This paper introduces LMAP (Local PCA Models with Global MDS Embeddings), a geometric method for nonlinear dimensionality reduction that combines local PCA-based tangent charts with global MDS alignment to obtain smooth embeddings with coherent local and global structure.Landmark points define locally linear models that approximate the manifold's tangent geometry, while classical multidimensional scaling aligns these charts into a consistent low-dimensional representation.The resulting atlas admits a closed-form out-of-sample extension via weighted blending of multiple tangent charts, yielding a continuous and reproducible mapping from the ambient space to the embedding.Experiments on synthetic manifolds analyze the influence of landmark density and neighborhood size and show that LMAP produces globally consistent embeddings that bridge the gap between linear PCA and stochastic neighbor-based methods, achieving low global distortion while maintaining reliable trustworthiness and out-of-sample stability.

Fitness evaluations often dominate the runtime of evolutionary algorithms (EAs), yet most runtime analyses assume that every offspring is evaluated on the true objective function.This work presents a unified runtime-theoretic framework for surrogate-assisted evolutionary algorithms that selectively schedule true evaluations based on predictive information.The framework characterizes expected optimization time directly in terms of evaluation frequency and improvement probabilities, abstracting from specific surrogate implementations.Focusing on a surrogate model with imperfect prediction accuracy, we show that linear Θ(n) scaling of true fitness evaluations can be achieved on OneMax when accuracy remains bounded away from zero and evaluations are performed periodically with logarithmic frequency.Simulation results confirm this prediction and clearly separate the surrogate-assisted process from the classical Θ(n log n) behavior of the (1+1)-EA.

We propose cognitive prompting as a novel approach to guide problem-solving in large language models (LLMs) through structured, human-like cognitive operations, such as goal clarification, decomposition, filtering, abstraction, and pattern recognition.By employing systematic, step-by-step reasoning, cognitive prompting enables LLMs to tackle complex, multi-step tasks more efficiently.We introduce three variants: a deterministic sequence of cognitive operations, a self-adaptive variant in which the LLM dynamically selects the sequence of cognitive operations, and a hybrid variant that uses generated correct solutions as few-shot chain-of-thought prompts.Experiments with LLaMA, Gemma 2, and Qwen models in each two sizes on the arithmetic reasoning benchmark GSM8K demonstrate that cognitive prompting significantly improves performance compared to standard question answering.

This paper introduces LLaMA-ES, an approach for tuning the hyperparameters of Evolution Strategies (ES), specifically the Covariance Matrix Adaptation Evolution Strategy (CMA-ES), by leveraging a Large Language Model (LLM).The proposed method uses the LLM to iteratively suggest parameter adjustments based on the optimization history, enabling dynamic fine-tuning of the algorithm.We validate our approach through experiments on numerical benchmark optimization problems, employing the LLaMA3 model with 70 billion parameters.The results demonstrate that LLaMA-ES significantly enhances the performance of CMA-ES, achieving competitive results in parameter tuning and demonstrating the potential of LLMs in optimization tasks.

This paper introduces an approach that integrates selfadaptive Evolution Strategies (ES) with Large Language Models (LLMs) to enhance the explainability of complex optimization processes.By employing a self-adaptive ES equipped with a restart mechanism, we effectively navigate the challenging landscapes of benchmark functions, capturing detailed logs of the optimization journey.The logs include fitness evolution, step-size adjustments and restart events due to stagnation.An LLM is then utilized to process these logs, generating concise, user-friendly summaries that highlight key aspects such as convergence behavior, optimal fitness achievements, and encounters with local optima.Our case study on the Rastrigin function demonstrates how our approach makes the complexities of ES optimization transparent.Our findings highlight the potential of using LLMs to bridge the gap between advanced optimization algorithms and their interpretability.

Evolution Strategies (ES) have emerged as a powerful and effective method for optimization and reinforcement learning tasks, largely due to their simplicity and scalability.However, current ES techniques can be limited in their capacity to quickly converge on the optimal solution.In this paper, we propose a novel approach to enhance ES by incorporating an evolution path-informed bias in the Gaussian mutation operator.This bias is designed to facilitate faster descent on decreasing functions.Our method leverages the evolution path, which represents the historical search directions, to intelligently bias the Gaussian mutation.By doing so, it enables the algorithm to be more sensitive to the underlying function's structure and adaptively exploit this information for more efficient exploration.We validate our approach through experiments on three benchmark functions: a linear function, we call Downhill function here, a Parabolic ridge, and a Sphere function.The results demonstrate that our evolution path-informed bias significantly accelerates convergence on in most of the cases.

