M. Pelikán, K. Sastry, D. Goldberg, Martin Volker Butz, Mark Hauschild
This paper presents a class of NK landscapes with nearest-neighbor interactions and tunable overlap. The considered class of NK landscapes is solvable in polynomial time using dynamic programming; this allows us to generate a large number of random problem instances with known optima. Several genetic and evolutionary algorithms are then applied to the generated problem instances. The results are analyzed and related to scalability theory for genetic algorithms and estimation of distribution algorithms.
{"title":"Performance of evolutionary algorithms on NK landscapes with nearest neighbor interactions and tunable overlap","authors":"M. Pelikán, K. Sastry, D. Goldberg, Martin Volker Butz, Mark Hauschild","doi":"10.1145/1569901.1570018","DOIUrl":"https://doi.org/10.1145/1569901.1570018","url":null,"abstract":"This paper presents a class of NK landscapes with nearest-neighbor interactions and tunable overlap. The considered class of NK landscapes is solvable in polynomial time using dynamic programming; this allows us to generate a large number of random problem instances with known optima. Several genetic and evolutionary algorithms are then applied to the generated problem instances. The results are analyzed and related to scalability theory for genetic algorithms and estimation of distribution algorithms.","PeriodicalId":193093,"journal":{"name":"Proceedings of the 11th Annual conference on Genetic and evolutionary computation","volume":"23 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2009-07-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123807342","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
We propose a crossover operator that works with genetic programming trees and is approximately geometric crossover in the semantic space. By defining semantic as program's evaluation profile with respect to a set of fitness cases and constraining to a specific class of metric-based fitness functions, we cause the fitness landscape in the semantic space to have perfect fitness-distance correlation. The proposed approximately geometric semantic crossover exploits this property of the semantic fitness landscape by an appropriate sampling. We demonstrate also how the proposed method may be conveniently combined with hill climbing. We discuss the properties of the methods, and describe an extensive computational experiment concerning logical function synthesis and symbolic regression.
{"title":"Approximating geometric crossover in semantic space","authors":"K. Krawiec, Pawel Lichocki","doi":"10.1145/1569901.1570036","DOIUrl":"https://doi.org/10.1145/1569901.1570036","url":null,"abstract":"We propose a crossover operator that works with genetic programming trees and is approximately geometric crossover in the semantic space. By defining semantic as program's evaluation profile with respect to a set of fitness cases and constraining to a specific class of metric-based fitness functions, we cause the fitness landscape in the semantic space to have perfect fitness-distance correlation. The proposed approximately geometric semantic crossover exploits this property of the semantic fitness landscape by an appropriate sampling. We demonstrate also how the proposed method may be conveniently combined with hill climbing. We discuss the properties of the methods, and describe an extensive computational experiment concerning logical function synthesis and symbolic regression.","PeriodicalId":193093,"journal":{"name":"Proceedings of the 11th Annual conference on Genetic and evolutionary computation","volume":"24 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2009-07-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126868654","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Session details: Track 13: real world application","authors":"M. O’Neill","doi":"10.1145/3257507","DOIUrl":"https://doi.org/10.1145/3257507","url":null,"abstract":"","PeriodicalId":193093,"journal":{"name":"Proceedings of the 11th Annual conference on Genetic and evolutionary computation","volume":"2 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2009-07-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127959942","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
It is a hard problem to understand the search process of particle swarm optimization over high-dimensional domain. The visualization depicts the total search process and then it will allow better understanding of how to tune the algorithm. For the investigation, we adopt Sammon's mapping, which is a well-known distance-preserving mapping. We demonstrate the usefulness of the proposed methodology by applying it to some function optimization problems.
{"title":"Visualizing the search process of particle swarm optimization","authors":"Yong-Hyuk Kim, K. Lee, Yourim Yoon","doi":"10.1145/1569901.1569909","DOIUrl":"https://doi.org/10.1145/1569901.1569909","url":null,"abstract":"It is a hard problem to understand the search process of particle swarm optimization over high-dimensional domain. The visualization depicts the total search process and then it will allow better understanding of how to tune the algorithm. For the investigation, we adopt Sammon's mapping, which is a well-known distance-preserving mapping. We demonstrate the usefulness of the proposed methodology by applying it to some function optimization problems.","PeriodicalId":193093,"journal":{"name":"Proceedings of the 11th Annual conference on Genetic and evolutionary computation","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2009-07-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128788201","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In this paper, we attempt to make accurate predictions of the movement of the stock market with the aid of an evolutionary artificial neural network (EANN). To facilitate this objective we constructed an EANN for multi-objective optimization (MOO) that was trained with macro-economic data and its effect on market performance. Experiments were conducted with EANNs that updated connection weights through genetic operators (crossover and mutation) and/or with the aid of back-propagation. The results showed that the optimal performance was achieved under natural complexification of the EANN and that back-propagation tended to over fit the data. The results also suggested that EANNs trained with multi-objectives were more robust than that of a single optimization approach. The MOO approach produced superior investment returns during training and testing over a single objective optimization (SOO).
