In this paper we investigate the properties of CEAs with populations structured as Watts-Strogatz... more In this paper we investigate the properties of CEAs with populations structured as Watts-Strogatz small-world graphs and Albert-Barabási scale-free graphs as problem solvers, using several standard discrete optimization problems as a benchmark. The EA variants employed include self-adaptation of mutation rates. Results are compared with the corresponding classical panmictic EA showing that topology together with self-adaptation drastically influences the search.
Obviously, it is not a good idea to apply an optimization algorithm with wrongly specified parame... more Obviously, it is not a good idea to apply an optimization algorithm with wrongly specified parameter settings, a situation which can be avoided by applying algorithm tuning. Sequential tuning procedures are considered more efficient than single-stage procedures.
A radius-based separation of selection and recombination spheres in diffusion model EAs is introd... more A radius-based separation of selection and recombination spheres in diffusion model EAs is introduced, enabling a new taxonomy, oriented towards information flow analysis. It also contains parallel hillclimbers, panmictic EA and an unexplored area. Experiments are performed systematically on five complex binary and real coded problems in search of the best performing variants w.r.t. available optimization time. Additionally, information flow through recombination and selection is emulated by means of a simple model, that produces qualitative similar results.
We provide a comprehensive, effective and very efficient methodology for the design and experimen... more We provide a comprehensive, effective and very efficient methodology for the design and experimental analysis of algorithms. We rely on modern statistical techniques for tuning and understanding algorithms from an experimental perspective. Therefore, we make use of the sequential parameter optimization (SPO) method that has been successfully applied as a tuning procedure to numerous heuristics for practical and theoretical optimization problems. Two case studies, which illustrate the applicability of SPO to algorithm tuning and model selection, are presented.
Elevator supervisory group control (ESGC) is a complex combinatorial optimization task that can b... more Elevator supervisory group control (ESGC) is a complex combinatorial optimization task that can be solved by modern search heuristics. To reduce its complexity and to enable a theoretical analysis, a simplified ESGC model (S-ring) is proposed. The S-ring has many desirable properties: Fast evaluation, reproducibility, scalability, and extensibility. It can be described as a Markov decision process and thus be analyzed theoretically and numerically. Algorithm based validation (ABV), as a new methodology for the validation of simulation models, is introduced. Based on ABV, we show that the S-ring is a valid ESGC model. Finally, the extensibility of the S-ring model is demonstrated.
Adjusting algorithm parameters to a given problem is of crucial importance for performance compar... more Adjusting algorithm parameters to a given problem is of crucial importance for performance comparisons as well as for reliable (first) results on previously unknown problems, or with new algorithms. This also holds for parameters controlling adaptability features, as long as the optimization algorithm is not able to completely self-adapt itself to the posed problem and thereby get rid of all parameters. We present the recently developed sequential parameter optimization (SPO) technique that reliably finds good parameter sets for stochastically disturbed algorithm output. SPO combines classical regression techniques and modern statistical approaches for deterministic algorithms as Design and Analysis of Computer Experiments (DACE). Moreover, it is embedded in a twelve-step procedure that targets at doing optimization experiments in a statistically sound manner, focusing on answering scientific questions.
This paper investigates the intelligent moving and path-finding of groups in real-time strategy (... more This paper investigates the intelligent moving and path-finding of groups in real-time strategy (RTS) games exemplified by the open source game Glest. We utilize the technique of Flocking for achieving a smooth and natural movement of a group of units and expect grouping to decrease the amount of unit losses in RTS games. Furthermore, we present a setting in which Flocking will improve the game progress. But we also demonstrate a situation where Flocking fails. To prevent these annoying situations, we combined Flocking with Influence Maps (IM) to find safe paths for the flock in real time. This combination turns out to be an excellent alternative to normal movement in Glest and most likely in other RTS games.