With growing environmental awareness, power generation from wind and other renewable sources is becoming increasingly important.Accurate short-term predictions of wind turbine power are needed to keep the grid stable and secure.This paper investigates the use of ETSformer, a time series approach based on the transformer architecture, for wind power prediction.ETSformer incorporates exponential smoothing principles and introduces mechanisms such as exponential smoothing attention and frequency attention to improve accuracy, efficiency and interpretability.This study compares ETSformer and LSTM on a sample dataset of a wind farm and its surrounding sites within a three kilometer radius from the Wind Integration National Dataset Toolkit with five minute interval measurements.The investigation shows promising results and improvements of ETSformer in ultra-short and short-term wind power prediction.

With the rise of A.I. methods the demand for efficient optimization methods that are easy to implement and use increases.This paper introduces a simple optimization method for numerical blackbox optimization.It proposes to apply covariance matrix estimation for the (1+1)-ES with Rechenberg's step size control.Experiments on a small set of benchmark functions demonstrate that the approach outperforms its isotropic variant allowing competitive convergence on problems with scaled and correlated dimensions.

For in silico drug discovery various representations have been established regarding storing and processing molecular data. The choice of representation has a great impact on employed methods and algorithms. Molecular fingerprints in the form of fixed-size bit vectors are a widely used representation which captures structural features of a molecule and enables a straight-forward way of estimating molecule similarities. However, since fingerprints are not invertible, they are rarely utilized for molecule generation tasks. This study presents an approach to the reconstruction of molecules from their fingerprint representation that is based on genetic algorithms. The algorithm assembles molecules from BRICS fragments and therefore only generates valid molecular structures. We demonstrate that the genetic algorithm is able to construct molecules similar to the specified target, or even reconstruct the original molecule. Furthermore, to illustrate how this genetic algorithm unlocks fingerprints as a representation for other in silico drug discovery methods, a novel Transformer neural language model trained on molecular fingerprints is introduced as a molecule generation model.

Multi-task learning is an approach to reduce the amount of required training data by learning multiple tasks at the same time.In the context of neural networks, multi-task learning is performed by sharing weights or creating dependencies between weights of task-specific networks.In this work, we propose an algorithm that uses a simple evolutionary algorithm, which is able to match and also surpass learned weight sharing.We evaluate the performance of this method on CIFAR-100, cast as a multi-tasking problem, using an 18-layer residual network, and compare our results to literature.

This work introduces an autoregressive generative model for graphs which is based on the transformer architecture and applied to the domain of molecular graph generation.Utilizing the multi-head self-attention mechanism to directly model distributions over atoms and bonds, it can sample new molecular graphs in an autoregressive manner.The benchmark framework MOSES is used to compare the proposed approach to other state-of-the-art molecule generation models.It is shown that the model is capable of generalizing from the training data to generate novel and realistic molecules.

Generative models play an important role in the discovery of new molecules. However, these models often operate on a string or graph based representation of molecules, without taking spatial information into consideration. In this paper we introduce an autoregressive generative neural network operating directly in the three-dimensional space of atoms. By leveraging the powerful multi-head self-attention mechanism of the transformer architecture the model predicts a set of atoms and a corresponding distance matrix to generate molecules in a translation and rotation invariant manner. We demonstrate that the model can be utilized to generate completely new molecules that resemble the training data in their structural properties. Furthermore, the model is capable of completing unconnected sets of atoms to valid molecular structures.

In deep multi-task learning, weights of task-specific networks are shared between tasks to improve performance on each single one. Since the question, which weights to share between layers, is difficult to answer, human-designed architectures often share everything but a last task-specific layer. In many cases, this simplistic approach severely limits performance. Instead, we propose an algorithm to learn the assignment between a shared set of weights and task-specific layers. To optimize the non-differentiable assignment and at the same time train the differentiable weights, learning takes place via a combination of natural evolution strategy and stochastic gradient descent. The end result are task-specific networks that share weights but allow independent inference. They achieve lower test errors than baselines and methods from literature on three multi-task learning datasets.

