采用半初始化和概率扰动策略改进的遗传算法

针对遗传算法在函数寻优过程中收敛速度慢、易陷入局部最优解的问题,提出一种采用半初始化和概率扰动策略改进的遗传算法DIAGA。首先,通过引入概率扰动策略增加了算法迭代后期的种群多样性,采用半初始化从根本上改变了算法在全局最优解比较过程中的局限性;然后利用马尔可夫链理论证明了DIAGA的收敛性;最后,对六个标准测试函数进行仿真测试。仿真实验结果表明,提出的DIAGA有效摆脱了局部收敛,在搜索精度、收敛速度上具有明显优势,就多维测试函数而言,寻优精度提高了约29%。     

Comparison of probabilistic and deterministic optimizations using genetic algorithms

This paper describes an application of genetic algorithms to deterministic and probabilistic (reliability-based) optimization of damping augmentation for a truss structure. The probabilistic formulation minimizes the probability of exceeding upper limits on the magnitude of the dynamic response of the structure due to uncertainties in the properties of the damping devices. The corresponding deterministic formulation maximizes a safety margin with respect to these limits. Because this work was done in the context of an experimental comparison of the reliabilities of the resulting designs, antioptimization was used to maximize the contrast between the probabilities of failure of the two designs. This contrast maximization was also performed with a genetic algorithm. The paper describes the genetic algorithm used for the optimization and antioptimization, and a number of modifications to the antioptimization formulation intended to reduce the computational expense to an acceptable level. Optimal designs are obtained for both formulations. The probabilistic design shows a very significant improvement in reliability.     

基于概率模型的遗传规划在故障诊断中的应用

采用在遗传规划中使用概率模型的新方法采解决一系列故障诊断问题。故障诊断可被看为是一个多级分类问题。遗传规划在解决复杂问题上有很大的优势,而这种优势在故障诊断中仍然显著。而且,使用概率模型作为适应函数能提高诊断的精确性,最后用这种方法解决机电设备的故障诊断。结果显示,使用基于概率模型的遗传规划解决机电设备的故障诊断比人工神经网络优越。     

On using deterministic functions to reduce randomness in probabilistic algorithms

We show the existence of nonuniform schemes for the following sampling problem: Given a sample space with n points, an unknown set of size n/2, and s random points, it is possible to generate deterministically from them s + k points such that the probability of not hitting the unknown set is exponentially smaller in k than 2^−s. Tight bounds are given for the quality of such schemes. Explicit, uniform versions of these schemes could be used for efficiently reducing the error probability of randomized algorithms. A survey of known constructions (whose quality is very far from the existential result) is included.     

Continuous probabilistic model building genetic network programming using reinforcement learning

Recently, a novel probabilistic model-building evolutionary algorithm (so called estimation of distribution algorithm, or EDA), named probabilistic model building genetic network programming (PMBGNP), has been proposed. PMBGNP uses graph structures for its individual representation, which shows higher expression ability than the classical EDAs. Hence, it extends EDAs to solve a range of problems, such as data mining and agent control. This paper is dedicated to propose a continuous version of PMBGNP for continuous optimization in agent control problems. Different from the other continuous EDAs, the proposed algorithm evolves the continuous variables by reinforcement learning (RL). We compare the performance with several state-of-the-art algorithms on a real mobile robot control problem. The results show that the proposed algorithm outperforms the others with statistically significant differences.     

Fitting optimal piecewise linear functions using genetic algorithms

Constructing a model for data in R² is a common problem in many scientific fields, including pattern recognition, computer vision, and applied mathematics. Often little is known about the process which generated the data or its statistical properties. For example, in fitting a piecewise linear model, the number of pieces, as well as the knot locations, may be unknown. Hence, the method used to build the statistical model should have few assumptions, yet, still provide a model that is optimal in some sense. Such methods can be designed through the use of genetic algorithms. We examine the use of genetic algorithms to fit piecewise linear functions to data in R². The number of pieces, the location of the knots, and the underlying distribution of the data are assumed to be unknown. We discuss existing methods which attempt to solve this problem and introduce a new method which employs genetic algorithms to optimize the number and location of the pieces. Experimental results are presented which demonstrate the performance of our method and compare it to the performance of several existing methods, We conclude that our method represents a valuable tool for fitting both robust and nonrobust piecewise linear functions     

