Using a Markov network as a surrogate fitness function in a genetic algorithm

Real life optimization problems often require finding optimal solution to complex high dimensional, multimodal problems involving computationally very expensive fitness function evaluations. Use of a population based iterative technique like evolutionary algorithms in such problem domains is thus practically prohibitive. An attractive alternative is to build surrogates or use an approximation of the actual fitness functions to be evaluated. Many regression and interpolation tools are available to build such meta models. This paper briefly discusses the architectures and use of such meta-modeling tools in an evolutionary optimization context. We further present an evolutionary algorithm framework which involves use of surrogate models for fitness function evaluation. The original framework namely, the dynamic approximate fitness based hybrid EA (DAFHEA) model [14] reduces computation time by controlled use of meta-models (in this case approximate model generated by support vector machine regression) to partially replace the actual function evaluation by approximate function evaluation. However, the underlying assumption in DAFHEA is that the training samples for the meta-model are generated from a single uniform model. This does not take into account uncertain scenarios involving noisy fitness functions. The enhanced model, DAFHEA-II, incorporates a multiple-model based learning approach for the support vector machine approximator to counter effects of noise [15]. Empirical results obtained by evaluating the frameworks using several benchmark functions (both non- noisy and noisy versions) are presented.

We propose a new approach for building recommender systems by adapting surrogate-assisted interactive genetic algorithms. A pool of user-evaluated items is used to construct an approximative model which serves as a surrogate fitness function in a genetic algorithm for optimizing new suggestions. The surrogate is used to recommend new items to the user, which are then evaluated according to the user's liking and subsequently removed from the search space. By updating the surrogate model after new recommendations have been evaluated by the user, we enable the model itself to evolve towards the user's preferences. In order to precisely evaluate the performance of that approach, the human's subjective evaluation is replaced by common continuous objective benchmark functions for evolutionary algorithms. The system's performance is compared to a conventional genetic algorithm and random search. We show that given a very limited amount of allowed evaluations on the true objective, our approach outperforms these baseline methods.

Temperature control is the key of Ruhrstahl-Heraeus (RH) process in steelmaking plant. The accuracy of RH control model greatly affects the molten steel temperature fluctuation. To obtain RH control model with higher accuracy, an improved case-based reasoning (CBR) model based on attribute weights optimized by genetic algorithm (GA) was proposed. The fitness function in GA was determined according to the prediction accuracy of end temperature of molten steel in RH; then, GA is used to optimize all the attribute weights based on known case base. An improved CBR model that contains the optimized attribute weights was applied to predict end temperature of molten steel in RH, and the prediction accuracy was calculated. Four methods, CBR based on attribute weights optimized by GA (CBR–GA), ordinary CBR, back propagation neural network (BPNN) and multiple linear regression (MLR) method were employed for comparison. The results show that in the error range of [− 3 °C, 3 °C], [− 5 °C, 5 °C], [− 7 °C, 7 °C] and [− 10 °C, 10 °C], the prediction accuracy of CBR–GA was improved by 19.99%, 28.19%, 27.11% and 16.3%, respectively, than that of MLR. Compared with BPNN, the prediction accuracy increased by 3.22%, 7.44%, 5.29% and 2.40%, respectively. Compared with ordinary CBR, the accuracy increased by 5.43%, 5.80%, 4.66% and 2.27%, respectively.

Location‐only and use‐availability data: analysis methods converge

Hybridization of multi-objective evolutionary algorithms and artificial neural networks for optimizing the performance of electrical drives

A byproduct benefit of using probabilistic model-building genetic algorithms is the creation of cheap and accurate surrogate models. Learning classifier systems -- and genetics-based machine learning in general -- can greatly benefit from such surrogates which may replace the costly matching procedure of a rule against large data sets. In this paper we investigate the accuracy of such surrogate fitness functions when coupled with the probabilistic models evolved by the $\chi$-ary extended compact classifier system ($\chi$eCCS). To achieve such a goal, we show the need that the probabilistic models should be able to represent all the accurate basis functions required for creating an accurate surrogate. We also introduce a procedure to transform populations of rules based into dependency structure matrices (DSMs) which allows building accurate models of overlapping building blocks -- a necessary condition to accurately estimate the fitness of the evolved rules.