Choosing the best hyperparameters for neural networks is a big challenge. This paper proposes a method that automatically initializes and adjusts hyperparameters during the training process of stacked autoencoders. A population of autoencoders is trained with gradient-descent-based weight updates, while hyperparameters are mutated and weights are inherited in a Lamarckian kind of way. The training is conducted layer-wise, while each new layer initiates a new neuroevolutionary optimization process. In the fitness function of the evolutionary approach a dimensionality reduction quality measure is employed. Experiments show the contribution of the most significant hyperparameters, while analyzing their lineage during the training process. The results confirm that the proposed method outperforms a baseline approach on MNIST, FashionMNIST, and the Year Prediction Million Song Database.

In order to improve particle swarm optimization (PSO) to tackle dynamic optimization problems, various strategies have been introduced, e. g., random restart, memory, and multi-swarm approaches. However, literature lacks approaches based on prediction. In this paper we propose three different PSO variants employing a prediction approach based on recurrent neural networks to adapt the swarm behavior after a change of the objective function. We compare the variants in an experimental study to a PSO algorithm that is solely based on re-randomization. The experimental study comprises the moving peaks benchmark and dynamic extensions of the Sphere, Rastrigin, and Rosenbrock functions for showing the strengths of the prediction-based PSO variants regarding convergence.

Automatic tool detection from surgical imagery has a multitude of useful applications, such as real-time computer assistance for the surgeon. Using the successful residual network architecture, a system that can distinguish 21 different tools in cataract surgery videos is created. The videos are provided as part of the 2017 CATARACTS challenge and pose difficulties found in many real-world datasets, for example a strong class imbalance. The construction of the detection system is guided by a wide array of experiments that explore different design decisions.

Although neural networks are very successful in the domain of image processing, they are vulnerable to adversarial images-slightly perturbed images, which a human cannot distinguish from the original image. However, for the neural network, the perturbation leads to a different classification of the image. A lot of research was done on adversarial attacks, and on defenses against those attacks. In this paper, we propose a new defense by applying adversarial attacks to adversarial images. The new type of adversarial images is called adv-1and by observing the properties of the different transitions-from original to adversarial images, and from adversarial to adv-1images-we are able to detect adversarial images with a high accuracy, even for unknown attacks. Furthermore we are able to identify the attack, used to create the adversarial image in the first place. Regarding classification, depending on the used attack, our approach reaches correct classification accuracies, comparable to other defenses.

In this work, we propose a new dimensionality reduction approach for generating low-dimensional embeddings of high-dimensional data based on an iterative procedure. The data set's dimensions are sorted depending on their variance. Starting with the highest variance, the dimensions are iteratively projected onto the embedding. The projection can be seen as taking a photo from a two-dimensional motive employing a depth effect. The approach is flexible and offers numerous extensions for future work. We introduce a basic variant and illustrate it working mechanisms with numerous visualizations. The approach is experimentally analyzed on a small set of benchmark problems. Exemplary embeddings and evaluations based on the Shepard-Kruskal measure and the co-ranking matrix complement the analysis. The new approach shows competitive results in comparison to well-established dimensionality reduction methods.

Virtual sensors are getting more and more important as replacement and quality control tool for expensive and fragile hardware sensors. We introduce a virtual sensor application with marine sensor data from two data sources. The virtual sensor models are built upon recurrent neural networks (RNNs). To take full advantage of past data, we employ the time dimensionality reduction method piecewise approximate aggregation (PAA). We present an extension of this method, called exponential PAA (ExPAA) that pulls finer details from recent values, but preserves less exact information about the past. Experimental results demonstrate that RNNs benefit from this extension and confirm the stability and usability of our virtual sensor models over a five-month period of multivariate marine time series data.

