Pub Date : 2018-07-01DOI: 10.1109/CEC.2018.8477931
Xiu Cheng, Will N. Browne, Mengjie Zhang
Learning Classifier Systems (LCSs) have been widely used to tackle Reinforcement Learning (RL) problems as they have a good generalization ability and provide a simple understandable rule-based solution. The accuracy-based LCS, XCS, has been most popularly used for single-objective RL problems. As many real-world problems exhibit multiple conflicting objectives recent work has sought to adapt XCS to Multi-Objective Reinforcement Learning (MORL) tasks. However, many of these algorithms need large storage or cannot discover the Pareto Optimal solutions. This is due to the complexity of finding a policy having multiple steps to multiple possible objectives. This paper aims to employ a decomposition strategy based on MOEA/D in XCS to approximate complex Pareto Fronts. In order to achieve multi-objective learning, a new MORL algorithm has been developed based on XCS and MOEA/D. The experimental results show that on complex bi-objective maze problems our MORL algorithm is able to learn a group of Pareto optimal solutions for MORL problems without huge storage. Analysis of the learned policies shows successful trade-offs between the distance to the reward versus the amount of reward itself.
{"title":"Decomposition Based Multi-Objective Evolutionary Algorithm in XCS for Multi-Objective Reinforcement Learning","authors":"Xiu Cheng, Will N. Browne, Mengjie Zhang","doi":"10.1109/CEC.2018.8477931","DOIUrl":"https://doi.org/10.1109/CEC.2018.8477931","url":null,"abstract":"Learning Classifier Systems (LCSs) have been widely used to tackle Reinforcement Learning (RL) problems as they have a good generalization ability and provide a simple understandable rule-based solution. The accuracy-based LCS, XCS, has been most popularly used for single-objective RL problems. As many real-world problems exhibit multiple conflicting objectives recent work has sought to adapt XCS to Multi-Objective Reinforcement Learning (MORL) tasks. However, many of these algorithms need large storage or cannot discover the Pareto Optimal solutions. This is due to the complexity of finding a policy having multiple steps to multiple possible objectives. This paper aims to employ a decomposition strategy based on MOEA/D in XCS to approximate complex Pareto Fronts. In order to achieve multi-objective learning, a new MORL algorithm has been developed based on XCS and MOEA/D. The experimental results show that on complex bi-objective maze problems our MORL algorithm is able to learn a group of Pareto optimal solutions for MORL problems without huge storage. Analysis of the learned policies shows successful trade-offs between the distance to the reward versus the amount of reward itself.","PeriodicalId":212677,"journal":{"name":"2018 IEEE Congress on Evolutionary Computation (CEC)","volume":"159 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123081286","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}
Pub Date : 2018-07-01DOI: 10.1109/CEC.2018.8477697
Junkun Zhong, Yuping Wang, Hui Du, Wuning Tong
In view of the shortcomings that many clustering algorithms such as K-means clustering algorithm are not suitable for the non-convex dataset and the Affinity Propagation (AP) algorithm may cluster two adjacent different class points into one class, we proposed a new clustering algorithm by using boundary information. The idea of the proposed algorithm in this paper is as follows: First, use the number of points contained in each point's neighborhood as its density, and consider the points whose density are below the average density as boundary points. Then, count the number of boundary points. If the number of boundary points is larger than a given threshold then clustering is carried out by transfer ideas directly, otherwise boundary points will be regarded as the cluster boundary wall. When the boundary points are encountered in the transitive clustering process, the transfer stopped and selected an unprocessed non-boundary point to start clustering process as above again until all non-boundary points are processed, so as to effectively prevent clustering two adjacent different class points into one class. Because of the clustering of transfer idea, the proposed algorithm is applicable to nonconvex datasets, and different clustering schemes are adopted according to the number of boundary points which increases the applicability of the algorithm. Experimental results on synthetic datasets and standard datasets show that the algorithm proposed in this paper is efficient.
