Jisong Kou

Coevolutionary learning of neural network ensemble for complex classification tasks

Ensemble approaches to classification have attracted a great deal of interest recently. This paper presents a novel method for designing the neural network ensemble using coevolutionary algorithm. The bootstrap resampling procedure is employed to obtain different training subsets that are used to estimate different component networks of the ensemble. Then the cooperative coevolutionary algorithm is developed to optimize the ensemble model via the divide-and-cooperative mechanism. All component networks are coevolved in parallel in the scheme of interacting co-adapted subpopulations. The fitness of an individual from a particular subpopulation is assessed by associating it with the representatives from other subpopulations. In order to promote the cooperation of all component networks, the proposed method considers both the accuracy and the diversity among the component networks that are evaluated using the multi-objective Pareto optimality measure. A hybrid output-combination method is designed to determine the final ensemble output. Experimental results illustrate that the proposed method is able to obtain neural network ensemble models with better classification accuracy in comparison with currently popular ensemble algorithms.

Crowding with nearest neighbors replacement for multiple species niching and building blocks preservation in binary multimodal functions optimization

Abstract This paper introduces a novel niching scheme called the q-nearest neighbors replacement (q-NNR) method in the framework of the steady-state GAs (SSGAs) for solving binary multimodal optimization problems. A detailed comparison of the main niching approaches are presented first. The niching paradigm and difference of the selection-recombination genetic algorithms (GAs) and the recombination-replacement SSGAs are discussed. Then the q-NNR is developed by adopting special replacement policies based on the SSGAs; a Boltzmann scheme for dynamically sizing the nearest neighbors set is designed to achieve a speed-up and control the proportion of individuals adapted to different niches. Finally, experiments are carried out on a set of test functions characterized by deception, epistasis, symmetry and multimodality. The results are satisfactory and illustrate the effectivity and efficiency of the proposed niching method.

Two novel encoding strategies based genetic algorithms for circuit partitioning

Circuit partitioning is a key phase in the VLSI design and partitioning algorithm is of great importance. Two styles of genetic algorithms based on different encoding strategies for circuit partitioning are presented. The first adopts the form of 0-1 encoding, and the second uses integer encoding based on modules number. Meanwhile, the corresponding fitness function and genetic operators are designed for each method. Then these two algorithms are implemented to test standard benchmark circuits. Compared with the traditional F-M algorithm, partition results by the two genetic algorithms are markedly improved.

An investigation on niching multiple species based on population replacement strategies for multimodal functions optimization

This paper studies the niching mechanism based on population replacement in the process of evolution to solve the multimodal functions optimization (MMFO) problems. In order to niche multiple species for the MMFO tasks, the overlapping population replacement is surely needed because the offspring population most probably does not inherit all of the genetic information contained in its parental population, and the basic procedure for niching genetic algorithms with overlapping population replacement is presented. Then four niching schemes, the nearest neighbors replacement crowding (NNRC), the species conservation technique (SCT), the HFC-I (implicit hierarchical fair competition), and the CPE (clearing procedure with elitist) are investigated. These niching schemes are characterized with regard to different niching strategies and parameterizations, and the corresponding niching procedures are outlined. Finally, experiments are carried out on a suite of test functions to compare different niching strategies regarding niching efficiency and scalability. Experimental results illustrate the intrinsic difference of the four niching schemes. The NNRC and HFC-I have a mechanism of multiple species coevolution via adapting multiple species to different niches, while the SCT and CPE tend to make use of a mandatory mechanism to conserve species just like the grid searching over the solution space based on species distance or clearing radius. All niching methods are able to deal with complex MMFO problems, while the NNRC and HFC-I show a better performance in terms of niching efficiency and scalability, and are more robust regarding the algorithm parameterization.