Geographical constraints are essential when planning the locations for wind turbines. In real-world scenarios, especially in densely populated countries, the designated area where turbines can be placed is not an empty map on which the turbines can be placed arbitrarily. Even in rural areas, streets, buildings, and rivers have to be considered. In this paper, we model two constrained turbine placement scenarios and use evolutionary algorithms to find optimized turbine locations. To evaluate the locations, we combine a proven wind model with real-world data of a wind prediction model from a meteorological service. Geographical data from a free map service is used to define constrained areas in the scenarios based on administrative rules. For the evolutionary optimization process, we consider five ways to handle penalties. Starting with a simple specification that can only achieve two different values, we end up in a definition that considers distances relative of the required minimum distances to all geographical objects for each turbine. We combine the penalty definitions with three types of penalty functions. In the experimental section, we compare the various configurations and show a detailed analysis of the results.

Beside the curse of dimensionality and imbalanced classes, unfavorable data distributions can hamper classification accuracy. This is particularly problematic with increasing dimensionality of the classification task. A classifier that can handle high-dimensional and imbalanced data sets is the cascade classification method for time series. The cascade classifier can compound unfavorable data distributions by projecting the high-dimensional data set onto low-dimensional subsets. A classifier is trained for each of the low-dimensional data subsets and their predictions are aggregated to an overall result. For the cascade classifier, the errors of each classifier accumulate in the overall result and therefore small improvements in each small classifier can improve the classification accuracy. Therefore we propose two methods for data preprocessing to improve the cascade classifier. The first method is instance selection, a technique to select representative examples for the classification task. Furthermore, artificial infeasible examples can improve classification performance. Even if high-dimensional infeasible examples are available, their projection to low-dimensional space is not possible due to projection errors. We propose a second data preprocessing method for generating artificial infeasible examples in low-dimensional space. We show for micro Combined Heat and Power plant power production time series and an artificial and complex data set that the proposed data preprocessing methods increase the performance of the cascade classifier by increasing the selectivity of the learned decision boundaries.

Classification of high-dimensional data with imbalanced classes poses problems. Especially such time series classification tasks are problematic, because the ordering of each time step (feature) is important and therefore dimensionality reduction and feature selection cannot be applied. The cascade classification model was developed for such time series classification tasks. The cascade classifier splits high-dimensional classification tasks into a cascade of low-dimensional tasks. But the cascade classification model can only handle data sets with a data structure that can be easily learned in low-dimensional space. In this paper, we propose a generalized version of the cascade classification model that can also deal with data sets with more complex data structures. Generalization is achieved with time series transformations and an ensemble of classifiers based on the time series classifier: transformation based ensembles. For this purpose the cascade classifier is integrated into transformation based ensembles with some adjustments. In a simulation study we apply the generalized cascade classification model to predict the realizability (feasibility) of power production time series for pools of different numbers of micro combined heat and power plants. We show that the choice of the aggregation scheme for the ensemble members in the generalized cascade classification model has a strong impact on the overall classification results. But the choice of a weighting scheme showed hardly any influences on the classification result. Furthermore, data sets of different complexity (different structures in data space) yielded very similar classification results.

The imputation of partially missing multivariate time series data is critical for its correct analysis. The biggest problems in time series data are consecutively missing values that would result in serious information loss if simply dropped from the dataset. To address this problem, we adapt the k-Nearest Neighbors algorithm in a novel way for multivariate time series imputation. The algorithm employs Dynamic Time Warping as distance metric instead of point-wise distance measurements. We preprocess the data with linear interpolation to create complete windows for Dynamic Time Warping. The algorithm derives global distance weights from the correlation between features and consecutively missing values are penalized by individual distance weights to reduce error transfer from linear interpolation. Finally, efficient ensemble methods improve the accuracy. Experimental results show accurate imputations on datasets with a high correlation between features. Further, our algorithm shows better results with consecutively missing values than state-of-the-art algorithms.

Wind energy is an important component in the renewable energy mix, but successful integration into existing grid infrastructure is a major challenge. In this context, the accurate prediction of future wind generation power is extremely valuable as it would facilitate more efficient and sustainable provision. In a previous work, we proposed a method for wind power prediction based on an ensemble of support vector regressors. While this approach was very promising, there are still many avenues for further improvement. In this paper, we present two key extensions to the existing methodology and show that these result in significant performance improvements. In the first approach, we seek to vary the parameter values for the component support vector regressors. This is done in two ways: random initialization, and using an evolutionary strategy to select appropriate parameter values. Secondly, we exploit correlations between the input features to reduce the dimensionality and noise. Again, two approaches are tested: a feature selection stage using correlation, and Principal Component Analysis (PCA), which is shown to greatly reduce computational requirements while increasing prediction accuracy. The combination of these two enhancements produces interesting improvements over the previous prediction system, and these are presented and discussed in this paper.