{"title":"A New Clustering Algorithm by Using Boundary Information","authors":"Junkun Zhong, Yuping Wang, Hui Du, Wuning Tong","doi":"10.1109/CEC.2018.8477697","DOIUrl":"https://doi.org/10.1109/CEC.2018.8477697","url":null,"abstract":"In view of the shortcomings that many clustering algorithms such as K-means clustering algorithm are not suitable for the non-convex dataset and the Affinity Propagation (AP) algorithm may cluster two adjacent different class points into one class, we proposed a new clustering algorithm by using boundary information. The idea of the proposed algorithm in this paper is as follows: First, use the number of points contained in each point's neighborhood as its density, and consider the points whose density are below the average density as boundary points. Then, count the number of boundary points. If the number of boundary points is larger than a given threshold then clustering is carried out by transfer ideas directly, otherwise boundary points will be regarded as the cluster boundary wall. When the boundary points are encountered in the transitive clustering process, the transfer stopped and selected an unprocessed non-boundary point to start clustering process as above again until all non-boundary points are processed, so as to effectively prevent clustering two adjacent different class points into one class. Because of the clustering of transfer idea, the proposed algorithm is applicable to nonconvex datasets, and different clustering schemes are adopted according to the number of boundary points which increases the applicability of the algorithm. Experimental results on synthetic datasets and standard datasets show that the algorithm proposed in this paper is efficient.","PeriodicalId":212677,"journal":{"name":"2018 IEEE Congress on Evolutionary Computation (CEC)","volume":"32 8 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123227320","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}
Pub Date : 2018-07-01DOI: 10.1109/CEC.2018.8477934
A. G. Bari, Alessio Gaspar, R. P. Wiegand, Anthony Bucci
The Population-based Pareto Hill Climber (P-PHC) algorithm exemplifies coevolutionary computation approaches that manage a group of candidate solutions both used as a population to explore the underlying search space as well as an archive preserving solutions that meet the adopted solution concept. In some circumstances when parsimonious evaluations are desired, inefficiencies can arise from using the same group of candidate solutions for both purposes. The reliance, in such algorithms, on the otherwise beneficial Pareto dominance concept can create bottlenecks on search progress as most newly generated solutions are non-dominated, and thus appear equally qualified to selection, when compared to the current ones they should eventually replace. We propose new selection conditions that include both Pareto dominated and Pareto non-dominated solutions, as well as other factors to help provide distinctions for selection. The potential benefits of also considering Pareto non-dominated solutions are illustrated by a visualization of the underlying interaction space in terms of levels. In addition, we define some new performance metrics that allow one to compare our various selection methods in terms of ideal evaluation of coevolution. Fewer duplicate solutions are retained in the final generation, thus allowing for more efficient usage of the fixed population size.
基于种群的Pareto Hill Climber (P-PHC)算法举例说明了协同进化计算方法，该方法管理一组候选解决方案，这些解决方案既用作探索底层搜索空间的种群，又用于满足所采用的解决方案概念的存档保存解决方案。在某些情况下，当需要进行简洁的评估时，为两个目的使用同一组候选解决方案可能会导致效率低下。在这样的算法中，对帕累托支配概念的依赖可能会对搜索过程造成瓶颈，因为大多数新生成的解决方案都是非支配的，因此与它们最终应该取代的当前解决方案相比，它们似乎同样适合选择。我们提出了新的选择条件，包括帕累托支配和帕累托非支配的解决方案，以及其他因素，以帮助提供选择的区别。考虑帕累托非主导解决方案的潜在好处，可以通过层次的潜在交互空间的可视化来说明。此外，我们定义了一些新的性能指标，允许人们根据共同进化的理想评估来比较我们的各种选择方法。在最后一代中保留较少的重复解决方案，从而允许更有效地使用固定的种群大小。
{"title":"Selection Methods to Relax Strict Acceptance Condition in Test-Based Coevolution","authors":"A. G. Bari, Alessio Gaspar, R. P. Wiegand, Anthony Bucci","doi":"10.1109/CEC.2018.8477934","DOIUrl":"https://doi.org/10.1109/CEC.2018.8477934","url":null,"abstract":"The Population-based Pareto Hill Climber (P-PHC) algorithm exemplifies coevolutionary computation approaches that manage a group of candidate solutions both used as a population to explore the underlying search space as well as an archive preserving solutions that meet the adopted solution concept. In some circumstances when parsimonious evaluations are desired, inefficiencies can arise from using the same group of candidate solutions for both purposes. The reliance, in such algorithms, on the otherwise beneficial Pareto dominance concept can create bottlenecks on search progress as most newly generated solutions are non-dominated, and thus appear equally qualified to selection, when compared to the current ones they should eventually replace. We propose new selection conditions that include both Pareto dominated and Pareto non-dominated solutions, as well as other factors to help provide distinctions for selection. The potential benefits of also considering Pareto non-dominated solutions are illustrated by a visualization of the underlying interaction space in terms of levels. In addition, we define some new performance metrics that allow one to compare our various selection methods in terms of ideal evaluation of coevolution. Fewer duplicate solutions are retained in the final generation, thus allowing for more efficient usage of the fixed population size.","PeriodicalId":212677,"journal":{"name":"2018 IEEE Congress on Evolutionary Computation (CEC)","volume":"28 5 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123418045","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}
