In this paper, we experiment with novelty-based variants of OpenAI-ES, NS-ES and NSR-ES algorithms, and evaluate their effectiveness in training complex transformer-based architectures designed for the problem of reinforcement learning, such as Decision Transformers. We also test whether we can accelerate the novelty-based training of these larger models by seeding the training with a pretrained model. The experimental results were mixed. NS-ES showed progress, but it would clearly need many more iterations for it to yield interesting results. NSR-ES, on the other hand, proved quite capable of being straightforwardly used on larger models, since its performance appears as similar between the feed-forward model and Decision Transformer, as it is for the baseline objective-based OpenAI-ES.

Performance prediction has been a key part of the neural architecture search (NAS) process, allowing to speed up NAS algorithms by avoiding resource-consuming network training. Although many performance predictors correlate well with ground truth performance, they require training data in the form of trained networks. Recently, zero-cost proxies have been proposed as an efficient method to estimate network performance without any training. However, they are still poorly understood, exhibit biases with network properties, and their performance is limited. Inspired by the drawbacks of zero-cost proxies, we propose neural graph features (GRAF), simple to compute properties of architectural graphs. GRAF offers fast and interpretable performance prediction while outperforming zero-cost proxies and other common encodings. In combination with other zero-cost proxies, GRAF outperforms most existing performance predictors at a fraction of the cost.

This paper presents a definition of a generalized cosine-based fuzzy set, its application into simulation problems and uncertainty quantification in order to cover both aleatoric and epistemic uncertainties commonly seen in non-probabilistic models. The proposed cosine fuzzy set is able to represent different shapes via a parameter b which operates as the power of a cosine-type function adding flexibility to its behavior. Some special cases, some of its properties and a random variate generation method are described and applied into three examples: a random variate example and uncertainty quantification for two benchmark functions.

Some data mining, AI and data processing tasks might have data loss whose estimation/imputation is an important problem to be solved. Genetic algorithms are efficient and flexible global optimization methods able to deal with both multiple missing observations and multiple features such as continuous/discrete/binary data which are often found in multivariate databases unlike classical missing data estimation methods which only deal with univariate–continuous data. This paper presents a genetic algorithm to impute multiple missing observations in multivariate data which minimizes a new multi–objective (fitness) function based on the Minkowski distance of the means, variances, covariances and skewness between available/completed data. To do so, two sets of examples were tested: a continuous/discrete dataset which is compared to both the EM algorithm and auxiliary regressions, and a comparison over seven benchmark datasets.

Machine learning tools have been employed for problem solutions in bioinformatics. However, the parameters tuning of these models cam imply additional difficulties around the specific technique used to classify. In this work data from protein sequences was applied to three auto machine learning strategies to determine the type of mutation for the Neurofibromatosis disease. Results show that the parameters in the machine learning models were found automatically. In addition, these tools were relevant to determine relations between the amino-acids in the protein sequence.

This paper presents experimental results about the relationship between the Footprint Of Uncertainty (FOU) and the centroid of Interval Type-2 Fuzzy Numbers (IT2FN) which are a subclass of Type-2 fuzzy numbers. Four types of IT2FNs are analyzed: singleton-core triangular, interval-core triangular, singleton-core Gaussian and interval-core Gaussian IT2FNs. A quadratic relationship was detected, characterized, and their results are presented and discussed using numerical experiments.

Machine learning models exhibit vulnerability to adversarial examples i.e., artificially created inputs that become misinterpreted. The goal of this paper is to explore non-targeted black-box adversarial attacks on deep networks performing image classification. The original evolutionary algorithm for generating adversarial examples is proposed that employs a guided multi-objective search through the space of perturbed images. The efficiency of attacks is validated by experiments with the CIFAR-10 data set. The experimental results verify the usability of our approach against deep convolutional neural networks.

Hyperparameters optimization is one of the most important metalearning features that is used in AutoML systems. In this paper we use hyperparameters-space search algorithms to optimize the settings of supervised machine learning methods and workflows. We focus on changes in performance of hyperparameters optimization algorithms with the growing complexity of the hyperparameters-space, when using the data preprocessings adds more parameters to the configuration and thus more dimensions to the searched space.

This paper describes an algorithm solving the domain-independent problem of partial order simple task network planning using tree-based genetic programming. It provides several enhancements and tailored solutions of the standard genetic programming approach to improve the efficiency of the algorithm, including distance heuristic, local search, and equivalence of the individuals. The algorithm was tested on several experiments from the hierarchical planning domain in order to demonstrate its abilities, strengths and weaknesses.

