The cross-entropy loss is widely used in neural networks for classification problems, however it may not be the best choice. In this study the problem of designing new loss function is considered, in particular the Taylor polynomial is applied. To tune the parameters of the polynomial, differential evolution is used. As the fitness evaluation requires training a neural network, it is a computationally expensive process. To overcome this problem, an approach based on surrogate models is used, in particular the Kriging method applied. Experiments on the CIFAR-10 test set have shown that the surrogate approach accelerates the search for the optimal solution, and also allows achieving better accuracy in comparison with the cross-entropy or the coefficient optimization method without using the surrogate approach. The found loss function curve has significantly different shape compared to the cross-entropy, and has two minima at probability of correct prediction equal to 0 and 1.

Constrained numerical optimization problems introduce significant challenges for optimization methods. One of the popular heuristic optimization techniques for such problems is the Differential Evolution algorithm. This study focuses on applying a variant of differential evolution with two populations to the set of benchmark problems from the CEC 2024 Constrained Single Objective Numerical Optimization competition. In the proposed CL-SRDE algorithm the adaptation of the scaling factor is performed based on the success rate, which is the number of replaced individuals during selection divided by population size. The constraints are handled by a modified epsilon-constraint method. The analysis of the experimental results show that the used adaptation scheme achieves better feasibility rates and overall performance, compared to the alternative approaches.

One of the most important problems of the Differ-ential Evolution algorithm is the adaptation of scaling factor, due to high sensitivity to this parameter. In this paper the L-SRTDE algorithm is proposed, where the scaling factor is set based on the ratio of improved solutions at each generation, i.e. success rate. The proposed algorithm is used to solve the benchmark problems from the CEC 2024 Bound Constrained Single Objective Numerical Optimization competition. Additional experiments are performed using the CEC 2017 and CEC 2022 benchmarks. The analysis of numerical results, considering both accuracy and speed of the algorithms shows that the proposed approach is capable of outperforming many alternative approaches, which use success history-based adaptation.

In this paper the adaptive differential evolution algorithm is presented, which includes a set of concepts, such as linear bias change in parameter adaptation, repetitive generation of points for bound constraint handling, as well as non-linear population size reduction and selective pressure. The proposed algorithm is used to solve the problems of the CEC 2022 Bound Constrained Single Objective Numerical Optimization bench-mark problems. The computational experiments and analysis of the results demonstrate that the NL-SHADE-LBC algorithm presented in this study is able to demonstrate high efficiency in solving complex optimization problems compared to the winners of the previous years' competitions.

This paper proposes a variant of the adaptive differential evolution algorithm, which combines several important concepts, including non-linear population size reduction, rank-based selective pressure in the mutation strategy, adaptive archive set usage, as well as a set of rules to control crossover rate. The developed approach is applied to solve the CEC 2021 Bound Constrained Single Objective Optimization Parametrized Benchmark problems. The performed computational experiments and their statistical analysis show that the proposed NL-SHADE-RSP algorithm is capable of demonstrating high efficiency of fining solutions to biased, shifted and rotated functions compared to other state-of-the-art algorithms, including the winners of the previous competitions.

Solving single objective constrained real-parameter optimization problems via population-based algorithms has attracted much attention. In this paper, a new self-tuning meta-heuristic approach called Fuzzy Controlled Cooperative Heterogeneous Algorithm (FCHA), which was proposed for constrained optimization, is introduced. The developed approach combines competition and cooperation between biology-inspired and evolutionary algorithms, regulated by fuzzy controller. It should be noted, that the epsilon-constrained method is utilized to handle the constraints for the solved optimization problems. The performance of the proposed FCHA algorithm is evaluated on 57 real-world constrained problems submitted for CEC 2020 special session. Its workability and usefulness are demonstrated; also ways of algorithm improvement are discussed.