Tandem mass spectrometry (MS/MS) is currently the most commonly used technology in proteomics for identifying proteins in complex biological samples. Mass spectrometers can produce a large number of MS/MS spectra each of which has hundreds of peaks. These peaks normally contain background noise, therefore a preprocessing step to filter the noise peaks can improve the accuracy and reliability of peptide identification. This paper proposes to preprocess the data by classifying peaks as noise peaks or signal peaks, i.e., a highly-imbalanced binary classification task, and uses genetic programming (GP) to address this task. The expectation is to increase the peptide identification reliability. Meanwhile, six different types of classification algorithms in addition to GP are used on various imbalance ratios and evaluated in terms of the average accuracy and recall. The GP method appears to be the best in the retention of more signal peaks as examined on a benchmark dataset containing 1, 674 MS/MS spectra. To further evaluate the effectiveness of the GP method, the preprocessed spectral data is submitted to a benchmark de novo sequencing software, PEAKS, to identify the peptides. The results show that the proposed method improves the reliability of peptide identification compared to the original un-preprocessed data and the intensity-based thresholding methods.
{"title":"Genetic Programming for Preprocessing Tandem Mass Spectra to Improve the Reliability of Peptide Identification","authors":"Samaneh Azari, Mengjie Zhang, Bing Xue, Lifeng Peng","doi":"10.1109/CEC.2018.8477810","DOIUrl":"https://doi.org/10.1109/CEC.2018.8477810","url":null,"abstract":"Tandem mass spectrometry (MS/MS) is currently the most commonly used technology in proteomics for identifying proteins in complex biological samples. Mass spectrometers can produce a large number of MS/MS spectra each of which has hundreds of peaks. These peaks normally contain background noise, therefore a preprocessing step to filter the noise peaks can improve the accuracy and reliability of peptide identification. This paper proposes to preprocess the data by classifying peaks as noise peaks or signal peaks, i.e., a highly-imbalanced binary classification task, and uses genetic programming (GP) to address this task. The expectation is to increase the peptide identification reliability. Meanwhile, six different types of classification algorithms in addition to GP are used on various imbalance ratios and evaluated in terms of the average accuracy and recall. The GP method appears to be the best in the retention of more signal peaks as examined on a benchmark dataset containing 1, 674 MS/MS spectra. To further evaluate the effectiveness of the GP method, the preprocessed spectral data is submitted to a benchmark de novo sequencing software, PEAKS, to identify the peptides. The results show that the proposed method improves the reliability of peptide identification compared to the original un-preprocessed data and the intensity-based thresholding methods.","PeriodicalId":212677,"journal":{"name":"2018 IEEE Congress on Evolutionary Computation (CEC)","volume":"6 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121690008","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 proposes an improved $epsilon$ constrained handling method (IEpsilon) for solving constrained single-objective optimization problems (CSOPs). The IEpsilon method adaptively adjusts the value of $epsilon$ according to the proportion of feasible solutions in the current population, which has an ability to balance the search between feasible regions and infeasible regions during the evolutionary process. The proposed constrained handling method is embedded to the differential evolutionary algorithm LSHADE44 to solve CSOPs. Furthermore, a new mutation operator DE/randr1*/1 is proposed in the LSHADE44-IEpsilon. In this paper, twenty-eight CSOPs given by “Problem Definitions and Evaluation Criteria for the CEC 2017 Competition on Constrained Real-Parameter Optimization” are tested by the LSHADE44-IEpsilon and four other differential evolution algorithms CAL-SHADE, LSHADE44+IDE, LSHADE44 and UDE. The experimental results show that the LSHADE44-IEpsilon outperforms these compared algorithms, which indicates that the IEpsilon is an effective constraint-handling method to solve the CEC2017 benchmarks.
