基于仿真的优化的粒子群算法参数选取研究

粒子群优化（Particle Swarm Optimization,PSO）算法参数较少、搜索机制简单,故一直是智能优化算法研究和应用的重点。然而PSO有易早熟、搜索精度不高及搜索性能对参数依赖性强的缺陷。针对此特点,在基于仿真的优化框架下,基于多Agent对融合传统全局最佳和局部最佳的PSO算法人工生命模型进行了仿真,以混合优化算法为计算引擎,对PSO的参数选取进行了重点讨论。利用一系列benchmark函数为例,进行了仿真优化实验和分析,取得了较为满意的结果,从而说明了本思想方法的可行性与可信性。     

A note on genetic algorithms for large-scale feature selection

We introduce the use of genetic algorithms (GA) for the selection of features in the design of automatic pattern classifiers. Our preliminary results suggest that GA is a powerful means of reducing the time for finding near-optimal subsets of features from large sets.     

Parallel algorithms for modules of learning automata

Parallel algorithms are presented for modules of learning automata with the objective of improving their speed of convergence without compromising accuracy. A general procedure suitable for parallelizing a large class of sequential learning algorithms on a shared memory system is proposed. Results are derived to show the quantitative improvements in speed obtainable using parallelization. The efficacy of the procedure is demonstrated by simulation studies on algorithms for common payoff games, parametrized learning automata and pattern classification problems with noisy classification of training samples.     

Attribute selection method based on a hybrid BPNN and PSO algorithms

High dimensional data contain many redundant or irrelevant attributes, which will be difficult for data mining and a variety of pattern recognition. When implementing data mining or a variety of pattern recognition on high dimensional space, it is necessary to reduce the dimension of high dimensional space. In this paper, a new attribute importance measure and selection methods based on attribute ranking was proposed. In proposed attribute selection method, input output correlation (IOC) is applied for calculating attribute’ importance, and then sorts them according to descending order. The hybrid of Back Propagation Neural Network (BPNN) and Particle Swarm Optimization (PSO) algorithms is also proposed. PSO is used to optimize weights and thresholds of BPNN for overcoming the inherent shortcoming of BPNN. The experiment results show the proposed attribute selection method is an effective preproceesing technology.     

监督学习模型指导的函数级编译优化参数选择方法研究

基于机器学习的迭代编译方法可以在对新程序进行迭代编译时,有效预测新程序的最佳优化参数组合。现有方法在模型训练过程中存在优化参数组合搜索效率较低、程序特征表示不恰当、预测精度不高的问题。因此,基于机器学习的迭代编译方法是当前迭代编译领域内的一个研究热点,其研究挑战在于学习算法选择、优化参数搜索以及程序特征表示等问题。基于监督学习技术,提出了一种程序优化参数预测方法。该方法首先通过约束多目标粒子群算法对优化参数空间进行搜索,找到样本函数的最佳优化参数;然后,通过动静结合的程序特征表示技术,对函数特征进行抽取;最后,通过由函数特征和优化参数形成的样本构建监督学习模型,对新程序的优化参数进行预测。分别采用k近邻法和softmax回归建立统计模型,实验结果表明,新方法在NPB测试集和大型科学计算程序上实现了较好的预测性能。     

基于克隆选择原理的自适应免疫性遗传算法研究

针对遗传算法中存在着收敛方向无法控制和没有记忆能力等缺陷,提出了具有免疫功能的克隆遗传算法.该算法把目标函数和制约条件作为抗原,保证所生成的抗体与问题直接相关联,使收敛方向得以控制;对抗原亲和力高的抗体进行克隆记忆,促使优良个体的发育成熟并能有效地遗传到下一代;同时,基于浓度的概念提出对抗体数量进行抑制,确保群体更新的多样性,避免早熟.通过理论分析和实验研究,证明该算法具有快的收敛速度和搜索能力,是一种有效的生物计算方法.     

