F. Hutter, H. Hoos, Kevin Leyton-Brown, Kevin P. Murphy
This work experimentally investigates model-based approaches for optimising the performance of parameterised randomised algorithms. We restrict our attention to procedures based on Gaussian process models, the most widely-studied family of models for this problem. We evaluated two approaches from the literature, and found that sequential parameter optimisation (SPO) [4] offered the most robust performance. We then investigated key design decisions within the SPO paradigm, characterising the performance consequences of each. Based on these findings, we propose a new version of SPO, dubbed SPO+, which extends SPO with a novel intensification procedure and log-transformed response values. Finally, in a domain for which performance results for other (model-free) parameter optimisation approaches are available, we demonstrate that SPO+ achieves state-of-the-art performance.
{"title":"An experimental investigation of model-based parameter optimisation: SPO and beyond","authors":"F. Hutter, H. Hoos, Kevin Leyton-Brown, Kevin P. Murphy","doi":"10.1145/1569901.1569940","DOIUrl":"https://doi.org/10.1145/1569901.1569940","url":null,"abstract":"This work experimentally investigates model-based approaches for optimising the performance of parameterised randomised algorithms. We restrict our attention to procedures based on Gaussian process models, the most widely-studied family of models for this problem. We evaluated two approaches from the literature, and found that sequential parameter optimisation (SPO) [4] offered the most robust performance. We then investigated key design decisions within the SPO paradigm, characterising the performance consequences of each. Based on these findings, we propose a new version of SPO, dubbed SPO+, which extends SPO with a novel intensification procedure and log-transformed response values. Finally, in a domain for which performance results for other (model-free) parameter optimisation approaches are available, we demonstrate that SPO+ achieves state-of-the-art performance.","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":"115500147","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}
Neural networks are a common choice for solving classification problems, but require experimental adjustments of the topology, weights and thresholds to be effective. Success has been seen in the development of neural networks with evolutionary algorithms, making the extension of this work to classification problems a logical step. This paper presents the first known use of the Gene Expression Programming-based GEP-NN algorithm to design neural networks for classification purposes. The system uses pairwise decomposition to produce a series of binary classifiers for a given multi-class problem, with the results of the classifier set being combined by majority vote.
{"title":"On the evolution of neural networks for pairwise classification using gene expression programming","authors":"Stephen Johns, M. Santos","doi":"10.1145/1569901.1570227","DOIUrl":"https://doi.org/10.1145/1569901.1570227","url":null,"abstract":"Neural networks are a common choice for solving classification problems, but require experimental adjustments of the topology, weights and thresholds to be effective. Success has been seen in the development of neural networks with evolutionary algorithms, making the extension of this work to classification problems a logical step. This paper presents the first known use of the Gene Expression Programming-based GEP-NN algorithm to design neural networks for classification purposes. The system uses pairwise decomposition to produce a series of binary classifiers for a given multi-class problem, with the results of the classifier set being combined by majority vote.","PeriodicalId":193093,"journal":{"name":"Proceedings of the 11th Annual conference on Genetic and evolutionary computation","volume":"77 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":"115718821","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 introduces a new component based model that makes it relatively simple to prove that certain types of landscapes are elementary. We use the model to reconstruct proofs for the Traveling Salesman Problem, Graph Coloring and Min-Cut Graph Partitioning. The same model is then used to efficiently compute the average values over particular partial neighborhoods for these same problems. For Graph Coloring and Min-Cut Graph Partitioning, this computation can be used to focus search on those moves that are most likely to yield an improving move, ignoring moves that cannot yield an improving move. Let x be a candidate solution with objective function value f(x). The mean value of the objective function over the entire landscape is denoted f. Normally in an elementary landscape one can only be sure that a neighborhood includes an improving move (assuming minimization) if f(x) > f. However, by computing the expected value of an appropriate partial neighborhood it is sometimes possible to know that an improving move exists in the partial neighborhood even when f(x) < f.