The single-objective numerical optimization is an important research field due to variety of real world applications. One of the most promising classes of numerical optimization algorithms is Differential Evolution. This paper proposes a new algorithm called RASP-SHADE to solve the CEC 2020 Bound Constrained Single Objective Optimization benchmark problems. The developed algorithm is based on the L-SHADE with Distance-based success history adaptation, incudes parameter adaptations of the jSO algorithm, and introduces several novelties. The ranking of population and archive according to fitness introduces the selective pressure, resulting in a new mutation strategy. A new archive update rule is applied with replacing only worst points and the parameters sampling scheme is changed. The experiments show that RASP-SHADE modifications result in significant improvements when compared to other state-of-the-art algorithms.

The three-stage development of a cyber-physical speech-controlled wheelchair for disabled persons is described. Initially, a small prototype was built to test the feasibility of using cloud-based speech recognition systems for real-time wheelchair maneuvering. In the second stage, the full-sized prototype was built and tested in a laboratory and in a clinical environment. In the third stage, the full-scale prototype was equipped with distance sensors and an advanced control algorithm for semi-autonomous drive. The system architecture is described with the important addition of edge computing for speech recognition. Six cloud speech recognition services and two offline services were used, as using multiple speech recognition systems improves system reliability and latency. The software and hardware technologies are described, in addition to an innovative application of multiple cloud/edge systems for wheelchair motion control.

Various data-driven models are often involved in epidemiological studies, wherein the availability of data is constantly increasing. Accurate and, at the same time, interpretable models are preferable from the practical point of view. Finding simple and compact dependences between predictors and outcome variables makes it easier to understand necessary interventions and preventive measures. In this study, we applied a Fuzzy Logic-based model, which meets these requirements, to predict the coronary heart disease (CHD) progression during a 30-year follow-up. The Fuzzy Logic-based model was automatically designed with an ad hoc Genetic Algorithm using the data from the Kuopio Ischemic Heart Disease Risk Factor (KIHD) Study, a Finnish cohort of 2682 men who were middle-aged at baseline in 1980s. Using cross-validation, we found out that the sample from the KIHD study is heterogeneous and after filtering out 10% of outliers, the predictive accuracy increased from 65% to 73%. The generated rule bases include 19 fuzzy rules on average with maximum 7 variables in one rule from the initial set of 638 predictor variables. The selected predictors of CHD progression are informative and diverse representing physical aspects, behavior, and socioeconomics. The Fuzzy Logic-based model creates a comprehensive set of predictors that enables us to better understand the complexity of illnesses and their progression. Moreover, the Fuzzy Logic-based model has potential to provide tools to analyse and deal with heterogeneity in large cohorts.

Genetic algorithm is a popular optimization method for solving binary optimization problems. However, its efficiency highly depends on the parameters of the algorithm. In this study the success history adaptation (SHA) mechanism is applied to genetic algorithm to improve its performance. The SHA method was originally proposed for another class of evolutionary algorithms, namely differential evolution (DE). The application of DE’s adaptation mechanisms for genetic algorithm allowed significant improvement of GA performance when solving different types of problems including binary optimization problems and continuous optimization problems. For comparison, in this study, a self-configured genetic algorithm is implemented, in which the adaptive mechanisms for probabilities of choosing one of three selection, three crossover and three mutation types are implemented. The comparison was performed on the set of functions, presented at the Congress on Evolutionary Computation for numerical optimization in 2017. The results demonstrate that the developed SHAGA algorithm outperforms the self-configuring GA on binary problems and the continuous version of SHAGA is competetive against other methods, which proves the importance of the presented modification.

The Differential Evolution (DE) is a highly competitive numerical optimization algorithm, with a small number of control parameters. However, it is highly sensitive to the setting of these parameters, which inspired many researchers to develop adaptation strategies. One of them is the popular Success-History based Adaptation (SHA) mechanism, which significantly improves the DE performance. In this study, the focus is on the choice of the metaparameters of the SHA, namely the settings of the Lehmer mean coefficients for scaling factor and crossover rate memory cells update. The experiments are performed on the LSHADE algorithm and the Congress on Evolutionary Computation competition on numerical optimization functions set. The results demonstrate that for larger dimensions the SHA mechanism with modified Lehmer mean allows a significant improvement of the algorithm efficiency. The theoretical considerations of the generalized Lehmer mean could be also applied to other adaptive mechanisms.