{"title":"Multi-objective optimization with an evolutionary artificial neural network for financial forecasting","authors":"Matthew Butler, Ali Daniyal","doi":"10.1145/1569901.1570096","DOIUrl":"https://doi.org/10.1145/1569901.1570096","url":null,"abstract":"In this paper, we attempt to make accurate predictions of the movement of the stock market with the aid of an evolutionary artificial neural network (EANN). To facilitate this objective we constructed an EANN for multi-objective optimization (MOO) that was trained with macro-economic data and its effect on market performance. Experiments were conducted with EANNs that updated connection weights through genetic operators (crossover and mutation) and/or with the aid of back-propagation. The results showed that the optimal performance was achieved under natural complexification of the EANN and that back-propagation tended to over fit the data. The results also suggested that EANNs trained with multi-objectives were more robust than that of a single optimization approach. The MOO approach produced superior investment returns during training and testing over a single objective optimization (SOO).","PeriodicalId":193093,"journal":{"name":"Proceedings of the 11th Annual conference on Genetic and evolutionary computation","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2009-07-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128945967","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Jella Pfeiffer, D. Duzevik, Franz Rothlauf, Koichi Yamamoto
In the field of decision-making a fundamental problem is how to uncover people's choice behavior. While choices them- selves are often observable, our underlying decision strategies determining these choices are not entirely understood. Previous research defined a number of decision strategies and conjectured that people do not apply only one strategy but switch strategies during the decision process. To the best of our knowledge, empirical evidence for the latter conjecture is missing. This is why we monitored the purchase decisions 624 consumers shopping online. We study how many of the observed choices can be explained by the existing strategies in their pure form, how many decisions can be explained if we account for switching behavior, and investigate switching behavior in detail. Since accounting for switching leads to a large search space of possible mixed decision strategies, we apply a genetic algorithm to find the set of mixed decision strategies which best explains the observed behavior. The results show that mixed strategies are used more often than pure ones and that a set of four mixed strategies is able to explain 93.9% of choices in a scenario with 4 alternatives and 75.4% of choices in a scenario with 7 alternatives.
{"title":"A genetic algorithm for analyzing choice behavior with mixed decision strategies","authors":"Jella Pfeiffer, D. Duzevik, Franz Rothlauf, Koichi Yamamoto","doi":"10.1145/1569901.1570113","DOIUrl":"https://doi.org/10.1145/1569901.1570113","url":null,"abstract":"In the field of decision-making a fundamental problem is how to uncover people's choice behavior. While choices them- selves are often observable, our underlying decision strategies determining these choices are not entirely understood. Previous research defined a number of decision strategies and conjectured that people do not apply only one strategy but switch strategies during the decision process. To the best of our knowledge, empirical evidence for the latter conjecture is missing. This is why we monitored the purchase decisions 624 consumers shopping online. We study how many of the observed choices can be explained by the existing strategies in their pure form, how many decisions can be explained if we account for switching behavior, and investigate switching behavior in detail. Since accounting for switching leads to a large search space of possible mixed decision strategies, we apply a genetic algorithm to find the set of mixed decision strategies which best explains the observed behavior. The results show that mixed strategies are used more often than pure ones and that a set of four mixed strategies is able to explain 93.9% of choices in a scenario with 4 alternatives and 75.4% of choices in a scenario with 7 alternatives.","PeriodicalId":193093,"journal":{"name":"Proceedings of the 11th Annual conference on Genetic and evolutionary computation","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2009-07-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130486444","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
This paper presents a coalition-based metaheuristic (CBM) to solve the uncapacitated facility location problem. CBM is a population-based metaheuristic where individuals encapsulate a single solution and are considered as agents. In comparison to classical evolutionary algorithms, these agents have additional capacities of decision, learning and cooperation. Our approach is also a case study to present how concepts from multiagent systems' domain may contribute to the design of new metaheuristics. The tackled problem is a well-known combinatorial optimization problem, namely the uncapacitated facility location problem, that consists in determining the sites in which some facilities must be set up to satisfy the requirements of a client set at minimum cost. A computational experiment is conducted to test the performance of learning mechanisms and to compare our approach with several existing metaheuristics. The results showed that CBM is competitive with powerful heuristics approaches and presents several advantages in terms of flexibility and modularity.