Evolving strategy for a probabilistic game of imperfect information using genetic programming

We provide the complete record of methodology that let us evolve BrilliAnt, the winner of the Ant Wars contest. Ant Wars contestants are virtual ants collecting food on a grid board in the presence of a competing ant. BrilliAnt has been evolved through a competitive one-population coevolution using genetic programming and fitnessless selection. In this paper, we detail the evolutionary setup that lead to BrilliAnt’s emergence, assess its direct and indirect human-competitiveness, and describe the behavioral patterns observed in its strategy.

---

Some properties of certain classes of meromorphically p-valent functions involving extended multiplier transformations

In this paper, the authors investigate interesting properties of certain subclasses of meromorphically multivalent functions which are defined here by means of extended multiplier transformations.     

Unsatisfying functions and multiobjective fuzzy satisfaction design using genetic algorithms

This paper describes a new fuzzy satisfaction method using genetic algorithms (GA) for multiobjective problems. First, an unsatisfying function, which has a one-to-one correspondence with the membership function, is introduced for expressing "fuzziness". Next, the multiobjective design problem is transformed into a satisfaction problem of constraints by introducing an aspiration level for each objective. Here, in order to handle the fuzziness involved in aspiration levels and constraints, the unsatisfying function is used, and the problem is formulated as a multiobjective minimization problem of unsatisfaction ratings. Then, a GA is employed to solve the problem, and a new strategy is proposed to obtain a group of Pareto-optimal solutions in which the decision maker (DM) is interested. The DM can then seek a satisfaction solution by modifying parameters interactively according to preferences.     

Learning directed probabilistic logical models: ordering-search versus structure-search

We discuss how to learn non-recursive directed probabilistic logical models from relational data. This problem has been tackled before by upgrading the structure-search algorithm initially proposed for Bayesian networks. In this paper we show how to upgrade another algorithm for learning Bayesian networks, namely ordering-search. For Bayesian networks, ordering-search was found to work better than structure-search. It is non-obvious that these results carry over to the relational case, however, since there ordering-search needs to be implemented quite differently. Hence, we perform an experimental comparison of these upgraded algorithms on four relational domains. We conclude that also in the relational case ordering-search is competitive with structure-search in terms of quality of the learned models, while ordering-search is significantly faster.     

Guaranteed cost regulator design: A probabilistic solution and a randomized algorithm

This paper presents a gradient-based randomized algorithm to design a guaranteed cost regulator for a plant with general parametric uncertainties. The algorithm either provides with high confidence a probabilistic solution that satisfies the design specification with high probability for a randomly sampled uncertainty or claims that the feasible set of the design parameters is too small to contain a ball with a given radius. In both cases, the number of iterations executed in the algorithm is of polynomial order of the problem size and is independent of the dimension of the uncertainty.     

Learning from partially supervised data using mixture models and belief functions

This paper addresses classification problems in which the class membership of training data are only partially known. Each learning sample is assumed to consist of a feature vector x_i∈X and an imprecise and/or uncertain “soft” label m_i defined as a Dempster-Shafer basic belief assignment over the set of classes. This framework thus generalizes many kinds of learning problems including supervised, unsupervised and semi-supervised learning. Here, it is assumed that the feature vectors are generated from a mixture model. Using the generalized Bayesian theorem, an extension of Bayes’ theorem in the belief function framework, we derive a criterion generalizing the likelihood function. A variant of the expectation maximization (EM) algorithm, dedicated to the optimization of this criterion is proposed, allowing us to compute estimates of model parameters. Experimental results demonstrate the ability of this approach to exploit partial information about class labels.     