{"title":"Partial neighborhoods of elementary landscapes","authors":"L. D. Whitley, Andrew M. Sutton","doi":"10.1145/1569901.1569954","DOIUrl":"https://doi.org/10.1145/1569901.1569954","url":null,"abstract":"This paper introduces a new component based model that makes it relatively simple to prove that certain types of landscapes are elementary. We use the model to reconstruct proofs for the Traveling Salesman Problem, Graph Coloring and Min-Cut Graph Partitioning. The same model is then used to efficiently compute the average values over particular partial neighborhoods for these same problems. For Graph Coloring and Min-Cut Graph Partitioning, this computation can be used to focus search on those moves that are most likely to yield an improving move, ignoring moves that cannot yield an improving move. Let x be a candidate solution with objective function value f(x). The mean value of the objective function over the entire landscape is denoted f. Normally in an elementary landscape one can only be sure that a neighborhood includes an improving move (assuming minimization) if f(x) > f. However, by computing the expected value of an appropriate partial neighborhood it is sometimes possible to know that an improving move exists in the partial neighborhood even when f(x) < f.","PeriodicalId":193093,"journal":{"name":"Proceedings of the 11th Annual conference on Genetic and evolutionary computation","volume":"7 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":"114402290","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 consider a formulation of the biobjective soft graph coloring problem so as to simultaneously minimize the number of colors used as well as the number of edges that connect vertices of the same color. We aim to evolve hyperheuristics for this class of problem using multiobjective genetic programming (MOGP). The major advantage being that these hyperheuristics can then be applied to any instance of this problem. We test the hyperheuristics on benchmark graph coloring problems, and in the absence of an actual Pareto-front, we compare the solutions obtained with existing heuristics. We then further improve the quality of hyperheuristics evolved, and try to make them closer to human-designed heuristics.
{"title":"Evolution of hyperheuristics for the biobjective graph coloring problem using multiobjective genetic programming","authors":"Paresh Tolay, Rajeev Kumar","doi":"10.1145/1569901.1570247","DOIUrl":"https://doi.org/10.1145/1569901.1570247","url":null,"abstract":"We consider a formulation of the biobjective soft graph coloring problem so as to simultaneously minimize the number of colors used as well as the number of edges that connect vertices of the same color. We aim to evolve hyperheuristics for this class of problem using multiobjective genetic programming (MOGP). The major advantage being that these hyperheuristics can then be applied to any instance of this problem. We test the hyperheuristics on benchmark graph coloring problems, and in the absence of an actual Pareto-front, we compare the solutions obtained with existing heuristics. We then further improve the quality of hyperheuristics evolved, and try to make them closer to human-designed heuristics.","PeriodicalId":193093,"journal":{"name":"Proceedings of the 11th Annual conference on Genetic and evolutionary computation","volume":"142 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":"114818848","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 1: ant colony optimization and swarm intelligence","authors":"M. Birattari, T. Stützle","doi":"10.1145/3257495","DOIUrl":"https://doi.org/10.1145/3257495","url":null,"abstract":"","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":"115023484","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}
Helicopter hovering is an important challenge problem in the field of reinforcement learning. This paper considers several neuroevolutionary approaches to discovering robust controllers for a generalized version of the problem used in the 2008 Reinforcement Learning Competition, in which wind in the helicopter's environment varies from run to run. We present the simple model-free strategy that won first place in the competition and also describe several more complex model-based approaches. Our empirical results demonstrate that neuroevolution is effective at optimizing the weights of multi-layer perceptrons, that linear regression is faster and more effective than evolution for learning models, and that model-based approaches can outperform the simple model-free strategy, especially if prior knowledge is used to aid model learning.
{"title":"Neuroevolutionary reinforcement learning for generalized helicopter control","authors":"Rogier Koppejan, Shimon Whiteson","doi":"10.1145/1569901.1569922","DOIUrl":"https://doi.org/10.1145/1569901.1569922","url":null,"abstract":"Helicopter hovering is an important challenge problem in the field of reinforcement learning. This paper considers several neuroevolutionary approaches to discovering robust controllers for a generalized version of the problem used in the 2008 Reinforcement Learning Competition, in which wind in the helicopter's environment varies from run to run. We present the simple model-free strategy that won first place in the competition and also describe several more complex model-based approaches. Our empirical results demonstrate that neuroevolution is effective at optimizing the weights of multi-layer perceptrons, that linear regression is faster and more effective than evolution for learning models, and that model-based approaches can outperform the simple model-free strategy, especially if prior knowledge is used to aid model learning.","PeriodicalId":193093,"journal":{"name":"Proceedings of the 11th Annual conference on Genetic and evolutionary computation","volume":"83 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":"114533787","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 addresses the problem of multiagent task allocation in extreme teams. An extreme team is composed by a large number of agents with overlapping functionality operating in dynamic environments with possible inter-task constraints. We present eXtreme-Ants, an approximate algorithm for task allocation in extreme teams. The algorithm is inspired by the division of labor in social insects and in the process of recruitment for cooperative transport observed in ant colonies. Division of labor offers fast and efficient decision-making, while the recruitment ensures the allocation of tasks that require simultaneous execution. We compare eXtreme-Ants with two other algorithms for task allocation in extreme teams and we show that it achieves balanced efficiency regarding quality of the solution, communication, and computational effort.