{"title":"A cooperative and self-adaptive metaheuristic for the facility location problem","authors":"D. Meignan, Jean-Charles Créput, A. Koukam","doi":"10.1145/1569901.1569946","DOIUrl":"https://doi.org/10.1145/1569901.1569946","url":null,"abstract":"This paper presents a coalition-based metaheuristic (CBM) to solve the uncapacitated facility location problem. CBM is a population-based metaheuristic where individuals encapsulate a single solution and are considered as agents. In comparison to classical evolutionary algorithms, these agents have additional capacities of decision, learning and cooperation. Our approach is also a case study to present how concepts from multiagent systems' domain may contribute to the design of new metaheuristics. The tackled problem is a well-known combinatorial optimization problem, namely the uncapacitated facility location problem, that consists in determining the sites in which some facilities must be set up to satisfy the requirements of a client set at minimum cost. A computational experiment is conducted to test the performance of learning mechanisms and to compare our approach with several existing metaheuristics. The results showed that CBM is competitive with powerful heuristics approaches and presents several advantages in terms of flexibility and modularity.","PeriodicalId":193093,"journal":{"name":"Proceedings of the 11th Annual conference on Genetic and evolutionary computation","volume":"9 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2009-07-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130492617","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The characterization of fractured reservoirs involves: (1) the design of geological models integrating statistical and/or deterministic fracture properties; (2) the validation of flow simulation models by calibrating with dynamic field data e.g. well tests. The latter validation step is critical since it also validates the underlying geological model, it allows one to reduce some uncertainties among the fracture geometrical and distribution properties, and it is often the only mean to characterize fracture conductivities. However this is usually an ill-posed inverse problem: field data are usually not sufficient to fully characterize the fracture system. It is of interest to explore the parameters space effectively, so that multiple solutions may be characterized, and many production development scenarii may be studied. This paper presents a well tests inversion method to characterize fracture sets conductivities. The Covariance Matrix Adaptation-Evolution Strategy (CMA-ES) has been used as the optimization algorithm. It has been tested with some local optimization algorithms for comparison, and extended in order to detect several solutions simultaneously using a local proxy of the response surface. Moreover, uncertainty analyses are performed in regions of interest. Applications are presented for a fracture system with two fracture sets.
{"title":"An extended evolution strategy for the characterization of fracture conductivities from well tests","authors":"J. Bruyelle, A. Lange","doi":"10.1145/1569901.1570095","DOIUrl":"https://doi.org/10.1145/1569901.1570095","url":null,"abstract":"The characterization of fractured reservoirs involves: (1) the design of geological models integrating statistical and/or deterministic fracture properties; (2) the validation of flow simulation models by calibrating with dynamic field data e.g. well tests. The latter validation step is critical since it also validates the underlying geological model, it allows one to reduce some uncertainties among the fracture geometrical and distribution properties, and it is often the only mean to characterize fracture conductivities. However this is usually an ill-posed inverse problem: field data are usually not sufficient to fully characterize the fracture system. It is of interest to explore the parameters space effectively, so that multiple solutions may be characterized, and many production development scenarii may be studied. This paper presents a well tests inversion method to characterize fracture sets conductivities. The Covariance Matrix Adaptation-Evolution Strategy (CMA-ES) has been used as the optimization algorithm. It has been tested with some local optimization algorithms for comparison, and extended in order to detect several solutions simultaneously using a local proxy of the response surface. Moreover, uncertainty analyses are performed in regions of interest. Applications are presented for a fracture system with two fracture sets.","PeriodicalId":193093,"journal":{"name":"Proceedings of the 11th Annual conference on Genetic and evolutionary computation","volume":"65 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2009-07-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132994158","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The covariance matrix adaptation evolution strategy (CMAES) has proven to be a powerful method for reinforcement learning (RL). Recently, the CMA-ES has been augmented with an adaptive uncertainty handling mechanism. Because uncertainty is a typical property of RL problems this new algorithm, termed UH-CMA-ES, is promising for RL. The UH-CMA-ES dynamically adjusts the number of episodes considered in each evaluation of a policy. It controls the signal to noise ratio such that it is just high enough for a sufficiently good ranking of candidate policies, which in turn allows the evolutionary learning to find better solutions. This significantly increases the learning speed as well as the robustness without impairing the quality of the final solutions. We evaluate the UH-CMA-ES on fully and partially observable Markov decision processes with random start states and noisy observations. A canonical natural policy gradient method and random search serve as a baseline for comparison.