This paper presents a search-based method for generating maps for the popular real-time strategy ... more This paper presents a search-based method for generating maps for the popular real-time strategy (RTS) game StarCraft. We devise a representation of StarCraft maps suitable for evolutionary search, along with a set of fitness functions based on predicted entertainment value of those maps, as derived from theories of player experience. A multiobjective evolutionary algorithm is then used to evolve complete Star-Craft maps based on the representation and selected fitness functions. The output of this algorithm is a Pareto front approximation visualizing the tradeoff between the several fitness functions used, and where each point on the front represents a viable map. We argue that this method is useful for both automatic and machine-assisted map generation, and in particular that the Pareto fronts are excellent design support tools for human map designers.
IEEE Transactions on Computational Intelligence and Ai in Games, 2010
In this paper, we overview the 2009 Simulated Car Racing Championship-an event comprising three c... more In this paper, we overview the 2009 Simulated Car Racing Championship-an event comprising three competitions held in association with the 2009 IEEE Congress on Evolutionary Computation (CEC), the 2009 ACM Genetic and Evolutionary Computation Conference (GECCO), and the 2009 IEEE Symposium on Computational Intelligence and Games (CIG). First, we describe the competition regulations and the software framework. Then, the five best teams describe the methods of computational intelligence they used to develop their drivers and the lessons they learned from the participation in the championship. The organizers provide short summaries of the other competitors. Finally, we summarize the championship results, followed by a discussion about what the organizers learned about 1) the development of high-performing car racing controllers and 2) the organization of scientific competitions.
Popular games often have a high-quality graphic design but quite simple-minded non player charact... more Popular games often have a high-quality graphic design but quite simple-minded non player characters (NPC). Recently, Computational Intelligence (CI) methods have been discovered as suitable methods to revive NPC, making games more interesting, challenging, and funny. We present a fairly large study of human players on the simple arcade game Pac-Man, controlling the ghosts behaviors by simple strategies, neural networks or evolutionary algorithms. The playerpsilas fun is of course a subjective experience, but we presume that it is related to the psychological flow concept. We deal with the question whether flow is a more reliable measure than asking human players directly for the fun experienced during the game. In order to detect flow, we introduce a measure based on the interaction time fraction between the human-controlled Pac-Man and the ghosts, and compare the outcome to the results of a fun measure suggested by Yannakakis and Hallam [1].
In multi-criterion optimization, Pareto-optimal solutions that appear very similar in the objecti... more In multi-criterion optimization, Pareto-optimal solutions that appear very similar in the objective space may have very different pre-images. In many practical applications the decision makers, who select a solution or preferred region on the Pareto-front, may want to know different pre-images of the selected solutions. Especially, this will be the case when they would like to present alternative design candidates in later stages of a multidisciplinary design process. In this paper we extend an existing CMA-ES niching framework, which has been previously applied successfully to multi-modal optimization, to the multi-criterion domain for boosting decision space diversity. At the same time, we introduce the concept of space aggregation for diversity maintenance in the aggregated spaces, i.e. search/decision and objective space. Empirical results on synthetic multi-modal bi-criteria test problems with known efficient sets and Pareto-fronts demonstrate that the diversity in the decision space can be significantly enhanced without hampering the convergence to a precise and diverse Pareto front approximation in the objective space of the original algorithm.
Recent works in evolutionary multiobjective optimization suggest to shift the focus from solely e... more Recent works in evolutionary multiobjective optimization suggest to shift the focus from solely evaluating optimization success in the objective space to also taking the decision space into account. They indicate that this may be a) necessary to express the users requirements of obtaining distinct solutions (distinct Pareto set parts or subsets) of similar quality (comparable locations on the Pareto front) in real-world applications, and b) a demanding task for the currently most commonly used algorithms. We investigate if standard EMOA are able to detect and preserve equivalent Pareto subsets and develop an own special purpose EMOA that meets these requirements reliably.