Evolution strategies are successful blackbox optimization algorithms for continuous solution spaces. The covariance matrix adaptation evolution strategy (CMA-ES) and variants have shown great success on various problems in the past. In this paper, we present an evolution strategy (ES) based on a (1+1)-ES with Rechenberg's 1/5th step size control and Ledoit-Wolf covariance estimation. We compare this algorithm with a variant based on empirical maximum likelihood estimation. In the experimental part, the methods are compared to each other on a short benchmark function set. The ES with Ledoit-Wolf estimation turns out to outperform empirical covariance estimation. The analysis of the covariance estimation population size and the influence of the problem dimensionality allows insights into the choice of parameters.

The visualization of evolutionary blackbox optimization runs is important to understand evolutionary processes that may require the interaction with or intervention by the practitioner. But high-dimensional processes are not easy to visualize. In this work, we introduce an approach based on isometric mapping (ISOMAP) that maps continuous evolutionary runs in high-dimensional decision spaces to low-dimensional latent spaces that can be visualized. The embeddings are post-processed by computing a convex hull of embeddings, interpolating contour plots, and finally tracking the evolutionary run by marking the best solutions of each generation. Example plots demonstrate the capabilities of the approach. Experiments with the co-ranking matrix measure show that ISOMAP performs equally or better locally linear embedding and principal component analysis in maintaining neighborhoods of high-dimensional solutions.

Optimizing the weighting of features significantly improves the predictions in regression tasks. In this paper, we employ evolution strategies to evolve distance measures in a spatio-temporal regression approach for short-term wind prediction. The well-understood nearest neighbor regression method is the basis of our study. We compare a classic feature selection approach based on binary representations to the evolvement of continuous feature weights with the CMA-ES. The latter scales the original feature space and turns out to be the most successful approach in an experimental analysis on five benchmark turbines. We compare to standard nearest neighbor regression and concentrate on the interplay of training, validation, and test sets with a focus on overfitting the prediction model.

Wind turbine placement, i.e., the geographical planning of wind turbine locations, is an important first step to an efficient integration of wind energy. The turbine placement problem becomes a difficult optimization problem due to varying wind distributions at different locations and due to the mutual interference in the wind field known as wake effect. Artificial and environmental geological constraints make the optimization problem even more difficult to solve. In our paper, we focus on the evolutionary turbine placement based on an enhanced wake effect model fed with real-world wind distributions. We model geo-constraints with real-world data from OpenStreetMap. Besides the realistic modeling of wakes and geo-constraints, the focus of the paper is on the comparison of various evolutionary optimization approaches. We propose four variants of evolution strategies with turbine-oriented mutation operators and compare to state-of-the-art optimizers like the CMA-ES in a detailed experimental analysis on three benchmark scenarios.

Dimensionality reduction methods compute a mapping from a high-dimensional space to a space with lower dimensions while preserving important information. The idea of hybridizing dimensionality reduction with evolution strategies is that the search in a space that employs a larger dimensionality than the original solution space may be easier. We propose a dimensionality reduction evolution strategy (DRES) based on a self-adaptive (μ, λ)-ES that generates points in a space with a dimensionality higher than the original solution space. After the population has been generated, it is mapped to the solution space with dimensionality reduction (DR) methods, the solutions are evaluated and the best w.r.t. the fitness in the original space are inherited to the next generation. We employ principal component analysis (PCA) as DR method and show a performance tweak on a small set of benchmark problems.

Constraints can make a hard optimization problem even harder. We consider the blackbox scenario of unknown fitness and constraint functions. Evolution strategies with their self-adaptive step size control fail on simple problems like the sphere with one linear constraint (tangent problem). In this paper, we introduce an adaptive penalty function oriented to Rechenberg's 1/5th success rule: if less than 1/5th of the candidate population is feasible, the penalty is increased, otherwise, it is decreased. Experimental analyses on the tangent problem demonstrate that this simple strategy leads to very successful results for the high-dimensional constrained sphere function. We accelerate the approach with two regression meta-models, one for the constraint and one for the fitness function.

