Search

I use a theorem from machine learning, called the “No Free Lunch” theorem (NFL) to support the claim that non-epistemic values are essential to theory choice. I argue that NFL entails that predictive accuracy is insufficient to favor a given theory over others, and that NFL challenges our ability to give a purely epistemic justification for using other traditional epistemic virtues in theory choice. In addition, I argue that the natural way to overcome NFL’s challenge is to use non-epistemic values. If my argument holds, non-epistemic values are entangled in theory choice regardless of human limitations and regardless of the subject matter. Thereby, my argument overcomes objections to the main lines of argument revealing the role of values in theory choice. At the end of the paper, I argue that, contrary to common conception, the epistemic challenge arising from NFL is distinct from Hume’s problem of induction and other forms of underdetermination.

As revealed by the no free lunch theorem, no single algorithm can outperform any others on all classes of optimization problems. To tackle this issue, methods for recommending an existing algorithm for solving given problems have been proposed. However, existing recommendation methods for continuous optimization suffer from low practicability and transferability, mainly due to the difficulty in extracting features that can effectively describe the problem structure and lack of data for training a recommendation model. This work proposes a generic recommender system to address the above two challenges. First, a novel method is proposed to represent an analytic objective function of a continuous optimization problem as a tree, which is directly used as the features of the problem. For black-box optimization problems whose objective function is unknown, a symbolic regressor is adopted to estimate the tree structure. Second, a large number of benchmark problems are randomly created based on the proposed tree representation, providing an abundant amount of training data with various levels of difficulty. By employing a deep recurrent neural network, a recommendation model is trained to recommend a most suitable metaheuristic algorithm for white- or black-box optimization, making a significant step forward towards fully automated algorithm recommendation for continuous optimization. Experimental results on 100 000 benchmark problems show that the proposed recommendation model achieves considerably better performance than existing ones, and exhibits high transferability to real-world problems.

The most critical concern in machine learning is how to make an algorithm that performs well both on training data and new data. No free lunch theorem implies that each specific task needs its own tailored machine learning algorithm to be designed. A set of strategies and preferences are built into learning machines to tune them for the problem at hand. These strategies and preferences, with the core concern of generalization improvement, are collectively known as regularization. In deep learning, because of a considerable number of parameters, a great many forms of regularization methods are available to the deep learning community. Developing more effective regularization strategies has been the subject of significant research efforts in recent years. However, it is difficult for developers to choose the most suitable strategy for their problem at hand, because there is no comparative study regarding the performance of different strategies. In this paper, at the first step, the most effective regularization methods and their variants are presented and analyzed in a systematic approach. At the second step, comparative research on regularization techniques is presented in which the testing errors and computational costs are evaluated in a convolutional neural network, using CIFAR-10 ( https://www.cs.toronto.edu/~kriz/cifar.html ) dataset. In the end, different regularization methods are compared in terms of accuracy of the network, the number of epochs for the network to be trained and the number of operations per input sample. Also, the results are discussed and interpreted based on the employed strategy. The experiment results showed that weight decay and data augmentation regularizations have little computational side effects so can be used in most applications. In the case of enough computational resources, Dropout family methods are rational to be used. Moreover, in the case of abundant computational resources, batch normalization family and ensemble methods are reasonable strategies to be employed.

For more than decades, the central drivers of business growth have been technological tools. The most important of these is Artificial Intelligence (AI). We name the paradigm “Business Artificial Intelligence (BAI)” the category of AI approaches that can create profitable new business models, earnest business-values, and more competitiveness, particularly Machine Learning (ML). However, ML algorithms are plentiful, and most of them necessitate specific Data structures, particular applicability features, and in-depth analysis of the business context. Because of these considerations, an increasing number of industrial ML-based applications fail. Therefore, it is natural to ask the question of applicability limits and ML algorithms selection, giving the business context. Accordingly, in this paper, we propose the evidence of the “No-Free-Lunch” (NFL) theorems to understand ML use’s applicability in business organizations.