We propose a robust approach for learning car racing track models from sensory data for the car r... more We propose a robust approach for learning car racing track models from sensory data for the car racing simulator TORCS. Our track recognition system is based on the combination of an advanced preprocessing step of the sensory data and a simple classifier that delivers six types of track shapes similar to the ones a human would recognize. Out of these, establishing a complete track model is straightforward. This model provides an information advantage to controller strategies, as it generally enables planning. We demonstrate how such a planning controller can be derived by a mixture of expert knowledge and a simple evolutionary learning approach and give experimental evidence that knowing not only the current conditions but also the big picture of the track is beneficial, as may be expected.
For solving multimodal problems by means of evolutionary algorithms, one often resorts to multist... more For solving multimodal problems by means of evolutionary algorithms, one often resorts to multistarts or niching methods. The latter approach the question: 'What is elsewhere?' by an implicit second criterion in order to keep populations distributed over the search space. Induced by a practical problem that appears to be simple but is not easily solved, a multiobjective algorithm is proposed for solving multimodal problems. It employs an explicit diversity criterion as second objective. Experimental comparison with standard methods suggests that the multiobjective algorithm is fast and reliable and that coupling it with a local search technique is straightforward and leads to enormous quality gain. The combined algorithm is still fast and may be especially valuable for practical problems with costly target function evaluations.
Recent research on evolutionary multiobjective optimization has mainly focused on Pareto fronts. ... more Recent research on evolutionary multiobjective optimization has mainly focused on Pareto fronts. However, we state that proper behavior of the utilized algorithms in decision/search space is necessary for obtaining good results if multimodal objective functions are concerned. Therefore, it makes sense to observe the development of Pareto sets as well. We do so on a simple, configurable problem, and detect interesting interactions between induced changes to the Pareto set and the ability of three optimization algorithms to keep track of Pareto fronts.
The present paper investigates the hybridization of two wellknown multimodal optimization methods... more The present paper investigates the hybridization of two wellknown multimodal optimization methods, i.e. species conservation and multinational algorithms. The topological species conservation algorithm embraces the vision of the existence of subpopulations around seeds (the best local individuals) and the preservation of these dominating individuals from one generation to another, but detects multimodality by means of the hill-valley mechanism employed by multinational algorithms. The aim is to inherit the strengths of both parent techniques and at the same time overcome their flaws. The species conservation algorithm efficiently keeps track of several good search space regions at once, but is difficult to parametrize without prior problem knowledge. Conversely, the multinational algorithms use many function evaluations to establish subpopulations, but do not depend on provided radius parameter values. Experiments with all three algorithms are made on a wide range of test problems in order to investigate their advantages and shortcomings.
A new radii-based evolutionary algorithm (EA) designed for multimodal optimization problems is pr... more A new radii-based evolutionary algorithm (EA) designed for multimodal optimization problems is proposed. The approach can be placed within the genetic chromodynamics framework and related to other EAs with local interaction, e.g. using species formation or clearing procedures. The underlying motivation for modifying the original algorithm was to preserve its ability to search for many optima in parallel while increasing convergence speed, especially for complex problems, through generational selection and different replacement schemes. The algorithm is applied to function optimization and classification; obtained experimental results, in part improved immensely by state-ofthe-art parameter tuning (SPO), encourage further investigation.
The aim of this paper is to validate the new paradigm of evolutionary support vector machines (ES... more The aim of this paper is to validate the new paradigm of evolutionary support vector machines (ESVMs) for binary classification also through an application to a real-world problem, i.e. the diagnosis of diabetes mellitus. ESVMs were developed through hybridization between the strong learning paradigm of support vector machines (SVMs) and the optimization power of evolutionary computation. Hybridization is achieved at the level of solving the constrained optimization problem within the SVMs, which is a difficult task to perform in its standard manner. ESVMs have been so far applied to the binary classification of two-dimensional points. In this paper, experiments are conducted on the benchmark problem concerning diabetes of the UCI repository of machine learning data sets. Obtained results proved the correctness and promise of the new hybridized learning technique and demonstrated its ability to solve any case of binary standard classification
In this paper we investigate the properties of CEAs with populations structured as Watts-Strogatz... more In this paper we investigate the properties of CEAs with populations structured as Watts-Strogatz small-world graphs and Albert-Barabási scale-free graphs as problem solvers, using several standard discrete optimization problems as a benchmark. The EA variants employed include self-adaptation of mutation rates. Results are compared with the corresponding classical panmictic EA showing that topology together with self-adaptation drastically influences the search.
