Improved whale optimization algorithm based on variable spiral position update strategy and adaptive inertia weight

In this paper, a whale optimization algorithm based on adaptive inertia weight and variable spiral position updating strategy is proposed. The improved algorithm is used to solve the problem that the whale optimization algorithm is more dependent on the randomness of the parameters, so that the algorithm’s convergence accuracy and convergence speed are insufficient. The adaptive inertia weight, which varies with the fitness of individual whales, is used to balance the algorithm’s global search ability and local exploitation ability. The variable spiral position update strategy based on the collaborative convergence mechanism is used to dynamically adjust the search range and search accuracy of the algorithm. The effective combination of the two can make the improved whale optimization algorithm converge to the optimal solution faster. It had been used 18 international standard test functions, including unimodal function, multimodal function, and fixed-dimensional function to test the improved whale optimization algorithm in this paper. The test results show that the improved algorithm has faster convergence speed and higher algorithm accuracy than the original algorithm and several classic algorithms. The algorithm can quickly converge to near the optimal value in the early stage, and then effectively jump out of the local optimal through adaptive adjustment, and has a certain ability to solve large-scale optimization problems.

EXPERIMENTAL ANALYSIS OF MULTINATIONAL GENETIC ALGORITHM AND ITS MODIFICATIONS

Context. Niching genetic algorithms are one of the most popular approaches to solve multimodal optimization problems. When classifying niching genetic algorithms it is possible to select algorithms explicitly analyzing topography of fitness function landscape; multinational genetic algorithm is one of the earliest examples of these algorithms. Objective. Development and analysis of the multinational genetic algorithm and its modifications to find all maxima of a multimodal function. Method. Experimental analysis of algorithms is carried out. Numerous runs of algorithms on well-known test problems are conducted and performance criteria are computed, namely, the percentage of convergence, real (global, local) and fake peak ratios; note that peak rations are computed only in case of algorithm convergence. Results. Software implementation of a multinational genetic algorithm has been developed and experimental tuning of its parameters has been carried out. Two modifications of hill-valley function used for determining the relative position of individuals have been proposed. Experimental analysis of the multinational genetic algorithm with classic hill-valley function and with its modifications has been carried out. Conclusions. The scientific novelty of the study is that hill-valley function modifications producing less number of wrong identifications of basins of attraction in comparison with classic hill-valley function are proposed. Using these modifications yields to performance improvements of the multinational genetic algorithm for a number of test functions; for other test functions improvement of the quality criteria is accompanied by the decrease of the convergence percentage. In general, the convergence percentage and the quality criterion values demonstrated by the algorithm studied are insufficient for practical use in comparison with other known algorithms. At the same time using modified hill-valley functions as a post-processing step for other niching algorithms seems to be a promising improvement of performance of these algorithms.

A comparative study of gradient-based and meta-heuristic optimization methods using Griewank benchmark function/ Um estudo comparativo de métodos de otimização baseados em gradientes e meta-heurísticos usando a função de benchmark do Griewank

Optimization methods are frequently applied to solve real-world problems such, engineering design, computer science, and computational chemistry. This paper aims to compare gradient-based algorithms and the meta-heuristic particle swarm optimization to minimize the multidimensional benchmark Griewank function, a multimodal function with widespread local minima. Several approaches of gradient-based methods such as steepest descent, conjugate gradient with Fletcher-Reeves and Polak-Ribiere formulations, and quasi-Newton Davidon-Fletcher-Powell approach were compared. The results presented showed that the meta-heuristic method is recommended for function with this behavior because is no needed prior information of the search space. The performance comparison includes computation time and convergence of global and local optimum.

Multi-user Remote Lab: Timetable Scheduling Using Simplex Nondominated Sorting Genetic Algorithm

The scheduling of multi-user remote laboratories is modeled as a multimodal function for the proposed optimization algorithm. The hybrid optimization algorithm, hybridization of the Nelder-Mead Simplex algorithm, and Non-dominated Sorting Genetic Algorithm (NSGA), named Simplex Non-dominated Sorting Genetic Algorithm (SNSGA), is proposed to optimize the timetable problem for the remote laboratories to coordinate shared access. The proposed algorithm utilizes the Simplex algorithm in terms of exploration and NSGA for sorting local optimum points with consideration of potential areas. SNSGA is applied to difficult nonlinear continuous multimodal functions, and its performance is compared with hybrid Simplex Particle Swarm Optimization, Simplex Genetic Algorithm, and other heuristic algorithms. The results show that SNSGA has a competitive performance to address timetable problems.

Multimodal Modular Meta-Learning

Many real world prediction problems involve structured tasks across multiple modalities. We propose to extend previous work in modular meta learning to the multimodal setting. Specifically, we present an algorithmic approach to apply task aware modulation to a modular meta learning system that decomposes structured multimodal problems into a set of modules that can be reassembled to learn new tasks. We also propose a series of experiments to compare this approach with state of the art modular and multimodal meta learning approaches on multimodal function prediction and image classification tasks.

Chaos-based second-order bas for multimodal function optimization

Aiming at solving a drawback of the second-order beetle antenna search (SOBAS), a variant of the beetle antenna search (BAS), that it is difficult to find the global optimal solution and the low convergence accuracy when applied to the multimodal optimization functions with high dimension or large variable region, a chaotic-based second-order BAS algorithm (CSOBAS) is proposed by introducing chaos theory into the SOBAS. The algorithm mainly has three innovations: 1) chaos initialization: choosing the one with the smallest fitness function value from twenty beetles with different positions for iterative search; 2) using chaotic map to tune the randomization parameter in the detection rule; 3) imposing a chaotic perturbation on the current beetle to hope to help the search to jump out the local optimal solution. Eight different chaotic maps are used to demonstrate their impact on the simulation results. With six typical multimodal functions, performance comparisons between the CSOBAS and the SOBAS are conducted, validating the effectiveness of the CSOBAS and its superiority compared to the SOBAS. What?s more, the CSOBAS with an appropriate chaotic map can achieve a very good convergence quality compared to other swarm intelligence optimization algorithms while maintaining an individual.

A Study of Normalized Population Diversity in Particle Swarm Optimization

The values and velocities of a Particle swarm optimization (PSO) algorithm can be recorded as a series of matrix and its population diversity can be considered as an observation of the distribution of matrix elements. Each dimension is measured separately in the dimension-wise diversity. On the contrary, the element-wise diversity measures all dimensions together. In this chapter, the PSO algorithm is first represented in the matrix format. Then, based on the analysis of the relationship between pairs of vectors in the PSO solution matrix, different normalization strategies are utilized for dimension-wise and element-wise population diversity, respectively. Experiments on benchmark functions are conducted. Based on the simulation results of 10 benchmark functions (including unimodal/multimodal function, separable/non-separable function), the properties of normalized population diversities are analyzed and discussed.