{"title":"Uncertainty handling CMA-ES for reinforcement learning","authors":"V. Heidrich-Meisner, C. Igel","doi":"10.1145/1569901.1570064","DOIUrl":"https://doi.org/10.1145/1569901.1570064","url":null,"abstract":"The covariance matrix adaptation evolution strategy (CMAES) has proven to be a powerful method for reinforcement learning (RL). Recently, the CMA-ES has been augmented with an adaptive uncertainty handling mechanism. Because uncertainty is a typical property of RL problems this new algorithm, termed UH-CMA-ES, is promising for RL. The UH-CMA-ES dynamically adjusts the number of episodes considered in each evaluation of a policy. It controls the signal to noise ratio such that it is just high enough for a sufficiently good ranking of candidate policies, which in turn allows the evolutionary learning to find better solutions. This significantly increases the learning speed as well as the robustness without impairing the quality of the final solutions. We evaluate the UH-CMA-ES on fully and partially observable Markov decision processes with random start states and noisy observations. A canonical natural policy gradient method and random search serve as a baseline for comparison.","PeriodicalId":193093,"journal":{"name":"Proceedings of the 11th Annual conference on Genetic and evolutionary computation","volume":"10 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2009-07-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132996748","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Mathematical modeling for gene regulative networks (GRNs) provides an effective tool for hypothesis testing in biology. A necessary step in setting up such models is the estimation of model parameters, i.e., an optimization process during which the difference between model output and given experimental data is minimized. This parameter estimation step is often difficult, especially for larger systems due to often incomplete quantitative data, the large size of the parameter space, and non-linearities in system behavior. Addressing the task of parameter estimation, we investigate the influence multiobjectivization can have on the optimization process. On the example of an established model for the segment polarity GRN in Drosophila, we test different multiobjectivization scenarios compared to a singleobjective function proposed earlier for the parameter optimization of the segment polarity network model. Since, instead of a single optimal parameter setting, a set of optimal parameter settings exists for this GRN, the comparison of the different optimization scenarios focuses on the capabilities of the different scenarios to identify optimal parameter settings showing good diversity in the parameter space. By embedding the objective functions in an evolutionary algorithm (EA), we show the superiority of the multiobjective approaches in exploring the model's parameter space.
{"title":"Multiobjectivization for parameter estimation: a case-study on the segment polarity network of drosophila","authors":"T. Hohm, E. Zitzler","doi":"10.1145/1569901.1569931","DOIUrl":"https://doi.org/10.1145/1569901.1569931","url":null,"abstract":"Mathematical modeling for gene regulative networks (GRNs) provides an effective tool for hypothesis testing in biology. A necessary step in setting up such models is the estimation of model parameters, i.e., an optimization process during which the difference between model output and given experimental data is minimized. This parameter estimation step is often difficult, especially for larger systems due to often incomplete quantitative data, the large size of the parameter space, and non-linearities in system behavior. Addressing the task of parameter estimation, we investigate the influence multiobjectivization can have on the optimization process. On the example of an established model for the segment polarity GRN in Drosophila, we test different multiobjectivization scenarios compared to a singleobjective function proposed earlier for the parameter optimization of the segment polarity network model. Since, instead of a single optimal parameter setting, a set of optimal parameter settings exists for this GRN, the comparison of the different optimization scenarios focuses on the capabilities of the different scenarios to identify optimal parameter settings showing good diversity in the parameter space. By embedding the objective functions in an evolutionary algorithm (EA), we show the superiority of the multiobjective approaches in exploring the model's parameter space.","PeriodicalId":193093,"journal":{"name":"Proceedings of the 11th Annual conference on Genetic and evolutionary computation","volume":"26 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2009-07-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133067286","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：