Obviously, it is not a good idea to apply an optimization algorithm with wrongly specified parame... more Obviously, it is not a good idea to apply an optimization algorithm with wrongly specified parameter settings, a situation which can be avoided by applying algorithm tuning. Sequential tuning procedures are considered more efficient than single-stage procedures.
A radius-based separation of selection and recombination spheres in diffusion model EAs is introd... more A radius-based separation of selection and recombination spheres in diffusion model EAs is introduced, enabling a new taxonomy, oriented towards information flow analysis. It also contains parallel hillclimbers, panmictic EA and an unexplored area. Experiments are performed systematically on five complex binary and real coded problems in search of the best performing variants w.r.t. available optimization time. Additionally, information flow through recombination and selection is emulated by means of a simple model, that produces qualitative similar results.
We provide a comprehensive, effective and very efficient methodology for the design and experimen... more We provide a comprehensive, effective and very efficient methodology for the design and experimental analysis of algorithms. We rely on modern statistical techniques for tuning and understanding algorithms from an experimental perspective. Therefore, we make use of the sequential parameter optimization (SPO) method that has been successfully applied as a tuning procedure to numerous heuristics for practical and theoretical optimization problems. Two case studies, which illustrate the applicability of SPO to algorithm tuning and model selection, are presented.
Elevator supervisory group control (ESGC) is a complex combinatorial optimization task that can b... more Elevator supervisory group control (ESGC) is a complex combinatorial optimization task that can be solved by modern search heuristics. To reduce its complexity and to enable a theoretical analysis, a simplified ESGC model (S-ring) is proposed. The S-ring has many desirable properties: Fast evaluation, reproducibility, scalability, and extensibility. It can be described as a Markov decision process and thus be analyzed theoretically and numerically. Algorithm based validation (ABV), as a new methodology for the validation of simulation models, is introduced. Based on ABV, we show that the S-ring is a valid ESGC model. Finally, the extensibility of the S-ring model is demonstrated.
Adjusting algorithm parameters to a given problem is of crucial importance for performance compar... more Adjusting algorithm parameters to a given problem is of crucial importance for performance comparisons as well as for reliable (first) results on previously unknown problems, or with new algorithms. This also holds for parameters controlling adaptability features, as long as the optimization algorithm is not able to completely self-adapt itself to the posed problem and thereby get rid of all parameters. We present the recently developed sequential parameter optimization (SPO) technique that reliably finds good parameter sets for stochastically disturbed algorithm output. SPO combines classical regression techniques and modern statistical approaches for deterministic algorithms as Design and Analysis of Computer Experiments (DACE). Moreover, it is embedded in a twelve-step procedure that targets at doing optimization experiments in a statistically sound manner, focusing on answering scientific questions.
This paper investigates the intelligent moving and path-finding of groups in real-time strategy (... more This paper investigates the intelligent moving and path-finding of groups in real-time strategy (RTS) games exemplified by the open source game Glest. We utilize the technique of Flocking for achieving a smooth and natural movement of a group of units and expect grouping to decrease the amount of unit losses in RTS games. Furthermore, we present a setting in which Flocking will improve the game progress. But we also demonstrate a situation where Flocking fails. To prevent these annoying situations, we combined Flocking with Influence Maps (IM) to find safe paths for the flock in real time. This combination turns out to be an excellent alternative to normal movement in Glest and most likely in other RTS games.