Energy-efficient virtual machine placement in data centres via an accelerated Genetic Algorithm with improved fitness computation

The chapter covers two main areas, these being an introduction to the technology and techniques associated with genetic algorithms and then the second part looks at how genetic algorithms can be used to search for good topological solutions to engineering design challenges. The start of the chapter places genetic algorithms in context compared to other evolutionary algorithms and also describes the reasons why genetic algorithms are potentially useful. This is then followed by a look at the concept of a search space. Section two looks at the canonical genetic algorithm as a basic introduction to the technology and includes an examination of the main techniques used to encode the genome, fitness functions, operators and selection. Section three looks at how genetic algorithms can be used for design and chooses the specific example of the conceptual design of commercial office buildings. Section four introduces the basic concepts of topological search and explains how having the right form of representation is vital before looking at example relating to structural components and the design of domes using a genetic algorithm linked to computational geometry techniques. The final section then looks at further methods using generative representations and generative geometries as possible solutions to the need to develop powerful forms of representation for handling topological search in genetic algorithms.

SUMMARY The seismic waveform inversion problem is usually cast into the framework of Bayesian statistics in which prior information on the model parameters is combined with the data and physics of the forward problem to estimate the a posteriori probability density (PPD) in model space. The PPD is a function of an objective or fitness function computed from the observed and synthetic data. In general, the PPD or the fitness function is multimodal and its shape is unknown. Global optimization methods such as simulated annealing (SA) and genetic algorithms (GA) do not require that the shape of the fitness function be known. In this paper, we investigate GA to rapidly sample the most significant portion or portions of the PPD, when very little prior information is available. First, we use a simple three operator (selection, crossover and mutation) GA acting on a randomly chosen finite population of haploid binary coded models. We use plane wave transformed synthetic seismic data and a normalized cross-correlation function [E(m)] in the frequency domain as a fitness function. A moderate value of crossover probability, a low value of mutation probability, a high value of update probability and a proper population size are required to reach very close to the global maximum of the fitness function. Next, with an attempt to accelerate convergence we show that the concepts from simulated annealing can be used in stretching of the fitness function, i.e., we use exp [E(m)/T] rather than E(m) as the fitness function, where T is a control parameter analogous to temperature in simulated annealing. By a schemata analysis, we show that at low temperatures, schemata with above average fitness values are reproduced in large numbers causing a much more rapid convergence of the algorithm. A high value of temperature T assigns nearly equal selection probability to most of the schemata and thus retains diversity among the members of the population. Thus a GA with a step function type cooling schedule (very high temperature in the beginning followed by rapid cooling to a very low temperature) improves the performance dramatically: high values of the fitness function are obtained rapidly using only half as many models as would be required by a conventional GA. Similar performance could also be achieved by first using a high mutation probability and then decreasing the mutation probability to a very low value, while retaining the same low temperature throughout. We also address the problem of ‘genetic drift’ which causes the finite GAs to converge to one peak or the other when the algorithm is applied to a highly multimodal fitness function with several peaks of nearly the same height. A parallel genetic algorithm based on the concept of ‘punctuated equilibria’ is implemented to circumvent the problem. We run several GAs each with a finite subpopulation in parallel and collect many good models from each one of these runs. These are then used to grasp the most significant portion(s) of the PPD in model space. We then compute the weighted mean model and use the derived good models to estimate uncertainty in the derived model parameters.