基于自适应图学习的多目标特征选择算法

针对多目标回归中的特征选择问题,提出一种基于自适应图学习的多目标特征选择算法,在单个框架中同时考虑3种关系结构:输入特征与目标输出、不同目标输出以及样本间的相关结构,并基于上述结构信息进行特征选择.首先,在传统稀疏回归模型中对系数矩阵施加低秩约束,利用低秩学习对特征间相关性以及目标间的依赖关系进行解耦学习;然后,构建基于样本局部结构信息的自适应图学习项,充分利用样本间的相似结构进行特征选择;进一步地,引入基于输出相关性的结构矩阵优化项,使模型能够更加充分地考虑目标间的相关性;最后,提出一种交替优化算法求解目标函数,并从理论上证明算法的收敛性.在公开数据集上的实验表明,所提方法相较于现有主流的多目标特征选择方法具有更好的性能和适用性.     

Prototype selection algorithms for distributed learning

Distributed learning from data is one of the typical tasks solved by distributed data-mining techniques and is seen as a fundamental computational problem. One of the approaches suitable for distributed learning is to select, by data reduction, relevant local patterns, called also prototypes, from geographically distributed databases. Next, locally selected prototypes can be moved to other sites and merged into the global knowledge model. The paper presents three agent-based population learning algorithms for distributed learning. The proposed algorithms are based on agent collaborations in distributed prototype selection processes and on agent collaborations when the learning global model is created. The basic property of the presented algorithms is that the prototypes are selected by agent-based population learning algorithm from data clusters induced at distributed sites. The main goal of the paper is to empirically compare how the way of inducing such clusters can influence the distributed learning performance. The paper investigates the agent-based population learning algorithms used to solve distributed data reduction and gives a brief discussion of the procedures for clusters initialization. Finally, computational experiment results are shown.     

A note on rebound automata

It was unknown whether there exists a context-free language not accepted by any deterministic rebound automaton. This paper solves this problem, and shows that there exists a context-free language not accepted by any nondeterministic rebound automaton. This paper also investigates closure properties of the class of rebound automata.     

An adaptive super-peer selection algorithm considering peers capacity utilizing asynchronous dynamic cellular learning automata

Super-peer networks refer to a class of peer-to-peer networks in which some peers called super-peers are in charge of managing the network. A group of super-peer selection algorithms use the capacity of the peers for the purpose of super-peer selection where the capacity of a peer is defined as a general concept that can be calculated by some properties, such as bandwidth and computational capabilities of that peer. One of the drawbacks of these algorithms is that they do not take into consideration the dynamic nature of peer-to-peer networks in the process of selecting super-peers. In this paper, an adaptive super-peer selection algorithm considering peers capacity based on an asynchronous dynamic cellular learning automaton has been proposed. The proposed cellular learning automaton uses the model of fungal growth as it happens in nature to adjust the attributes of the cells of the cellular learning automaton in order to take into consideration the dynamicity that exists in peer-to-peer networks in the process of super-peers selection. Several computer simulations have been conducted to compare the performance of the proposed super-peer selection algorithm with the performance of existing algorithms with respect to the number of super-peers, and capacity utilization. Simulation results have shown the superiority of the proposed super-peer selection algorithm over the existing algorithms.     

一种PID参数量子粒子群自整定方法

传统的PID参数整定方法由于需要决策者具有较强的工程经验,难以处理非连续、非线性或时滞的复杂系统。针对这种情况,提出一种新的基于量子粒子群优化的PID参数自整定方法。该算法采用问题的时间绝对偏差乘积积分方程来评价粒子的适应值;设计一种时变变异算子,用来均衡粒子的全局和局部开发能力。实验结果表明,该算法在超调量和调节时间等指标上皆优于传统粒子群优化算法。     

Unsupervised measures for parameter selection of binarization algorithms

In this paper, we propose a mechanism for systematic comparison of the efficacy of unsupervised evaluation methods for parameter selection of binarization algorithms in optical character recognition (OCR). We also analyze these measures statistically and ascertain whether a measure is suitable or not to assess a binarization method. The comparison process is streamlined in several steps. Given an unsupervised measure and a binarization algorithm we: (i) find the best parameter combination for the algorithm in terms of the measure, (ii) use the best binarization of an image on an OCR, and (iii) evaluate the accuracy of the characters detected. We also propose a new unsupervised measure and a statistical test to compare measures based on an intuitive triad of possible results: better, worse or comparable performance. The comparison method and statistical tests can be easily generalized for new measures, binarization algorithms and even other accuracy-driven tasks in image processing. Finally, we perform an extensive comparison of several well known measures, binarization algorithms and OCRs, and use it to show the strengths of the WV measure.     