This paper presents a search-based method for generating maps for the popular real-time strategy ... more This paper presents a search-based method for generating maps for the popular real-time strategy (RTS) game StarCraft. We devise a representation of StarCraft maps suitable for evolutionary search, along with a set of fitness functions based on predicted entertainment value of those maps, as derived from theories of player experience. A multiobjective evolutionary algorithm is then used to evolve complete Star-Craft maps based on the representation and selected fitness functions. The output of this algorithm is a Pareto front approximation visualizing the tradeoff between the several fitness functions used, and where each point on the front represents a viable map. We argue that this method is useful for both automatic and machine-assisted map generation, and in particular that the Pareto fronts are excellent design support tools for human map designers.
IEEE Transactions on Computational Intelligence and Ai in Games, 2010
In this paper, we overview the 2009 Simulated Car Racing Championship-an event comprising three c... more In this paper, we overview the 2009 Simulated Car Racing Championship-an event comprising three competitions held in association with the 2009 IEEE Congress on Evolutionary Computation (CEC), the 2009 ACM Genetic and Evolutionary Computation Conference (GECCO), and the 2009 IEEE Symposium on Computational Intelligence and Games (CIG). First, we describe the competition regulations and the software framework. Then, the five best teams describe the methods of computational intelligence they used to develop their drivers and the lessons they learned from the participation in the championship. The organizers provide short summaries of the other competitors. Finally, we summarize the championship results, followed by a discussion about what the organizers learned about 1) the development of high-performing car racing controllers and 2) the organization of scientific competitions.
Popular games often have a high-quality graphic design but quite simple-minded non player charact... more Popular games often have a high-quality graphic design but quite simple-minded non player characters (NPC). Recently, Computational Intelligence (CI) methods have been discovered as suitable methods to revive NPC, making games more interesting, challenging, and funny. We present a fairly large study of human players on the simple arcade game Pac-Man, controlling the ghosts behaviors by simple strategies, neural networks or evolutionary algorithms. The playerpsilas fun is of course a subjective experience, but we presume that it is related to the psychological flow concept. We deal with the question whether flow is a more reliable measure than asking human players directly for the fun experienced during the game. In order to detect flow, we introduce a measure based on the interaction time fraction between the human-controlled Pac-Man and the ghosts, and compare the outcome to the results of a fun measure suggested by Yannakakis and Hallam [1].
In multi-criterion optimization, Pareto-optimal solutions that appear very similar in the objecti... more In multi-criterion optimization, Pareto-optimal solutions that appear very similar in the objective space may have very different pre-images. In many practical applications the decision makers, who select a solution or preferred region on the Pareto-front, may want to know different pre-images of the selected solutions. Especially, this will be the case when they would like to present alternative design candidates in later stages of a multidisciplinary design process. In this paper we extend an existing CMA-ES niching framework, which has been previously applied successfully to multi-modal optimization, to the multi-criterion domain for boosting decision space diversity. At the same time, we introduce the concept of space aggregation for diversity maintenance in the aggregated spaces, i.e. search/decision and objective space. Empirical results on synthetic multi-modal bi-criteria test problems with known efficient sets and Pareto-fronts demonstrate that the diversity in the decision space can be significantly enhanced without hampering the convergence to a precise and diverse Pareto front approximation in the objective space of the original algorithm.
Recent works in evolutionary multiobjective optimization suggest to shift the focus from solely e... more Recent works in evolutionary multiobjective optimization suggest to shift the focus from solely evaluating optimization success in the objective space to also taking the decision space into account. They indicate that this may be a) necessary to express the users requirements of obtaining distinct solutions (distinct Pareto set parts or subsets) of similar quality (comparable locations on the Pareto front) in real-world applications, and b) a demanding task for the currently most commonly used algorithms. We investigate if standard EMOA are able to detect and preserve equivalent Pareto subsets and develop an own special purpose EMOA that meets these requirements reliably.