Regularized least-squares classification is one of the most promising alternatives to standard support vector machines, with the desirable property of closed-form solutions that can be obtained analytically, and efficiently. While the supervised, and mostly binary case has received tremendous attention in recent years, unsupervised multi-class settings have not yet been considered. In this work we present an efficient implementation for the unsupervised extension of the multi-class regularized least-squares classification framework, which is, to the best of the authors' knowledge, the first one in the literature addressing this task. The resulting kernel-based framework efficiently combines steepest descent strategies with powerful meta-heuristics for avoiding local minima. The computational efficiency of the overall approach is ensured through the application of matrix algebra shortcuts that render efficient updates of the intermediate candidate solutions possible. Our experimental evaluation indicates the potential of the novel method, and demonstrates its superior clustering performance over a variety of competing methods on real-world data sets.

In many optimization problems in practice, multiple objectives have to be optimized at the same time. Some multi-objective problems are characterized by multiple connected Pareto-sets at different parts in decision space — also called equivalent Pareto-subsets. We assume that the practitioner wants to approximate all Pareto-subsets to be able to choose among various solutions with different characteristics. In this work, we propose a clustering-based niching framework for multi-objective population-based approaches that allows to approximate equivalent Pareto-subsets. Iteratively, the clustering process assigns the population to niches, and the multi-objective optimization process concentrates on each niche independently. Two exemplary hybridizations, rake selection and DBSCAN, as well as SMS-EMOA and kernel density clustering demonstrate that the niching framework allows enough diversity to detect and approximate equivalent Pareto-subsets.

In systems optimization and machine learning multiple alternative solutions may exist in different parts of decision space for the same parts of the Pareto-front. The detection of equivalent Pareto-subsets may be desirable. In this paper we introduce a niching method that approximates Pareto-optimal solutions with diversity mechanisms in objective and decision space. For diversity in objective space we use rake selection, a selection method based on the distances to reference lines in objective space. For diversity in decision space we introduce a niching approach that uses the density based clustering method DBSCAN. The clustering process assigns the population to niches while the multi-objective optimization process concentrates on each niche independently. We introduce an indicator for the adaptive control of clustering processes, and extend rake selection by the concept of adaptive corner points. The niching method is experimentally validated on parameterized test function with the help of the S-metric.

We present a classification-based approach to identify quasi-stellar radio sources (quasars) in the Sloan Digital Sky Survey and evaluate its performance on a manually labeled training set. While reasonable results can already be obtained via approaches working only on photometric data, our experiments indicate that simple but problem-specific features extracted from spectroscopic data can significantly improve the classification performance. Since our approach works orthogonal to existing classification schemes used for building the spectroscopic catalogs, our classification results are well suited for a mutual assessment of the approaches' accuracies.

Kernel based techniques have shown outstanding success in data mining and machine learning in the recent past. Many optimization problems of kernel based methods suffer from multiple local optima. Evolution strategies have grown to successfulmethods in non-convex optimization. This work shows how both areas can profit from each other. We investigate the application of evolution strategies to Nadaraya-Watson based kernel regression and vice versa. The Nadaraya-Watson estimator is used as meta-model during optimization with the covariance matrix self-adaptation evolution strategy. An experimental analysis evaluates the meta-model assisted optimization process on a set of test functions and investigates model sizes and the balance between objective function evaluations on the real function and on the surrogate. In turn, evolution strategies can be used to optimize the embedded optimization problem of unsupervised kernel regression. The latter is fairly parameter dependent, and minimization of the data space reconstruction error is an optimization problem with numerous local optima. We propose an evolution strategy based unsupervised kernel regression method to solve the embedded learning problem. Furthermore, we tune the novel method by means of the parameter tuning technique sequential parameter optimization.

In the recent past, hybrid metaheuristics became famous as successful optimization methods. The motivation for the hybridization is a notion of combining the best of two worlds: evolutionary black box optimization and local search. Successful hybridizations in large combinatorial solution spaces motivate to transfer the idea of combining the two worlds to continuous domains as well. The question arises: Can local search also improve the convergence to the Pareto front in continuous multiobjective solutions spaces? We introduce a relay and a concurrent hybridization of the successful multiobjective optimizer SMS-EMOA and local optimization methods like Hooke & Jeeves and the Newton method. The concurrent approach is based on a parameterized probability function to control the local search. Experimental analyses on academic test functions show increased convergence speed as well as improved accuracy of the solution set of the new hybridizations.