{"title":"LSHADE44 with an Improved $epsilon$ Constraint-Handling Method for Solving Constrained Single-Objective Optimization Problems","authors":"Zhun Fan, Yi Fang, Wenji Li, Yutong Yuan, Zhaojun Wang, Xinchao Bian","doi":"10.1109/CEC.2018.8477943","DOIUrl":"https://doi.org/10.1109/CEC.2018.8477943","url":null,"abstract":"This paper proposes an improved $epsilon$ constrained handling method (IEpsilon) for solving constrained single-objective optimization problems (CSOPs). The IEpsilon method adaptively adjusts the value of $epsilon$ according to the proportion of feasible solutions in the current population, which has an ability to balance the search between feasible regions and infeasible regions during the evolutionary process. The proposed constrained handling method is embedded to the differential evolutionary algorithm LSHADE44 to solve CSOPs. Furthermore, a new mutation operator DE/randr1*/1 is proposed in the LSHADE44-IEpsilon. In this paper, twenty-eight CSOPs given by “Problem Definitions and Evaluation Criteria for the CEC 2017 Competition on Constrained Real-Parameter Optimization” are tested by the LSHADE44-IEpsilon and four other differential evolution algorithms CAL-SHADE, LSHADE44+IDE, LSHADE44 and UDE. The experimental results show that the LSHADE44-IEpsilon outperforms these compared algorithms, which indicates that the IEpsilon is an effective constraint-handling method to solve the CEC2017 benchmarks.","PeriodicalId":212677,"journal":{"name":"2018 IEEE Congress on Evolutionary Computation (CEC)","volume":"67 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121872391","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}
Pub Date : 2018-07-01DOI: 10.1109/CEC.2018.8477833
Tulio P. Vieira, P. E. M. Almeida, M. Meireles, M. Souza
This work aims to study the operational scheduling of hydraulic pumps in a Treated Water Lift Station (TWLS) using computational intelligence techniques. This scheduling is very important to reduce electricity consumption of TWLS. For the experiments, a typical TWLS composed of two pumps and a reservoir is simulated. The choice of operation periods is obtained to minimize expenses with electrical energy, by means of an optimization task. From the hydraulic power spent, the TWLS electrical consumption is calculated. A factor $lambda$ is used to correlate number of pumps starts and corresponding maintenance costs. An electrical consumption function, adjusted with this maintenance factor, is used as the objective function to be optimized. In this context, two meta-heuristics are compared: Simulated Annealing (SA) and a hybrid instance of Genetic Algorithms (HGA). Both meta-heuristic approaches were chosen because the reduction of energy and maintenance expenses can be seen as a nonlinear optimization problem, in addition to both techniques being used successfully to solve several real World problems. A statistical inference based objective comparison is done between results of both algorithms, and SA showed to achieve better results. After optimizing the activities related to this scheduling, it is possible to verify a reduction of up to 28.0% in electrical energy expenses, when compared to actual non-optimized operation.
{"title":"Use of Computational Intelligence for Scheduling of Pumps in Water Distribution Systems: a comparison between optimization algorithms","authors":"Tulio P. Vieira, P. E. M. Almeida, M. Meireles, M. Souza","doi":"10.1109/CEC.2018.8477833","DOIUrl":"https://doi.org/10.1109/CEC.2018.8477833","url":null,"abstract":"This work aims to study the operational scheduling of hydraulic pumps in a Treated Water Lift Station (TWLS) using computational intelligence techniques. This scheduling is very important to reduce electricity consumption of TWLS. For the experiments, a typical TWLS composed of two pumps and a reservoir is simulated. The choice of operation periods is obtained to minimize expenses with electrical energy, by means of an optimization task. From the hydraulic power spent, the TWLS electrical consumption is calculated. A factor $lambda$ is used to correlate number of pumps starts and corresponding maintenance costs. An electrical consumption function, adjusted with this maintenance factor, is used as the objective function to be optimized. In this context, two meta-heuristics are compared: Simulated Annealing (SA) and a hybrid instance of Genetic Algorithms (HGA). Both meta-heuristic approaches were chosen because the reduction of energy and maintenance expenses can be seen as a nonlinear optimization problem, in addition to both techniques being used successfully to solve several real World problems. A statistical inference based objective comparison is done between results of both algorithms, and SA showed to achieve better results. After optimizing the activities related to this scheduling, it is possible to verify a reduction of up to 28.0% in electrical energy expenses, when compared to actual non-optimized operation.","PeriodicalId":212677,"journal":{"name":"2018 IEEE Congress on Evolutionary Computation (CEC)","volume":"3 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125280748","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}
Pub Date : 2018-07-01DOI: 10.1109/CEC.2018.8477730
Ye Tian, Xiaoshu Xiang, Xing-yi Zhang, Ran Cheng, Yaochu Jin
The effectiveness of evolutionary algorithms have been verified on multi-objective optimization, and a large number of multi-objective evolutionary algorithms have been proposed during the last two decades. To quantitatively compare the performance of different algorithms, a set of uniformly distributed reference points sampled on the Pareto fronts of benchmark problems are needed in the calculation of many performance metrics. However, not much work has been done to investigate the method for sampling reference points on Pareto fronts, even though it is not an easy task for many Pareto fronts with irregular shapes. More recently, an evolutionary multi-objective optimization platform was proposed by us, called PlatEMO, which can automatically generate reference points on each Pareto front and use them to calculate the performance metric values. In this paper, we report the reference point sampling methods used in PlatEMO for different types of Pareto fronts. Experimental results show that the reference points generated by the proposed sampling methods can evaluate the performance of algorithms more accurately than randomly sampled reference points.