Deep neural networks enjoy high interest and have become the state-of-art methods in many fields of machine learning recently.Still, there is no easy way for a choice of network architecture.However, the choice of architecture can significantly influence the network performance.This work is the first step towards an automatic architecture design.We propose a genetic algorithm for an optimization of a network architecture.The algorithm is inspired by and designed directly for the Keras library [1] that is one of the most common implementations of deep neural networks.The target application is the prediction of air pollution based on sensor measurements.The proposed algorithm is evaluated on experiments on sensor data and compared to several fixed architectures and support vector regression.

We present a parallel evolutionary algorithm with interleaving generations. The algorithm uses a careful analysis of genetic operators and selection in order to evaluate individuals from following generations while the current generation is still not completely evaluated. This brings significant advantages in cases where each fitness evaluation takes different amount of time, the evaluations are performed in parallel, and a traditional generational evolutionary algorithm has to wait for all evaluations to finish. The proposed algorithm provides better utilization of computational resources in these cases. Moreover, the algorithm is functionally equivalent to the generational evolutionary algorithm, and thus it does not have any evaluation time bias, which is often present in asynchronous evolutionary algorithms.

The problem of automated algorithm discovery has been mainly approached by means of Genetic programming. Recently, Monte-Carlo tree search methods - well known from games - have been used for program discovery, using stack-based program representations. In this paper, we analyze the behavior of the stack-based representations and describe an approach that provides finer control over generated program sizes and fast uniform playouts. Our approach utilizes type system with parametric polymorphism to generate typed programs. We evaluate the proposed solution with two Monte-Carlo tree search algorithms, and conclude that it is a good alternative which has a better control of exploration.

Kernel-based neural networks are popular machine learning approach with many successful applications. Regularization networks represent a their special subclass with solid theoretical background and a variety of learning possibilities. In this paper, we focus on single and multi-kernel units, in particular, we describe the architecture of a product unit network, and describe an evolutionary learning algorithm for setting its parameters including different kernels from a dictionary, and optimal split of inputs into individual products. The approach is tested on real-world data from calibration of air-pollution sensor networks, and the performance is compared to several different regression tools.

In decomposition based algorithms the quality of the resulting solutions depends on the weights used in the decomposition scheme. Usually the weights are generated in the beginning and remain fixed during the evolution, which may lead to poor distribution of solutions along the Pareto front. In this paper, we describe an extension of the popular MOEA/D algorithm which is able to tune the weights in order to find a set of solutions which maximizes a user specified objective. This adaptation is added as a new step to the algorithm which uses an approximation of the Pareto front to find suitable points in the objective space. These points are translated back into weights in such way to lead MOEA/D to find these points.

This paper describes an algorithm for the automated design of whole machine learning workflows, including preprocessing of the data and automatic creation of several types of ensembles. The algorithm is based on strongly typed genetic programming which ensures the validity of the workflows. The evolution of the individuals in the population is asynchronous in order to improve the utilization of computational resources. The approach is validated on four data sets from the UCI machine learning repository.

Metalearning deals with an important problem in machine-learning, namely selecting the right techniques to model the data at hand. In most of the metalearning approaches, a notion of similarity between datasets is needed. Our approach derives the similarity measure by combining arbitrary attribute similarity functions ordered by the optimal attribute assignment. In this paper, we propose a genetic programming based approach to the evolution of an attribute similarity inducing function. The function is composed of two parts - one describes the similarity of categorical attributes, the other describes the similarity of numerical attributes. Co-evolution is used to put these two parts together to form the similarity function. We use a repairing approach to guarantee some of the metric features for this function, and also discuss which of these features are important in metalearning.

The resulting set of solutions obtained by MOEA/D depends on the weights used in the decomposition. In this work, we use this feature to incorporate user preferences into the search. We use co-evolutionary approach to change the weights adaptively during the run of the algorithm. After the user specifies their preferences by assigning binary preference values to the individuals, the co-evolutionary step improves the distribution of weights by creating new (offspring) weights and selecting those that better match the user preferences. The algorithm is tested on a set of benchmark functions with a set of different user preferences.