Modelling air pollution time-series by using wavelet functions and genetic algorithms

This paper proposes a novel approach based on the use of wavelet functions to model air pollution time series. One peculiarity of the approach is that of combining the use of wavelets and genetic algorithms to search for the best wavelet parameters. A case study, referring to the modelling of daily averages of SO₂ time series recorded in the industrial area of Syracuse (Italy) is reported in order to compare the performance of a wavelet-based prediction model and a Multi-layer perceptron neural model. The results obtained show that there are no significant differences between the neural and the wavelet approach in terms of model performance and computational effort. There is however an appreciable advantage in using the proposed wavelet-based technique in terms of model readability.The paper has been financially supported by the EU in the framework of the APPETISE project (Contract N. IST-99–11764). The author is also grateful to the Municipal and Provincial authorities in Syracuse (Italy) for providing the pollution and meteorological data considered in the paper. Finally the author is grateful to Dr Libero Bertucco who helped to code part of the software package considered in this work.     

Aggregation functions with given super-additive and sub-additive transformations

Aggregation functions and their transformations have found numerous applications in various kinds of systems as well as in economics and social science. Every aggregation function is known to be bounded above and below by its super-additive and sub-additive transformations. We are interested in the “inverse” problem of whether or not every pair consisting of a super-additive function dominating a sub-additive function comes from some aggregation function in the above sense. Our main results provide a negative answer under mild extra conditions on the super- and sub-additive pair. We also show that our results are, in a sense, best possible.     

Comparing networks with differing neural-node functions using transputer-based genetic algorithms

Different neural net node computational functions are compared using feedforward backpropagation networks. Three node types are examined: the standard model, ellipsoidal nodes and quadratic nodes. After preliminary experiments on simple small problems, in which quadratic nodes performed very well, networks of differing nodes types are applied to the speech recognition 104 speaker E-task using a fixed architecture. Ellipsoidal nodes were found to work well, but not as well as the standard model. Quadratic nodes did not perform well on the larger task. To facilitate an architecture independent comparison a transputer-based genetic algorithm is then used to compare ellipsoidal and mixed ellipsoidal and standard networks with the standard model. These experiments confirmed the earlier conclusion that ellipsoidal networks could not compare favourably with the standard model on the 104 speaker E-task. In an evolutionary search in which layer node types were free to adjust ellipsoidal nodes had a tendency to become extinct or barely survive. If the presence of ellipsoidal nodes was enforced then the resulting networks again performed poorly when compared with the standard model.     

Learning and re-using information in space layout planning problems using genetic engineering

We describe the use of a genetic engineering version of genetic algorithms as a natural tool for gathering and re-using information about some classes of design problems. This information is stored in the form of sets of ‘evolved genes’ which are linked to beneficial qualities of the ‘good’ designs within this class of problems. The approach is illustrated on a space layout planning problem.     

Learning to hash: forgiving hash functions and applications

The problem of efficiently finding similar items in a large corpus of high-dimensional data points arises in many real-world tasks, such as music, image, and video retrieval. Beyond the scaling difficulties that arise with lookups in large data sets, the complexity in these domains is exacerbated by an imprecise definition of similarity. In this paper, we describe a method to learn a similarity function from only weakly labeled positive examples. Once learned, this similarity function is used as the basis of a hash function to severely constrain the number of points considered for each lookup. Tested on a large real-world audio dataset, only a tiny fraction of the points (~0.27%) are ever considered for each lookup. To increase efficiency, no comparisons in the original high-dimensional space of points are required. The performance far surpasses, in terms of both efficiency and accuracy, a state-of-the-art Locality-Sensitive-Hashing-based (LSH) technique for the same problem and data set.     

Simultaneous design of membership functions and rule sets for fuzzycontrollers using genetic algorithms

This paper examines the applicability of genetic algorithms (GA's) in the simultaneous design of membership functions and rule sets for fuzzy logic controllers. Previous work using genetic algorithms has focused on the development of rule sets or high performance membership functions; however, the interdependence between these two components suggests a simultaneous design procedure would be a more appropriate methodology. When GA's have been used to develop both, it has been done serially, e.g., design the membership functions and then use them in the design of the rule set. This, however, means that the membership functions were optimized for the initial rule set and not the rule set designed subsequently. GA's are fully capable of creating complete fuzzy controllers given the equations of motion of the system, eliminating the need for human input in the design loop. This new method has been applied to two problems, a cart controller and a truck controller. Beyond the development of these controllers, we also examine the design of a robust controller for the cart problem and its ability to overcome faulty rules