We propose to extend the probability model of UMDA by means of random affine maps on bit vectors. To limit the complexity, the bit matrix of affine maps is represented as short sequences of simple and reversible bit operations called transvections. We consider two versions of UMDA with random affine maps (RAMU) updating one or two probability vectors. We provide fixed-budget experiments to explore the behavior of RAMU and compare it with UMDA.
Dispatching rules are simple heuristics primarily used for solving dynamic scheduling problems. Since designing such rules manually is time-consuming and requires expert knowledge, genetic programming is employed to develop them automatically. However, the quality of those automatically developed rules is limited and must therefore be further improved by integrating genetic programming with more sophisticated methods, most notably local search. This paper explores the potential of substituting local search algorithms with repeated applications of genetic operators, specifically mutation and crossover, within genetic programming to obtain better performing dispatching rules for the unrelated machines scheduling environment. Experimental results indicate that certain configurations of repeated genetic operators significantly outperform the baseline genetic programming approach and can be comparable or, at times, superior to the variants enhanced by local search. This finding indicates a significant advancement in evolutionary algorithm design for scheduling problems, proposing the strategic repetition of genetic operators as a potentially simpler yet effective alternative to more complex local search strategies.
In the course of apprehending of biological system, the modeling process of gene interactions plays a crucial rocketing role. Unfortunately, the main bottleneck of the modeling process is always the determination of the model parameters. This study focuses on the problem of identifying Gene Regulatory Network (GRN) variables in a discrete framework, represented by gene product concentration thresholds, that separate discrete states of genes contained in the GRN. We propose to compute thresholds from graphs representing interactions between biological genes, gene product concentrations experimentally measured by biologists, and observable behaviors. In this setting, we have developed some adaptations of bio-inspired methods to assist modelers and to propose compatible models to biologists. Since the parameter identification problem is constrained, in this article we focus on adapting and comparing three different heuristics of the CEC’2020 competition for solving single-objective constrained problems using the aforementioned bio-inspired methods by defining a dedicated fitness. To validate our approach, we used an abstract model of the cell cycle and found the performances of the three heuristics consistent with the results of the CEC’2020 competition. This serves as a proof of concept for the development of methods to identify parameters of GRN models using available data and relying less on biological expertise.
The covariance matrix adaptation evolution strategy (CMA-ES) is one of the most widely used evolutionary algorithms in single-objective black-box optimisation. Unlike the popularity of the single-objective CMA-ES, its multi-objective variant - MO-CMA-ES - has not been extensively studied. In MO-CMA-ES, like CMA-ES the step size, which determines how far the mutation can go, depends on whether a newly generated solution is better than its parent in the population. However, unlike single-objective optimisation where all solutions are distinguishable about their objective values, in multi-objective optimisation there may exist many solutions incomparable (i.e., non-dominated to each other). Updating the step size based on the qualitative comparison between solutions may not work best for the multi-objective case. In this paper, we propose a simple step-size update strategy for MO-CMA-ES. The proposed strategy considers how much change the newly generated solution has compared with its parent. Specifically, we factor in the difference in the non-domination levels that the offspring and parent solutions are located in, attempting to make use of quantitative information that may better reflect the current search progress. Experimental results show the effectiveness of the proposed strategy - it can accelerate the convergence speed on almost all the test problems considered.
Local optima networks (LONs) are a compact graph-based visualisation and characterisation approach to represent the fitness landscape of an optimisation problem. LONs are most frequently generated for combinatorial problems, where neighbourhoods can be crisply defined. A handful of approaches exist for continuous spaces, e.g. using derivatives-based local search approaches and basin-hopping, or by discretising the continuous space via gridding. Here we propose a new approach. Neighbourhoods are defined as balls around quasi-random samples from the search space. Basins and local optima are identified by greedily traversing these sampled neighbourhoods. One interpretation of such a formulation is that it approximates the LON induced by a (1+λ ) –Evolution Strategy (ES) with a ball mutation. The proposal allows the generation of LONs for problems with non-linear constraints, and discontinuous functions. We detail construction methods and illustrate the effective landscapes for some different objective functions using both the proposed methodology and derivatives-based alternatives.