Evolutionary algorithms are among the best multi-objective optimizers, but the large number of objective function evaluations they require makes it hard to use them to solve certain real-life tasks. In this work we present a surrogate-based local search for the multi-objective covariance matrix adaption evolution strategy (MO-CMA-ES). The local search is based on the estimation of hypervolume contribution of each individual and maximization of this contribution. This work extends our previous work and makes such surrogate models applicable to problems with more than two objectives. Moreover, it uses a unique feature of MO-CMA-ES to make the local search more effective. The results indicate that the algorithm can find solutions of the same quality as MO-CMA-ES while using 30-50 percent less objective function evaluations.

In this paper we present a new initialization method for genetic programming based on randomized exhaustive enumeration. It naturally enables complete sharing of sub trees among individuals which in turn allows an efficient reuse of computations. Moreover, it can be implemented as a random one pass initialization. We present experimental results on different instances of simple symbolic regression exploring the landscape of possible initializations based on our approach and confirming the usability of these initializations.

Tree based genetic programming (GP) traditionally uses simple S-expressions to represent programs, however more expressive representations, such as lambda calculus, can exhibit better results while being better suited for typed GP. In this paper we present population initialization methods within a framework of GP over simply typed lambda calculus that can be also used in the standard GP approach. Initializations can be parameterized by different search strategies, leading to wide spectrum of methods corresponding to standard ramped half-and-half initialization on one hand, or exhaustive systematic search on the other. A novel geometric strategy is proposed that balances those two approaches. Experiments on well known benchmark problems show that the geometric strategy outperforms the standard generating method in success rate, best fitness value, time consumption and average individual size.

Several contemporary multi-objective surrogate-based algorithms use some kind of local search operator. The search technique used in this operator can largely affect the performance of the multi-objective optimizer as a whole, however, little attention is often paid to the selection of this technique. In this paper, we compare three different local search techniques and evaluate their effect on the performance of two different surrogate based multi-objective optimizers. The algorithms are evaluated using the well known ZDT and WFG benchmark suites and recommendations are made based on the results.

Metalearning approach to the model selection problem - exploiting the idea that algorithms perform similarly on similar datasets - requires a suitable metric on the dataset space. One common approach compares the datasets based on fixed number of features describing the datasets as a whole. The information based on individual attributes is usually aggregated, taken for the most relevant attributes only, or omitted altogether. In this paper, we propose an approach that aligns complete sets of attributes of the datasets, allowing for different number of attributes. By supplying the distance between two attributes, one can find the alignment minimizing the sum of individual distances between aligned attributes. We present two methods that are able to find such an alignment. They differ in computational complexity and presumptions about the distance function between two attributes supplied. Experiments were performed using the proposed methods and the results were compared with the baseline algorithm.

Lambda calculus representation of programs offers a more expressive alternative to traditional S-expressions. In this paper we discuss advantages of this representation coming from the use of reductions (beta and eta) and a way to overcome disadvantages caused by variables occurring in the programs by use of the abstraction elimination algorithm. We discuss the role of those reductions in the process of generating initial population and propose two novel crossover operations based on abstraction elimination capable of handling general form of typed lambda term while being a straight generalization of the standard crossover operation. We compare their performances using the even parity benchmark problem.

This paper describes a surrogate based multi-objective evolutionary algorithm with hyper-volume contribution-based local search. The algorithm switches between an NSGA-II phase and a local search phase. In the local search phase, a model for each of the objectives is trained and CMA-ES is used to optimize the hyper-volume contribution of each individual with respect to its two neighbors on the non-dominated front. The performance of the algorithm is evaluated using the well known ZDT and WFG benchmark suites.

In surrogate evolutionary algorithms, usually the type of surrogate model is chosen beforehand, and it is never changed during the run of the evolution. Moreover, the reasoning why a particular type of model was chosen is often missing. In this paper, we present a framework which in each generation selects the most suitable surrogate from a set of models based on some pre-defined criteria. The results based on different types of model selectors are compared, and the dynamics of the evolution together with the change of the selected model type during the run of the evolutionary algorithm are discussed.

With the growing amount of data available in today's world, the emphasis is laid on the automatic configuration of data analysis - metal earning. This paper elaborates one of the metal earning sub problems, the data mining method recommendation. Based on a metric over the data features called metadata, we have proposed a solution exploiting clustering of datasets. The agglomerative algorithm is used to construct clustering over the metadata, and the average methods' performance is computed in each cluster. The ranking of data mining methods is then deduced from the classification of a dataset to a particular cluster. The recommendation algorithm, which is implemented within our data mining multi-agent system, has been tested in various configurations, and the results of these experiments have been compared.