We propose a robust approach for learning car racing track models from sensory data for the car r... more We propose a robust approach for learning car racing track models from sensory data for the car racing simulator TORCS. Our track recognition system is based on the combination of an advanced preprocessing step of the sensory data and a simple classifier that delivers six types of track shapes similar to the ones a human would recognize. Out of these, establishing a complete track model is straightforward. This model provides an information advantage to controller strategies, as it generally enables planning. We demonstrate how such a planning controller can be derived by a mixture of expert knowledge and a simple evolutionary learning approach and give experimental evidence that knowing not only the current conditions but also the big picture of the track is beneficial, as may be expected.
For solving multimodal problems by means of evolutionary algorithms, one often resorts to multist... more For solving multimodal problems by means of evolutionary algorithms, one often resorts to multistarts or niching methods. The latter approach the question: 'What is elsewhere?' by an implicit second criterion in order to keep populations distributed over the search space. Induced by a practical problem that appears to be simple but is not easily solved, a multiobjective algorithm is proposed for solving multimodal problems. It employs an explicit diversity criterion as second objective. Experimental comparison with standard methods suggests that the multiobjective algorithm is fast and reliable and that coupling it with a local search technique is straightforward and leads to enormous quality gain. The combined algorithm is still fast and may be especially valuable for practical problems with costly target function evaluations.
Recent research on evolutionary multiobjective optimization has mainly focused on Pareto fronts. ... more Recent research on evolutionary multiobjective optimization has mainly focused on Pareto fronts. However, we state that proper behavior of the utilized algorithms in decision/search space is necessary for obtaining good results if multimodal objective functions are concerned. Therefore, it makes sense to observe the development of Pareto sets as well. We do so on a simple, configurable problem, and detect interesting interactions between induced changes to the Pareto set and the ability of three optimization algorithms to keep track of Pareto fronts.
The present paper investigates the hybridization of two wellknown multimodal optimization methods... more The present paper investigates the hybridization of two wellknown multimodal optimization methods, i.e. species conservation and multinational algorithms. The topological species conservation algorithm embraces the vision of the existence of subpopulations around seeds (the best local individuals) and the preservation of these dominating individuals from one generation to another, but detects multimodality by means of the hill-valley mechanism employed by multinational algorithms. The aim is to inherit the strengths of both parent techniques and at the same time overcome their flaws. The species conservation algorithm efficiently keeps track of several good search space regions at once, but is difficult to parametrize without prior problem knowledge. Conversely, the multinational algorithms use many function evaluations to establish subpopulations, but do not depend on provided radius parameter values. Experiments with all three algorithms are made on a wide range of test problems in order to investigate their advantages and shortcomings.
A new radii-based evolutionary algorithm (EA) designed for multimodal optimization problems is pr... more A new radii-based evolutionary algorithm (EA) designed for multimodal optimization problems is proposed. The approach can be placed within the genetic chromodynamics framework and related to other EAs with local interaction, e.g. using species formation or clearing procedures. The underlying motivation for modifying the original algorithm was to preserve its ability to search for many optima in parallel while increasing convergence speed, especially for complex problems, through generational selection and different replacement schemes. The algorithm is applied to function optimization and classification; obtained experimental results, in part improved immensely by state-ofthe-art parameter tuning (SPO), encourage further investigation.
The aim of this paper is to validate the new paradigm of evolutionary support vector machines (ES... more The aim of this paper is to validate the new paradigm of evolutionary support vector machines (ESVMs) for binary classification also through an application to a real-world problem, i.e. the diagnosis of diabetes mellitus. ESVMs were developed through hybridization between the strong learning paradigm of support vector machines (SVMs) and the optimization power of evolutionary computation. Hybridization is achieved at the level of solving the constrained optimization problem within the SVMs, which is a difficult task to perform in its standard manner. ESVMs have been so far applied to the binary classification of two-dimensional points. In this paper, experiments are conducted on the benchmark problem concerning diabetes of the UCI repository of machine learning data sets. Obtained results proved the correctness and promise of the new hybridized learning technique and demonstrated its ability to solve any case of binary standard classification
Uploads
Papers by Mike Preuss