{"title":"Sampling Reference Points on the Pareto Fronts of Benchmark Multi-Objective Optimization Problems","authors":"Ye Tian, Xiaoshu Xiang, Xing-yi Zhang, Ran Cheng, Yaochu Jin","doi":"10.1109/CEC.2018.8477730","DOIUrl":"https://doi.org/10.1109/CEC.2018.8477730","url":null,"abstract":"The effectiveness of evolutionary algorithms have been verified on multi-objective optimization, and a large number of multi-objective evolutionary algorithms have been proposed during the last two decades. To quantitatively compare the performance of different algorithms, a set of uniformly distributed reference points sampled on the Pareto fronts of benchmark problems are needed in the calculation of many performance metrics. However, not much work has been done to investigate the method for sampling reference points on Pareto fronts, even though it is not an easy task for many Pareto fronts with irregular shapes. More recently, an evolutionary multi-objective optimization platform was proposed by us, called PlatEMO, which can automatically generate reference points on each Pareto front and use them to calculate the performance metric values. In this paper, we report the reference point sampling methods used in PlatEMO for different types of Pareto fronts. Experimental results show that the reference points generated by the proposed sampling methods can evaluate the performance of algorithms more accurately than randomly sampled reference points.","PeriodicalId":212677,"journal":{"name":"2018 IEEE Congress on Evolutionary Computation (CEC)","volume":"15 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125376879","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}
Pub Date : 2018-07-01DOI: 10.1109/CEC.2018.8477921
Y. Sun, Bing Xue, Mengjie Zhang, G. Yen
Deep learning algorithms have shown their superiority especially in addressing challenging machine learning tasks. The best performance of deep learning algorithms can be reached only when their hyper-parameters have been successfully optimized. However, the hyper-parameter optimization problem is non-convex and non-differentiable, and traditional optimization algorithms are incapable of addressing them well. Evolutionary algorithms are a class of meta-heuristic search algorithms, preferred for optimizing real-world problems due largely to their no mathematical requirements on the problems to be optimized. Although most researchers from the community of deep learning are aware of the effectiveness of evolutionary algorithms in optimizing the hyper-parameters of deep learning algorithms, they still believe that the grid search method is more effective when the number of hyper-parameters is small. To clarify this, we design a hyper-parameter optimization method by using particle swarm optimization that is a widely used evolutionary algorithm, to perform 192 experimental comparisons for stacked auto-encoders that are a class of deep learning algorithms with a relative small number of hyper-parameters, investigate and compare the classification accuracy and computational complexity with those of the grid search method on eight widely used image classification benchmark datasets. The experimental results show that the proposed algorithm can achieve the comparative classification accuracy but saving 10x-100x computational complexity compared with the grid search method.