{"title":"An ant based algorithm for task allocation in large-scale and dynamic multiagent scenarios","authors":"F. Santos, A. Bazzan","doi":"10.1145/1569901.1569912","DOIUrl":"https://doi.org/10.1145/1569901.1569912","url":null,"abstract":"This paper addresses the problem of multiagent task allocation in extreme teams. An extreme team is composed by a large number of agents with overlapping functionality operating in dynamic environments with possible inter-task constraints. We present eXtreme-Ants, an approximate algorithm for task allocation in extreme teams. The algorithm is inspired by the division of labor in social insects and in the process of recruitment for cooperative transport observed in ant colonies. Division of labor offers fast and efficient decision-making, while the recruitment ensures the allocation of tasks that require simultaneous execution. We compare eXtreme-Ants with two other algorithms for task allocation in extreme teams and we show that it achieves balanced efficiency regarding quality of the solution, communication, and computational effort.","PeriodicalId":193093,"journal":{"name":"Proceedings of the 11th Annual conference on Genetic and evolutionary computation","volume":"22 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":"114954994","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 12: parallel evolutionary systems","authors":"E. Alba","doi":"10.1145/3257506","DOIUrl":"https://doi.org/10.1145/3257506","url":null,"abstract":"","PeriodicalId":193093,"journal":{"name":"Proceedings of the 11th Annual conference on Genetic and evolutionary computation","volume":"5 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":"116818081","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 explores the capabilities of multi-objective genetic algorithms to cluster genomic data. We used multiple objective functions not only to further expand the clustering abilities of the algorithm, but also to give more biological significance to the results. Particularly, we grouped a large set of proteins described by a set collection of genomic attributes to infer functional interactions among them. We conducted various computational experiments that demonstrated the proficiency of the proposed method when compared to algorithms that rely on a single biological parameter.
{"title":"MOCEA: a multi-objective clustering evolutionary algorithm for inferring protein-protein functional interactions","authors":"J. Tapia, E. Vallejo, E. Morett","doi":"10.1145/1569901.1570164","DOIUrl":"https://doi.org/10.1145/1569901.1570164","url":null,"abstract":"This paper explores the capabilities of multi-objective genetic algorithms to cluster genomic data. We used multiple objective functions not only to further expand the clustering abilities of the algorithm, but also to give more biological significance to the results. Particularly, we grouped a large set of proteins described by a set collection of genomic attributes to infer functional interactions among them. We conducted various computational experiments that demonstrated the proficiency of the proposed method when compared to algorithms that rely on a single biological parameter.","PeriodicalId":193093,"journal":{"name":"Proceedings of the 11th Annual conference on Genetic and evolutionary computation","volume":"58 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":"124664588","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}
Roberto Santana, C. Bielza, J. A. Lozano, P. Larrañaga
One of the uses of the probabilistic models learned by estimation of distribution algorithms is to reveal previous unknown information about the problem structure. In this paper we investigate the mapping between the problem structure and the dependencies captured in the probabilistic models learned by EDAs for a set of multi-objective satisfiability problems. We present and discuss the application of different data mining and visualization techniques for processing and visualizing relevant information from the structure of the learned probabilistic models. We show that also in the case of multi-objective optimization problems, some features of the original problem structure can be translated to the probabilistic models and unveiled by using algorithms that mine the model structures.
{"title":"Mining probabilistic models learned by EDAs in the optimization of multi-objective problems","authors":"Roberto Santana, C. Bielza, J. A. Lozano, P. Larrañaga","doi":"10.1145/1569901.1569963","DOIUrl":"https://doi.org/10.1145/1569901.1569963","url":null,"abstract":"One of the uses of the probabilistic models learned by estimation of distribution algorithms is to reveal previous unknown information about the problem structure. In this paper we investigate the mapping between the problem structure and the dependencies captured in the probabilistic models learned by EDAs for a set of multi-objective satisfiability problems. We present and discuss the application of different data mining and visualization techniques for processing and visualizing relevant information from the structure of the learned probabilistic models. We show that also in the case of multi-objective optimization problems, some features of the original problem structure can be translated to the probabilistic models and unveiled by using algorithms that mine the model structures.","PeriodicalId":193093,"journal":{"name":"Proceedings of the 11th Annual conference on Genetic and evolutionary computation","volume":"1696 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":"129391753","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：