Petr Bujok

Adaptive Variants of Differential Evolution: Towards Control-Parameter-Free Optimizers

Seven up-to-date adaptive variants of differential evolution were compared in six benchmark problems of two levels of dimension (D = 30 and D = 100). The opposition-based optimization was also implemented to each adaptive variant and compared in experiments. It was found that all the algorithms perform very reliably in the problems of D = 30, whereas their reliability rate in the problems of D = 100 differs substantially among the test problems. Only two algorithms (JADE and b6e6rl variant of competitive DE) operate with acceptable reliability in all the problems. Considering the computational costs, the rank of the algorithms is different in various problems. When the average performance over all the problems is taken into account, JADE was the most efficient and b6e6rl the most reliable. The implementation of opposition-based optimization into adaptive variants of differential evolution does not increase the reliability and its positive influence on the efficiency is rare. Based on the results, recommendations to application of adaptive algorithms are formed and the source code of the algorithms is available online.

Controlled restart in differential evolution applied to CEC2014 benchmark functions

A controlled restart in differential evolution (DE) is proposed. The conditions of restart are derived from the difference of maximum and minimum values of the objective function and the estimated maximum distance among the points in the current population. The restart is applied in a competitive-adaptation variant of DE. This DE algorithm with the controlled restart is used in the solution of the benchmark problems defined for the CEC 2014 competition. Two control parameters of restart are set up intuitively. The population size, which is the only control parameter of competitive-adaptation variant of DE, is set up to the values based on a short preliminary experimentation.

Differential evolution with rotation-invariant mutation and competing-strategies adaptation

A new variant of the adaptive differential evolution algorithm was proposed and tested experimentally on the CEC 2014 test suite. In the new variant, the adaptation is based on the competition of several strategies. A part of strategies in the pool uses the rotation-invariant current-to-pbest mutation in the novel algorithm. The aim of the experimental comparison was to find whether the presence of the rotation-invariant strategy is able to improve the efficiency of the differential evolution algorithm, especially in problems with rotated objective functions. The results of the experiments showed that the new variant performed well in a few of the test problems, while no apparent benefit was observed in the majority of the benchmark problems.

L-SHADE with competing strategies applied to CEC2015 learning-based test suite

Successful adaptive variant of differential evolution, the Success-history based parameter adaptation of Differential Evolution using linear population size reduction algorithm (L-SHADE), was improved. Adaptive mechanisms used in the algorithm were joined with adaptive mechanism proposed for competitive differential evolution algorithm. Four strategies, including the original one and strategies with exponential crossover, compete in the new LSHADE44 algorithm. The proposed algorithm is applied to the benchmark set defined for Learning-based case of Special Session and Competitions on Real-Parameter Single Objective Optimization on CEC2016. According to preliminary experiments, the proposed algorithm with competing strategies outperformed the original L-SHADE in the most of the test problems.

SYNCHRONOUS AND ASYNCHRONOUS MIGRATION IN ADAPTIVE DIFFERENTIAL EVOLUTION ALGORITHMS

The influence of synchronous and asynchronous migration on the per- formance of adaptive differential evolution algorithms is investigated. Six adaptive differential evolution variants are employed by the parallel migration model with a star topology. Synchronous and asynchronous migration models with various parameters settings were experimentally compared with non-parallel adaptive al- gorithms in six shifted benchmark problems of dimension D = 30. Three different ways of exchanging individuals are applied in a synchronous island model with a fixed number of islands. Three different numbers of sub-populations are set up in an asynchronous island model. The parallel synchronous and asynchronous migration models increase performance in most problems.

Parallel migration model employing various adaptive variants of differential evolution

The influence of migration on the performance of differential evolution algorithm is studied. Six adaptive variants of differential evolution are applied to a parallel migration model with a star topology. The parallel algorithm with several different settings of parameters controlling the migration was experimentally compared with the adaptive serial algorithms in six benchmark problems of dimension D=30. The parallel algorithm was more efficient than the best serial adaptive DE variant in a half of the problems.

Unconventional modelling of complex system via cellular automata and differential evolution

Abstract The article deals with principles and utilization possibilities of cellular automata and differential evolution within task resolution and simulation of an epidemic process. The modelling of the spread of epidemics is one of the most widespread and commonly used areas of a modelling of complex systems. The origins of such complexity can be investigated through mathematical models termed ‘cellular automata’. Cellular automata consist of many identical components, each simple, but together capable of complex behaviour. They are analysed both as discrete dynamical systems, and as information-processing systems. Cellular Automata (CA) are well known computational substrates for studying emergent collective behaviour, complexity, randomness and interaction between order and chaotic systems. For the purpose of the article, cellular automata and differential evolution are recognized as an intuitive modelling paradigm for complex systems. The proposed cellular automata supports to find rules of the transition function that represents the model of a studied epidemic. Search for models a studied epidemic belongs to inverse problems whose solution lies in a finding of local rules guaranteeing a desired global behaviour. The epidemic models have the control parameters and their setting significantly influences the behaviour of the models. One way how to get proper values of the control parameters is use evolutionary algorithms, especially differential evolution (DE). Simulations of illness lasting from one to ten days were performed using both described approaches. The aim of the paper is to show a course of simulations for different rules of the transition function and how to find a suitable model of a studied epidemic in the case of inverse problems using a sufficient amount of local rules of a transition function.

Evaluating the performance of L-SHADE with competing strategies on CEC2014 single parameter-operator test suite

A new variant of differential evolution algorithm is proposed. The new variant is a modification of the success-history based parameter adaptation of differential evolution using linear population size reduction (L-SHADE). In the newly proposed variant, adaptive mechanism of competing strategies is added. Four different strategies combining two kinds of mutation and two types of crossover compete in generating the new trial points. The selection of the strategy to be used in the current step is based on the success in previous search steps. The proposed algorithm is applied to the benchmark set defined for Single parameter-operator set based case of Special Session and Competitions on Real-Parameter Single Objective Optimization on CEC2016. According to preliminary experiments curried out on a different benchmark set, the proposed algorithm with competing strategies outperformed the original L-SHADE. However, the performance of the proposed algorithm on the benchmark set of Single parameter-operator set based case of Special Session and Competitions on Real-Parameter Single Objective Optimization on CEC2016 is not so much higher than the performance of the original L-SHADE algorithm.

Analysis of attackers against windows emulating honeypots in various types of networks and regions

The paper is devoted to an analysis of a one-year-long period of operation of a honeynet composed of 6 Dionaea honeypots emulating Windows services. The analysis focused on the frequency of attacks according to the location of individual honeypots (sensors) as well as to the geographical location of attackers. From the statistical processing of the results, it was demonstrated that the most frequently attacking malware was well-known Conficker worm. Moreover, attacking OS were studied with the conclusion that Windows is the most frequent OS. Regarding the geographical location of the attackers, several non-western countries and autonomous systems were indicated as being the most frequent origin of the attacks.

Adaptive Differential Evolution: SHADE with Competing Crossover Strategies

Possible improvement of a successful adaptive SHADE variant of differential evolution is addressed. Exploitation of exponential crossover was applied in two newly proposed SHADE variants. The algorithms were compared experimentally on CEC 2013 test suite used as a benchmark. The results show that the variant using adaptive strategy of the competition of two types of crossover is significantly more efficient than other SHADE variants in 7 out of 28 problems and not worse in the others. Thus, this SHADE with competing crossovers can be considered superior to original SHADE algorithm.