Competitive coevolution is an important technique for fields such as defense and security which inherently involve adversarial games. One advantage that the field of computer security has in particular is that games in this space are often naturally able to be simulated at a high fidelity by interacting with the involved software or hardware directly. However, such high-fidelity evaluations are typically slow, so it is especially important in these cases to get as much useful information out of as few evaluations as possible. This paper proposes a new competitive coevolutionary evaluation method of Similar-Strength Opponent Sampling, which selects opponent pairings of similar skill levels so that evaluations can more efficiently distinguish the performances of similar individuals. This is enabled through the use of Elo ratings as a surrogate fitness function that prevents bias against individuals assigned stronger opponents. Care is taken to ensure that this technique is applicable to complex games where there is no explicit winner or loser, allowing ratings to be based on relative fitness. Mixed results are presented, showing that significant benefits are gained from pairing similar-strength opponents, but finding that the use of Elo rating instead of raw fitness harms evolution for intransitive games.

Proceeding of: IEEE Congress on Evolutionary Computation (CEC 2009), May 18-21 (Monday - Thursday), 2009, Trondheim, Norway.

The paper proposes a fast online learning method for neural network structures by using genetic algorithm (GA) and dynamic back propagation algorithm (BP) jointly. GA is used in the coarse tuning process which adjusts interconnection weights of the neural network. The dynamic back propagation algorithm is subsequently applied to achieve fine adjusting of the network weights. The fitness function, based on the squared error between the teaching signal and the network output value, is redefined at every time step and the proposed GA based algorithm solves a nonstationary function optimization task. At every time step the solution with the best fitness function is used for current representation of the neural network weights and biases. It is shown through the simulations and real time temperature control of drying oven that this learning algorithm has faster convergence ability and better performance on reducing mapping error in the online learning neural network structures. This leads to an improvement of the transient response of neuro adaptive systems. The proposed method has the potential to be applied to many practical areas such as system modeling and control, signal processing and pattern recognition.

In the present paper, optimization of functions with uncertainty by means of Genetic Algorithms (GA) is discussed. For such problems, there have been proposed methods of sampling fitness function several times and taking average of them for evaluation of each individual. However, important applications having uncertain fitness functions are online adaptation of real systems and complicated computer simulation using random variables. In such applications, possible number of fitness evaluation is quite limited. Hence, methods achieving optimization with less number of fitness evaluation is needed. In the present paper, the authors propose a GA for optimization of continuous fitness functions with observation noise utilizing history of search so as to reduce number of fitness evaluation. In the proposed method, value of fitness function at a novel search point is estimated not only by the sampled fitness value at that point but also by utilizing the fitness values of individuals stored in the history of search. Computer experiments using quadric fitness functions show that the proposed method outperforms the conventional GA of sampling fitness values several times at each search point in noisy environment.

Fisher’s iris data constitute a hard benchmark for clustering procedures, and attracted much work based on statistical methods and new approaches related to evolutionary algorithms and neural networks. We suggest two genetic algorithms effective for simultaneously determining both the optimal number of groups and the assignment of items to groups. The grouping genetic algorithm proposed by Falkenauer (1998) forms the basis of our method, where the variance ratio criterion and the Marriott’s method provide two fitness functions that both allow for fast computation and include the number of groups explicitly as a parameter. Specialized crossover operators, specific for each of the two fitness functions, are designed to accelerate convergence and minimize the number of iterations. Some simple implementations of our genetic algorithms are presented, that allow to classify correctly as many iris plants as the best alternative procedures proposed for this data set. Therefore genetic algorithms seem to constitute a good alternative choice for handling clustering problems.KeywordsClassificationCrossoverEncodingEvolutionary ComputationFitness FunctionMutation

Genetic Algorithm (GA) classifier can automatically search a proper clustering number according to fitness evaluation instead of assignment by users. In this study, a GA classifier with various fitness functions is adopted to search the cluster centers and a suitable cluster number for digital images to overcome the disadvantages of the conventional unsupervised classifier. By employing a proper clustering index as fitness, a GA with length-variable chromosome can determine the most suitable number of clusters and the most proper cluster centers.