{"title":"An Experimental Study on Hyper-parameter Optimization for Stacked Auto-Encoders","authors":"Y. Sun, Bing Xue, Mengjie Zhang, G. Yen","doi":"10.1109/CEC.2018.8477921","DOIUrl":"https://doi.org/10.1109/CEC.2018.8477921","url":null,"abstract":"Deep learning algorithms have shown their superiority especially in addressing challenging machine learning tasks. The best performance of deep learning algorithms can be reached only when their hyper-parameters have been successfully optimized. However, the hyper-parameter optimization problem is non-convex and non-differentiable, and traditional optimization algorithms are incapable of addressing them well. Evolutionary algorithms are a class of meta-heuristic search algorithms, preferred for optimizing real-world problems due largely to their no mathematical requirements on the problems to be optimized. Although most researchers from the community of deep learning are aware of the effectiveness of evolutionary algorithms in optimizing the hyper-parameters of deep learning algorithms, they still believe that the grid search method is more effective when the number of hyper-parameters is small. To clarify this, we design a hyper-parameter optimization method by using particle swarm optimization that is a widely used evolutionary algorithm, to perform 192 experimental comparisons for stacked auto-encoders that are a class of deep learning algorithms with a relative small number of hyper-parameters, investigate and compare the classification accuracy and computational complexity with those of the grid search method on eight widely used image classification benchmark datasets. The experimental results show that the proposed algorithm can achieve the comparative classification accuracy but saving 10x-100x computational complexity compared with the grid search method.","PeriodicalId":212677,"journal":{"name":"2018 IEEE Congress on Evolutionary Computation (CEC)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125562667","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}
Pub Date : 2018-07-01DOI: 10.1109/CEC.2018.8477729
Soheila Sadeghiram, Hui Ma, Gang Chen
Automatic Web service composition has received much interest in the last decades. Data-intensive concepts have provided a promising computing paradigm for data-intensive Web service composition. Due to the complexity of the problem, metaheuristics in particular Evolutionary Computing (EC) techniques have been used for solving this composition problem. However, most of the current works neglected the distributed nature of data-intensive Web services. In this paper, we study the problem of distributed data-intensive service composition and propose a model which integrates attributes of constituent data-intensive Web services and attributes of the network. The core idea is to propose a communication cost and time model of a composed Web service considering communication delay and cost. We therefore propose a novel method based on Genetic Algorithm (GA) which uses a variation of K-means clustering algorithm.
{"title":"Cluster-Guided Genetic Algorithm for Distributed Data-intensive Web Service Composition","authors":"Soheila Sadeghiram, Hui Ma, Gang Chen","doi":"10.1109/CEC.2018.8477729","DOIUrl":"https://doi.org/10.1109/CEC.2018.8477729","url":null,"abstract":"Automatic Web service composition has received much interest in the last decades. Data-intensive concepts have provided a promising computing paradigm for data-intensive Web service composition. Due to the complexity of the problem, metaheuristics in particular Evolutionary Computing (EC) techniques have been used for solving this composition problem. However, most of the current works neglected the distributed nature of data-intensive Web services. In this paper, we study the problem of distributed data-intensive service composition and propose a model which integrates attributes of constituent data-intensive Web services and attributes of the network. The core idea is to propose a communication cost and time model of a composed Web service considering communication delay and cost. We therefore propose a novel method based on Genetic Algorithm (GA) which uses a variation of K-means clustering algorithm.","PeriodicalId":212677,"journal":{"name":"2018 IEEE Congress on Evolutionary Computation (CEC)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129250741","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}
Pub Date : 2018-07-01DOI: 10.1109/CEC.2018.8477663
N. Nnamoko, A. Hussain, D. England
An exploratory research is presented to gauge the impact of feature selection on heterogeneous ensembles. The task is to predict diabetes onset with healthcare data obtained from UC Irvine (VCI) database. Evidence suggests that accuracy and diversity are the two vital requirements to achieve good ensembles. Therefore, the research presented in this paper exploits diversity from heterogeneous base classifiers; and the optimisation effect of feature subset selection in order to improve accuracy. Five widely used classifiers are employed for the ensembles and a meta-classifier is used to aggregate their outputs. The results are presented and compared with similar studies that used the same dataset within the literature. It is shown that by using the proposed method, diabetes onset prediction can be done with higher accuracy.
{"title":"Predicting Diabetes Onset: An Ensemble Supervised Learning Approach","authors":"N. Nnamoko, A. Hussain, D. England","doi":"10.1109/CEC.2018.8477663","DOIUrl":"https://doi.org/10.1109/CEC.2018.8477663","url":null,"abstract":"An exploratory research is presented to gauge the impact of feature selection on heterogeneous ensembles. The task is to predict diabetes onset with healthcare data obtained from UC Irvine (VCI) database. Evidence suggests that accuracy and diversity are the two vital requirements to achieve good ensembles. Therefore, the research presented in this paper exploits diversity from heterogeneous base classifiers; and the optimisation effect of feature subset selection in order to improve accuracy. Five widely used classifiers are employed for the ensembles and a meta-classifier is used to aggregate their outputs. The results are presented and compared with similar studies that used the same dataset within the literature. It is shown that by using the proposed method, diabetes onset prediction can be done with higher accuracy.","PeriodicalId":212677,"journal":{"name":"2018 IEEE Congress on Evolutionary Computation (CEC)","volume":"14 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129293925","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：