A note on causal and CITE iterative learning control algorithms

The convergence properties of causal and current iteration tracking error (CITE) discrete time iterative learning control (ILC) algorithms are studied using time and frequency domain convergence criteria. Of particular interest are conditions for monotone convergence, and these are evaluated using a discrete-time version of Bode's integral theorem.     

Balancing exploration and exploitation in memetic algorithms: A learning automata approach

One of the problems with traditional genetic algorithms (GAs) is premature convergence, which makes them incapable of finding good solutions to the problem. The memetic algorithm (MA) is an extension of the GA. It uses a local search method to either accelerate the discovery of good solutions, for which evolution alone would take too long to discover, or reach solutions that would otherwise be unreachable by evolution or a local search method alone. In this paper, we introduce a new algorithm based on learning automata (LAs) and an MA, and we refer to it as LA‐MA. This algorithm is composed of 2 parts: a genetic section and a memetic section. Evolution is performed in the genetic section, and local search is performed in the memetic section. The basic idea of LA‐MA is to use LAs during the process of searching for solutions in order to create a balance between exploration performed by evolution and exploitation performed by local search. For this purpose, we present a criterion for the estimation of success of the local search at each generation. This criterion is used to calculate the probability of applying the local search to each chromosome. We show that in practice, the proposed probabilistic measure can be estimated reliably. On the basis of the relationship between the genetic section and the memetic section, 3 versions of LA‐MA are introduced. LLA‐MA behaves according to the Lamarckian learning model, BLA‐MA behaves according to the Baldwinian learning model, and HLA‐MA behaves according to both the Baldwinian and Lamarckian learning models. To evaluate the efficiency of these algorithms, they have been used to solve the graph isomorphism problem. The results of computer experimentations have shown that all the proposed algorithms outperform the existing algorithms in terms of quality of solution and rate of convergence.     

Analysis of adaptive step-size SA algorithms for parameter tracking

We present proofs and data for adaptive step-size algorithms for tracking time-varying parameters when recursive stochastic approximation type algorithms are used. A classical problem in adaptive control and communication theory concerns the tracking of the best fit of a given form when the statistics or the parameters change slowly. A major, and yet unresolved, problem has been the choice of the step sizes in the tracking algorithm. An algorithm for adapting the step size using the same system measurements which are used for the tracking was suggested by Benveniste and various examples worked out by Brossier. The numerical results were very encouraging. But proofs were lacking. These proofs are supplied here together with supporting numerical data. The proofs are based on recent results in stochastic approximation. The adaptive step-size technique works very well indeed. Much supporting analysis is presented, particularly concerning the interpretation of certain stationary processes as “stationary” pathwise derivatives. Finite difference forms are also treated. These are mathematically simpler and can be applied to a wide variety of systems, even when the system is not well modeled. The data shows that they work well     

A note on three-dimensional finite automata

This paper investigates the relationships between the accepting powers of three-dimensional six-way finite automata (3-FA's) and three-dimensional five-way Turing machines (5WTM's), where the input tapes of these automata are restricted to cubic ones. A 3-FA (5WTM) can be considered as a natural extension of the two-dimensional four-way finite automaton (two-dimensional three-way Turing machine) to three dimensions. The main results are: (1) n²logn (n³) space is necessary and sufficient for deterministic 5WTM's to simulate deterministic (nondeterministic) 3-FA's; (2) n² (n²) space is necessary and sufficient for nondeterministic 5WTM's to simulate deterministic (nondeterministic) 3-FA's.     

A note on the variance of rank-based selection strategies for genetic algorithms and genetic programming

This paper evaluates different forms of rank-based selection that are used with genetic algorithms and genetic programming. Many types of rank based selection have exactly the same expected value in terms of the sampling rate allocated to each member of the population. However, the variance associated with that sampling rate can vary depending on how selection is implemented. We examine two forms of tournament selection and compare these to linear rank-based selection using an explicit formula. Because selective pressure has a direct impact on population diversity, we also examine the interaction between selective pressure and different mutation strategies.