Machine Learning practice in general offers significant opportunities for parallel computing and practicing sound software engineering. More often than not, practitioners routinely write dataset specific scripts and learners focus on model building and refining. Focusing on particular models is not consistent with NFL, a fundamental theorem in Machine Learning. Not minding time-honored software engineering principles is inefficient. In this paper, we present our implementation of MISD machine, consistent with No Free Lunch Theorem, problems we encountered and our approach to solve those problems.

Machine learning and artificial intelligence have evolved beyond simple hype and have integrated themselves in business and in popular conversation as an increasing number of smart applications profoundly transform the way we work and live. This article defines machine learning in terms of potential benefits and pitfalls for a nontechnical audience, and gives examples of popular and powerful machine learning algorithms: k-means clustering, principal component analysis, and artificial neural networks. Three important philosophical challenges of machine learning are introduced: the no free lunch theorem, the curse of dimensionality, and the bias–variance trade-off.

Feature selection has become an indispensable preprocessing step in data mining problems as high amount of data become prevalent with the advances in technology. The objective of feature selection is twofold: reducing data amount and improving learning performance. In this study, we leverage the multi-core nature of a regular PC to build a robust framework for feature selection. This framework executes the feature selection algorithm on four processors, in parallel. As per the No Free Lunch Theorem, we facilitate 40 different execution settings for the processors by employing two multiobjective selection algorithms, four initial population generation methods, and five machine learning techniques. Besides, we introduce six setting selection schemes to decide the most fruitful setting for each processor. We carry out extensive experiments on 11 UCI benchmark datasets and analyze the results with statistical tests. Finally, we compare our proposed method with state-of-the-art studies and record remarkable improvement in terms of maximum accuracy.

Several reasons such as no free lunch theorem indicates that any learning algorithm in combination with a specific feature selection (FS) technique may give more accurate estimation than other learning algorithms. Therefore, there is not a universal approach that outperforms other algorithms. Moreover, due to the large number of FS techniques, some recommended solutions such as using synthetic dataset or combining different FS techniques are very tedious and time consuming. In this study to tackle the issue of more accurate estimation of NPPs parameters, the performance of the major supervised learning algorithms in combination with the different FS techniques which are appropriate for parameters estimation is considered. The target parameters/transients of the Bushehr nuclear power plant (BNPP) are examined as the case study. By comparing three major supervised learning algorithms (i.e. the MLP-BR, the MLP-LM, and the SVM) in combination with six principal FS techniques (i.e. the NCA, the F-test, the Kendall’s tau, the Pearson, the Spearman, and the Relief) for estimation of three important parameters of NPP (i.e. FMT, CMT, and the DNBR), the BR learning algorithm gives the more accurate results. Therefore, the results show that if the number of FS techniques is m and the number of learning algorithms is n, the search space for more accurate estimation of the NPPs important parameters can be reduced from n × m to 1 × m.

The no free lunch theorem is a radicalized version of Hume’s induction skepticism. It asserts that relative to a uniform probability distribution over all possible worlds, all computable prediction algorithms—whether ‘clever’ inductive or ‘stupid’ guessing methods (etc.)—have the same expected predictive success. This theorem seems to be in conflict with results about meta-induction. According to these results, certain meta-inductive prediction strategies may dominate other (non-meta-inductive) methods in their predictive success (in the long run). In this article this conflict is analyzed and dissolved, by means of probabilistic analysis and computer simulation.