Solving single-objective real-parameter optimization problems can still cause difficulties, for example if the optimized function is multimodal or has rotated trap problems. Such optimization problems can be found in various areas in real-world applications. Usually, these problems are very complex and computationally expensive. A new algorithm, which is a modification of the LSHADE algorithm with a rank-based selective pressure strategy, called LSHADE-RSP, is presented in this paper. The proposed algorithm is a new variant of the LSHADE algorithm, the basic idea of which consists in the adaptation of its mutation strategy using selective pressure. The experiments were performed on CEC 2018 benchmark functions. A comparison of the proposed LSHADE-RSP algorithm and the algorithm-participants of the CEC 2017 competition is presented. From the obtained results it can be concluded that LSHADE-RSP performs better in comparison with most alternative algorithms: using the CEC 2018 evaluation method, LSHADE-RSP obtained one of the best final scores among the algorithms that were winners of the previous competition.

Automatically generated fuzzy rule-based classifiers for opinion mining are presented in this paper. A collective nature-inspired self-tuning meta-heuristic for solving unconstrained real-valued optimization problems called Co-Operation of Biology Related Algorithms and its modification with a biogeography migration operator for binary-parameter optimization problems were used for the design of classifiers. The basic idea consists in the representation of a fuzzy classifier rule base as a binary string and the parameters of the membership functions of the fuzzy classifier as a string of real-valued variables. Three opinion mining problems from the DEFT’07 competition were solved using the proposed classifiers. Experiments showed that the fuzzy classifiers developed in this way outperform many alternative methods at the given problems. The workability and usefulness of the proposed algorithm are confirmed.

The paper describes the development of educational Cyber-Physical Robotic Platforms, remotely controlled via cloud technologies. The platform is implemented using Parallax ActivityBot kit (only for mechanical part), controlled by Arduino, and includes a Quad core ARM-processor Mini PC MK802 V5 LE running Xubuntu Linux. The programming on the developed platform could be performed in JavaScript and HTML, providing the web interface for controlling the system through Wi-Fi. Programming the platform is possible through the Cloud9 IDE web interface, enabling rewriting the code or running different programs by the user. Four equal platforms were implemented to address the need for the easily accessible educational Cyber-Physical Robotic Platforms / Internet of Things hardware for students and tested in the experiment.

A novel approach for active training example selection in classification problems is presented. This active selection of training examples is designed to decrease the amount of computation resources required and increase the classification quality achieved. The approach changes the training sample during the evolutionary process so that the algorithm concentrates on problematic instances that are hard to classify. A fuzzy classifier designed with a self-configuring modification of a hybrid evolutionary algorithm is applied as a classification problem solver. The benchmark containing 9 data sets from KEEL is used to prove the usefulness of the approach proposed.

This paper addresses the problem of the hierarchical manpower system control in the restructuring process. The restructuring case study is described where eight topmost ranks are considered. The desired and actual structure of the system is given by the actual numbers of men in a particular rank. The system was modelled in the dicrete state space with state elements and flows representing the recruitment, wastages and retirements. The key issues were identified in the process as the stating of the criteria function, which are time variant boundaries on the parameter values, the chain stucture of the system and the tendency for the system to oscilate at given initial conditions. The oscillatory case is presented and the dynamic programming approach was considered in the optimization as unsuitable, examining the oscillations. The boundary space and optimal solution space were considered by indicating the small area where the solution could be optimal. The augmented finite automaton was defined which was used in the optimization with the adaptive genetic algorithm. The developed optimization method enabled us to successfully determine proper restructuring strategy for the defined manpower system.

Self-configuring evolutionary algorithm of fuzzy rule bases automated deign for solving classification problems, which combines Pittsburgh and Michigan approaches, is introduced. The evolutionary algorithm is based on the Pittsburgh approach where every individual is a rule base and the Michigan approach is used as a mutation operator. A self-configuration method is used to adjust probabilities of the usage of selection, mutation and Michigan part operators. Testing the algorithm on a number of real-world problems demonstrates its efficiency comparing to several other commonly used approaches.