HEURISTIC BIVARIATE FORECASTING MODEL OF MULTI-ATTRIBUTE FUZZY TIME SERIES BASED ON FUZZY CLUSTERING

Fuzzy time series has been applied to forecast various domain problems because of its capability to deal with vagueness and incompleteness inherent in data. However, most existing fuzzy time series models cannot cope with multi-attribute time series and remain too subjective in the partition of the universe of discourse. Moreover, these models do not consider the trend factor and the corresponding external time series, which are highly relevant to target series. In the current paper, a heuristic bivariate model is proposed to improve forecasting accuracy, and the proposed model applies fuzzy c-means clustering algorithm to process multi-attribute fuzzy time series and to partition the universe of discourse. Meanwhile, the trend predictors are extracted in the training phase and utilized to select the order of fuzzy relations in the testing phase. Finally, the proper full use of the external series to assist forecasting is discussed. The performance of the proposed model is tested using actual time series including the enrollments at the University of Alabama, the Taiwan Stock Exchange Capitalization Weighted Stock Index (TAIEX) and a sensor dataset. The experimental results show that the proposed model can be utilized for multi-attribute time series and significantly improves the average MAER to 1.19% when compared with other forecasting models.

A novel type of niching methods based on steady-state genetic algorithm

In this paper, a novel niching approach to solve the multimodal function optimization problems is proposed. We firstly analyze and compare the characteristics and behaviors of a variety of niching methods as the fitness sharing, the crowding and deterministic crowding, the restricted mating, and the island model GA with regard to the competition, exploration & exploitation, genetic drift, and the ability to locate and maintain niches. Then we put forward the idea that the local competition of individuals is crucial to realize the distribution equilibria among niches of the optimization functions, and two types of niching methods, q-nearest neighbor replacement and parental neighbor replacement, are formulated by adopting special replacement policies in the setting of the SSGA. Finally, we use a set of test functions to illustrate the efficacy and efficiency of the proposed methods and the DC scheme based on the SSGA.

Candidate Vectors Selection for Training Support Vector Machines

In this paper, a novel and concise method for the selection of candidate vectors (SCV) is proposed based on the structural information of two classes in the input space. First, the Euclidean distance of all samples to the boundary of the other classes is calculated. Then the relative distance is computed to reorder training samples ascendingly, and boundary samples will rank in front of others and have a higher probability to be candidate support vectors. A certain proportion of the foremost ranked samples are selected to form examples subset for training the SVM classification function by using the SMO. For linearly non-separable datasets with noise, an abnormal examples filtering (AEF) procedure is designed to find abnormal examples or outliers that may give rise to the distortion of structural information on the boundaries of two classes. Finally, two datasets are used to test the prediction accuracy of the SVM decision function estimated by the SMO and the AEF+SCV+SMO.

Comprehensive Review of Sleep/Wake Scheduling in Wireless Sensor Networks

Sleep/wake scheduling is an essential consideration in sensor network applications. Finding an optimal sleep/wake scheduling strategy that would minimize computation and communication overhead, be resilient to node failures, and provide high-quality data service is extremely challenging. In this paper, we present and compare several state-of-the-art algorithms and techniques that aim to address the sleep/wake scheduling issue, which are divided into distributed and centralized manners. Meanwhile, the advantages and disadvantages of these scheduling models and algorithms are discussed, together with open research issues.

Dynamics of fitness sharing evolutionary algorithms for coevolution of multiple species

This paper builds the normal model of fitness sharing with proportionate selection on real-valued functions, and derives the dynamic formula to describe the evolution process of the population with the fitness sharing. The normal modeling simulation is investigated on specific test functions, and experimental results illustrate that the normal model is able to describe exactly the dynamics of the fitness sharing EAs and is a good platform to study the behavior of the fitness sharing EAs with regard to niching radius. The experimental results of the normal modeling simulation and the fitness sharing EAs verify the dilemma in finding optimal niche radius to achieve both good niching convergence and niching efficiency, for which a hybrid scheme is proposed to carry out the niching task.

Adaptive simulated annealing for standard cell placement

A standard cell placement algorithm based on adaptive simulated annealing is presented in this paper. Considering the characters of different circuits to be placed, adaptively initial temperature and adaptive searching region are added to traditional simulated annealing algorithm. At the same time, the punishment item in objective function and initial placement approach are improved for the standard cell placement problem. This algorithm is applied to test a set of benchmark circuits, and experiments reveal its advantages in placement results and time performance when compared with the traditional simulated annealing algorithm.