We define a new class of 2D-stock cutting problems, the so called semi-quillotinable problems, and show its practical importance in solving the task of creating optimal cutting plans for a circular saw. Furthermore, we create a new algorithm suited for solving of semi-guillotinable problems by adapting existing evolutionary algorithms for both guillotinable and non-guillotinable 2D stock cutting problems. This algorithm is compared to standard algorithms on a selected set of both benchmark and real-life problems.

The paper presents a surrogate-based evolutionary strategy for multiobjective optimization. The evolutionary strategy uses distance based aggregate surrogate models in two ways: as a part of memetic search and as way to pre-select individuals in order to avoid evaluation of bad individuals. The model predicts the distance of individuals to the currently known Pareto set. The newly proposed algorithm is compared to other algorithms which use similar surrogate models on a set of benchmark functions.

The goal of our data-mining multi-agent system is to facilitate data-mining experiments without the necessary knowledge of the most suitable machine learning method and its parameters to the data. In order to replace the expertâs knowledge, the meta-learning subsystems are proposed including the parameter-space search and method recommendation based on previous experiments. In this paper we show the results of the parameter-space search with several search algorithms â" tabulation, random search, simmulated annealing, and genetic algorithm.

Most existing surrogate based evolutionary algorithms deal with only one model selected by the authors and different models are not considered. In this paper we propose a framework which enables automatic selection of types of surrogate models, and evaluate the effect of the type of selection on the overall performance of the resulting evolutionary algorithm. Two different types of model selection are tested and compared both in pre-selection scenario and in local search scenario.

In this paper we present a multiobjective evolutionary algorithm which uses surrogate models in two different ways -- during a local search and during pre-selection. Two different approaches to surrogate modeling are used, and the algorithm provides multiple individuals in each generation to enable easy parallelization. The algorithm is tested and compared to standard multiobjective evolutionary algorithms and to our previously developed surrogate evolution strategy. We also discuss the importance of the use of two different approaches and show that it improves the convergence speed significantly.

Evolutionary algorithms generally require a large number of objective function evaluations which can be costly in practice. These evaluations can be replaced by evaluations of a cheaper meta-model (surrogate model) of the objective functions. In this paper we present a novel distance based aggregate surrogate model for multiobjective optimization and describe a memetic multiobjective algorithm based on this model. Various variants of the models are tested and discussed and the algorithm is compared to standard multiobjective evolutionary algorithms. We show that our algorithm greatly reduces the number of required objective function evaluations.

In the field of multi-objective optimization there have been attempts to reduce the number of objective function evaluations by the use of surrogate models. However, in many-objective optimization, this work still has to be done to make the optimizers more practically usable. In this paper we show, that aggregate meta-models can be used even for the many-objective optimization and that they can also improve the performance of the many-objective optimizer. Moreover, meta-models are discussed from another point of view and compared to scalarization techniques in many-objective optimization. Two algorithms using our models are compared to IBEA on a set of selected benchmark functions with 5, 10, and 15 objectives.

In this paper, we formalize the role-based model of computational multi-agent system (MAS) solving data mining tasks in description logic (DL). MAS represents a promising approach for complicated data mining tasks due to its configuration flexibility and parallel processing. This is exploited in encapsulation of computational modeling, pre-processing and post-processing methods reflecting industrial standard specifications. The ontology of computational agents' capabilities is proposed and included in the model. The ontology agent, which implements the DL model, supports matchmaking, system integrity checking and automated creation of the system.

This paper deals with learning possibilities of regularization networks with product kernel units. Approximation problems formulated as regularized minimization problems with kernel-based stabilizers lead to solutions of the shape of linear combination of kernel functions. These can be expressed as one-hidden layer feed-forward neural network schemes, called regularization networks. We propose a novel evolutionary algorithm utilizing for regularization networks with product kernels. This algorithm utilizes genetic search for suitable network parameters as well as kernel functions.

The majority of multiobjective genetic algorithms is computationally expensive, therefore they often need to be parallelized before they can be used to solve practical tasks. Parallelization of multiobjective genetic algorithms is a relatively studied area, but no clearly winning approach has appeared yet. In this paper we present a novel parallel hybrid algorithm which combines multiobjective and single-objective genetic algorithms. We show that this algorithm can be successfully used to solve multiobjective optimization problems while outperforming more traditional parallel versions of multiobjective genetic algorithms.