The motif discovery problem consists of finding over-represented patterns in a collection of sequences. Its difficulty stems partly from the large number of possibilities to define both the motif space to be searched and the notion of over-representation. Since the size of the search space is generally exponential in the motif length, many heuristic methods, including evolutionary algorithms, have been developed. However, comparatively little attention has been devoted to the adequate evaluation of motif quality, especially when comparing motifs of different lengths. We propose an evolution strategy to solve the motif discovery problem based on a new fitness function that simultaneously takes into account (1) the number of motif occurrences, (2) the motif length, and (3) its information content. Experimental results show that the proposed method succeeds in uncovering the correct motif positions and length with high accuracy.

The maximum margin clustering approach is a recently proposed extension of the concept of support vector machines to the clustering problem. Briefly stated, it aims at finding an optimal partition of the data into two classes such that the margin induced by a subsequent application of a support vector machine is maximal. We propose a method based on stochastic search to address this hard optimization problem. While a direct implementation would be infeasible for large data sets, we present an efficient computational shortcut for assessing the "quality" of intermediate solutions. Experimental results show that our approach outperforms existing methods in terms of clustering accuracy.

Constrained continuous optimization is still an interesting field of research. Many heuristics have been proposed in the last decade. Most of them are based on penalty functions. Here, we experimentally investigate the two constraint handling heuristics proposed by Kramer and Schwefel. The two sexes evolution strategy (TSES) is inspired by the biological concept of sexual selection and pairing. The death penalty step control evolution strategy (DSES) is based on the controlled reduction of a minimum step size depending on the distance to the infeasible search space. These two methods are able to overcome the problem of premature mutation strength reduction, a result of the self-adaptation mechanism of evolution strategies in constrained environments. All methods are experimentally evaluated on a couple of typical constrained test problems. These experiments offer recommendations for the TSES population ratios and the speed of the ε-reduction process of the DSES.

The success of evolutionary algorithms (EAs) depends crucially on finding suitable parameter settings. Doing this by hand is a very time consuming job without the guarantee to finally find satisfactory parameters. Of course, there exist various kinds of parameter control techniques, but not for parameter tuning. The Design of Experiment (DoE) paradigm offers a way of retrieving optimal parameter settings. It is still a tedious task, but it is known to be a robust and well tested suite, which can be beneficial for giving reason to parameter choices besides human experience. In this paper we analyse evolution strategies (ES) and particle swarm optimisation (PSO) with and without optimal parameters gathered with DoE. Reasonable improvements have been observed for the two ES variants.

Self-adaptive crossover is a step towards exploiting the structure of problems automatically by evolution. We present a self-adaptive extension of the partially mapped crossover (PMX) operator that controls the crossover points. Because the link between strategy parameters and fitness is weak for self-adaptive crossover, superior results were hard to gather in the past. We can now report encouraging experimental results for the PMX on the traveling salesman problem (TSP) as an example for combinatorial problems.

Modern computer games have become far more sophisticated than their ancestors. In this process the requirements to the intelligence of artificial gaming characters have become more and more complex. This paper describes an approach to evolve human-competitive artificial players for modern computer games. The agents are evolved from scratch and successfully learn how to survive and defend themselves in the game. Agents trained with standard evolution against a training partner and agents trained by coevolution are presented. Both types of agents were able to defeat or even to dominate the original agents supplied by the game. Furthermore, we have made a detailed analysis of the obtained results to gain more insight into the resulting agents.

We propose a new mutation operator - the biased mutation operator (BMO) -for evolution strategies, which is capable of handling problems for constrained fitness landscapes. The idea of our approach is to bias the mutation ellipsoid in relation to the parent and therefore lead the mutations into a beneficial direction self-adaptively. This helps to improve the success rate to reproduce better offspring. Experimental results show this bias enhances the solution quality within constrained search domains. The number of the additional strategy parameters used in our approach equals to the number of dimensions of the problem. Compared to the correlated mutation, the BMO needs much less memory and supersedes the computation of the rotation matrix of the correlated mutation and the asymmetric probability density function of the directed mutation.

No abstract available.