simultaneously for improving the search efficiency. This study introduces the EMO into genetic fuzzy data mining to address the above issue. Specifically, this study incorporates a state-of the-art genetic fuzzy data mining method, the structure-based representation genetic algorithm, with the well-known multifactorial evolutionary algorithm (MFEA). A series of experiments is conducted to validate the effectiveness and efficiency of the proposed method. The results indicate that the proposed method improves the structure-based representation genetic algorithm in terms of convergence speed and solution quality on all sizes of datasets. The results also show that the proposed method is about 20 times faster than the structure-based representation genetic algorithm with respect to the exploited number of evaluations.

Evolutionary multitasking is a significant emerging search paradigm that utilizes evolutionary algorithms to concurrently optimize multiple tasks. The multi-factorial evolutionary algorithm renders an effectual realization of evolutionary multitasking on two or three tasks. However, there remains room for improvement on the performance and capability of evolutionary multitasking. Beyond three tasks, this paper proposes a novel framework, called the symbiosis in biocoenosis optimization (SBO), to address evolutionary many-tasking optimization. The SBO leverages the notion of symbiosis in biocoenosis for transferring information and knowledge among different tasks through three major components: 1) transferring information through inter-task individual replacement, 2) measuring symbiosis through intertask paired evaluations, and 3) coordinating the frequency and quantity of transfer based on symbiosis in biocoenosis. The inter-task individual replacement with paired evaluations caters for estimation of symbiosis, while the symbiosis in biocoenosis provides a good estimator of transfer. This study examines the effectiveness and efficiency of the SBO on a suite of many-tasking benchmark problems, designed to deal with 30 tasks simultaneously. The experimental results show that SBO leads to better solutions and faster convergence than the state-of-the-art evolutionary multitasking algorithms. Moreover, the results indicate that SBO is highly capable of identifying the similarity between problems and transferring information appropriately.

Evolutionary algorithms (EAs) have shown their great capability of handling optimization problems. In the domain of large scale global optimization, many distributed EAs (dEAs) have been proposed for maintaining population diversity so as to enhance their efficacy. One well-known variant of dEA is the island model EA, in which several populations form islands communicating through migration. There are several key factors that affect the performance and search behavior of island model EA, such as population size of each island and migration topology. While most studies of dEA focus on low or medium dimensional problems, an investigation into the effects of these factors on high dimensional problems is greatly needed. This study presents an empirical analysis of island model EA on large scale global optimization problems. The analysis examines the solution quality, convergence speed, and population diversity of island model EA with different migration topologies, population sizes, migration rates, and migration frequencies on four benchmark function of 1,000 dimensions. The results render guidelines for using island model EA to solve large scale global optimization problems.

Many real-world problems encounter varying environments while searching for the optimal solutions. Problems of these kinds are known as dynamic optimization problems. Evolutionary algorithms (EAs) have shown their high capability of solving dynamic optimization problems. The large number of fitness evaluations, however, limits the utility of EAs in dynamic optimization with expensive fitness evaluation. This study proposes a fitness inheritance method based on k-nearest neighbors (kNN) to reduce the number of fitness evaluations. This fitness inheritance method is applied to the covariance matrix adaptation evolution strategy (CMAES), a state-of-the-art evolutionary algorithm for complex numerical optimization problems, to save fitness evaluations and improve search efficiency. This study further presents a convergence checking operator to maintain a reasonable accuracy of fitness inheritance by reinitializing the population and multivariate normal distribution model timely. Experimental results verify the effectiveness and efficiency of the proposed method on six benchmark problems of dynamic optimization.

Evolutionary multitasking is an emergent topic in evolutionary computation area. Recently, a well-known evolutionary multitasking method, the multi-factorial evolutionary algorithm (MFEA), has been proposed and applied to concurrently solve two or three problems. In MFEA, individuals of different tasks are recombined in a predefined random mating probability. As the number of tasks increases, such recombination of different tasks becomes very frequent, thereby detracting the search from any specific problems and limiting the MFEA's capability to solve many-tasking problems. This study proposes a general framework, called the evolution of biocoenosis through symbiosis (EBS), for evolutionary algorithms to deal with the many-tasking problems. The EBS has two main features: the selection of candidates from concatenate offspring and the adaptive control of information exchange among tasks. The concatenate offspring represent a set of offspring used for all tasks. Moreover, this study presents a test suite of many-tasking problems (MaTPs), modified from the CEC 2014 benchmark problems. The Spearman correlation is adopted to analyze the effect of the shifts of optima on the MaTPs. Experimental results show that the effectiveness of EBS is superior to that of single task optimization and MFEA on the four MaTPs. The results also validate that EBS is capable of exploiting the synergy of fitness landscapes.

Evolution strategy (ES) has shown to be effective in many search and optimization problems. In particular, the ES with covariance matrix adaptation (CMAES) achieves great successes and is viewed as a state-of-the-art evolutionary algorithm for complex numerical optimization. The CMAES models the population by a multivariate normal distribution, which requires a considerable amount of fitness evaluation results and thus degrades its efficiency. This paper proposes using fitness inheritance to reduce the computational cost at fitness evaluation. More specifically, the proposed FI-CMAES adopts fitness inheritance to approximate the fitness of offspring. The survivors are selected according to the approximated fitness; thereafter, the survival offspring are evaluated by the original fitness function. By this way, several original fitness evaluations on offspring can be saved. Experiments examine the effectiveness and efficiency of FI-CMAES on the CEC2014 test suite. The results show that FI-CMAES can outperform CMAES in terms of solution quality and convergence speed.

Multi-objective optimization is a blooming research area since many real-world problems comprise multiple objectives. Multi-objective evolutionary algorithms (MOEAs) have been widely used to solve the multi-objective optimization problems. In particular, the decomposition based MOEA (MOEA/D) has achieved considerable successes in tackling multi-objective optimization problems. The covariance matrix adaptation evolution strategy (CMAES) is known for its effectiveness in solving complex numerical optimization problems. This study integrates CMAES into MOEA/D as the MOEA/D-CMAES for the merits of MOEA/D framework in multi-objective optimization and CMAES in complex numerical optimization. In MOEA/D-CMAES, each subproblem is handled with one CMAES. To avoid the drastic increase in the number of offspring generated and their fitness evaluations, MOEA/D-CMAES generates only one offspring in each subproblem. The multivariate normal distribution in each CMAES is updated by the collaboration of the offspring generated in the present subproblem and those of other subproblems. Experimental results show that MOEA/D-CMAES outperforms MOEA/D using differential evolution in terms of hypervolume and convergence speed, which validate the effectiveness and efficiency of MOEA/D-CMAES in multi-objective optimization.