In recent works concerning open multi-agent systems (MAS), the emphasis has been laid on organizational aspects of the development of agent societies. Many models of collaboration and cooperation of agents have been proposed which allow reusability of design patterns in MAS and separation of concerns. One of these approaches is the role-based model, inspired by the importance of roles in real human communities. In this paper we apply the concepts of role-based models on a concrete scenario in the field of Computational MAS.

A decision making architecture for intelligent agents is presented in this paper. It allows agents to perform autonomous behavior with respect to its own goals and cooperation with other agents. We focus on computational agents that encapsulate various methods of computational intelligence, such as neural networks, genetic algorithms, and similar methods. The goal of these agents is typically to solve a data mining problem characterized by a data set together with problem description. The architecture is based on the vertically-layered and belief-desire-intention architectures. Several experiments with computational agents were conducted to demonstrate the benefits of the architecture.

An emergence of intelligent behaviour within a simple robotic agent is studied in this paper. The radial basis function neural network is used as the control mechanism of the robot. Evolutionary algorithm is used to train the agent to perform several tasks. A comparison to multilayer perceptron neural networks and reinforcement learning is made and the results are discussed.

Abstract Web Services providing access to datasources with structured data have an important place in the SOA. In this paper we focus on modeling and discovery of generic data providing services (DPS), with the goal of making data providing services available for interactions with service requesters in contexts such as service composition and mediation. In our model RDF Views are used to represent the content provided by the DPS. A characterization of match between description of DPS as RDF Views and the OWL-S service request is specified, based on which we developed a flexible matchmaking algorithm for discovery of data providing services. Finally, we propose a realization of the DPS using a SOAP version of the SPARQL protocol and a dynamic configuration interface allowing easy interactions of service requesters with data providing services.

Intelligent embodied agents should be able to adopt to changes of the environment and to modify their behavior according to acquired knowledge. The goal of this work is the study of emergence of intelligent behavior within a simple intelligent agent. Cognitive agent functions will be realized by mechanisms based on neural networks of the perceptron type. The adaptation mechanism is realized by the evolutionary algorithms which is responsible for setting the weights of a neural network in a simulated environment. Several tasks including obstacle avoidance and efficient maze exploration are presented in the experimental section. The behaviors developed during the adaptation process compare favorably with hard coded strategies.

Hybrid models combining neural networks and genetic algorithms have been studied recently with the goal of achieving either better performance of the resulting network or faster training. In this paper we deal with variants of genetic learning applied for the structure optimization and weights evolution of multi-layer perceptron networks. Several genetic operators are tested, including memetic-type local search, that produce good results in terms of network performace. It is shown, that combining evolutionary algorithms with neural networks can lead to better results than relying on neural networks alone in terms of the quality of the solution (both training and generalization error). Comparison to gradient algorithms in terms of time complexity is discussed which does not bring overly optimistic results sometimes met in literature.

Rapidly growing amount of the data available on World Wide Web (WWW) as the ultimate collection of public accessible text documents complicates the task of acquiring usable and relevant information from raw data. Several methods of improvement and optimization of information retrieval (IR) systems providing user interaction with WWW documents have been introduced and discussed. In this paper, theory and application of genetic programming (GP) as an optimization method of modern user oriented IR and fuzzy IR (FIR) systems optimization of search queries is presented.

A functional equivalence of feedforward networks has been proposed to reduce the search space of learning algorithms. The description of equivalence classes has been used to introduce a unique parametrization property and consequently the so-called canonical parameterizations as representatives of functional equivalence classes. A novel genetic learning algorithm for neural networks that outperforms standard genetic learning has been proposed based on these results. In this paper we summarize previous results and present a geometrical approach that illustrates the situation and also leads to further open problems.

A functional equivalence of feed-forward networks has been proposed to reduce the search space of learning algorithms. The description of equivalence classes has been used to introduce a unique parametrization property and consequently the so-called canonical parameterizations as representatives of functional equivalence classes. A novel genetic learning algorithm for RBF networks and perceptrons with one hidden layer that operates only on these parameterizations has been proposed. Experimental results show that our procedure outperforms the standard genetic learning. An important extension of theoretical results demonstrates that our approach is also valid in the case of approximation.

A functional equivalence property of feedforward networks has been proposed to reduce the search space of learning algorithms. We summarize previous results, describing the form of functional equivalence for one-hidden-layer perceptron networks and radial basis function (RBF) networks with Gaussians. The description of equivalence classes is used in a proposition of a genetic learning algorithm which is tested on two simple problems and which outperforms the standard genetic learning procedure.