Construction heuristics play an important role in solving combinatorial optimization problems. These heuristics are usually used to create an initial solution to the problem which is improved using optimization techniques such as metaheuristics. For examination timetabling and university course timetabling problems essentially graph colouring heuristics have been used for this purpose. The process of deriving heuristics manually for educational timetabling is a time consuming task. Furthermore, according to the no free lunch theorem different heuristics will perform well for different problems and problem instances. Hence, automating the induction of construction heuristics will reduce the man hours involved in creating such heuristics, allow for the derivation of problem specific heuristics and possibly result in the derivation of heuristics that humans have not thought of. This paper presents generation construction hyper-heuristics for educational timetabling. The study investigates the automatic induction of two types of construction heuristics, namely, arithmetic heuristics and hierarchical heuristics. Genetic programming is used to evolve arithmetic heuristics. Genetic programming, genetic algorithms and the generation of random heuristic combinations is examined for the generation of hierarchical heuristics. The hyper-heuristics generating both types of heuristics are applied to the examination timetabling and the curriculum based university course timetabling problems. The evolved heuristics were found to perform much better than the existing graph colouring heuristics used for this domain. Furthermore, it was found that the while the arithmetic heuristics were more effective for the examination timetabling problem, the hierarchical heuristics produced better results than the arithmetic heuristics for the curriculum based course timetabling problem. Genetic algorithms proved to be the most effective at inducing hierarchical heuristics.

This chapter is a treatise on various fundamental principles that lie at the heart of machine learning theory and are quite often used in practice implicitly or explicitly. A new comer to this area of machine learning will find this chapter of immense value and will suddenly feel lot more confident in understanding the multitude of concepts that are built on top of these principles. I would strongly advise to not skip this chapter. All these concepts will be visited in the rest of the book multiple times as well as in many other books/articles the reader will encounter in the field.

The extraction of Bag of Visual Words (BoVW) features from retinal images for automated classification has been shown to be effective but computationally expensive. Histogram and co-variance matrix features do not generally result in models that have the same predictive accuracy as BoVW and are still computationally expensive. The discovery of features that result in accurate image classification on computationally constrained devices such as smartphones would enable new and promising applications for image classification. For example, smartphone retinal cameras can conceivably make diabetic retinopathy widely available and potentially reduce undiagnosed retinopathy if it could be achieved with computationally simple classification algorithms. A novel image feature extraction technique inspired by N-grams in text mining, called 'Pixel N-grams' is described that can serve this purpose. Results on mammogram and texture classification have shown high accuracy despite the reduced computational complexity. However retinal scan classification results using Pixel N-grams lag behind BoVW approaches. An explanation for the relative poor performance of Pixel N-grams with diabetic retinopathy that draws on concepts associated with the No Free Lunch theorem are presented.

Forecasting of stock market returns is a challenging research activity that is now expanding with the availability of new data sources, markets, financial instruments, and algorithms. At its core, the predictability of prices still raises important questions. Here, we discuss the economic significance of the prediction accuracy. To develop this question, we collect the daily series prices of almost half of the publicly traded companies around the world over a period of ten years and formulate some trading strategies based on their prediction. Proper visualization of these data together with the use of the No Free Lunch theoretical framework gives some unexpected results that show how the a priori less accurate algorithms and inefficient strategies can offer better results than the a priori best alternatives in some particular subsets of data that have a clear interpretation in terms of economic sectors and regions. This work was supported in part by the Universidad Carlos III de Madrid under Strategic Action 2013/00199/002.

There is a wide variety of studies that propose different classifiers to solve a large amount of problems in distinct classification scenarios. The no free lunch theorem states that if we use a big enough set of varied problems, all classifiers would be equivalent in performance. From another point of view, the performance of the classifiers is dependant of the scope and properties of the datasets. In this sense, new proposals on the topic often focus on a given context, aiming at improving the related state-of-the-art approaches. Data complexity metrics have been traditionally used to determine the inner characteristics of datasets. This way, researchers are able to categorize the problems in different scenarios. Then, this taxonomy can be applied to determine inner characteristics of the datasets in order to determine intervals of good and bad behavior for a given classifier. In this paper, we will take advantage of the data complexity metrics in order to design a fuzzy metaclassifier. The final goal is to create decision rules based on the inner characteristics of the data to apply a different version of the fuzzy classifier for a given problem. To do so, we will make use of the FARC-HD classifier, an evolutionary fuzzy system that has led to different extensions in the specialized literature. Experimental results show the goodness of this novel approach as it is able to outperform all versions of FARC-HD on a wide set of problems, and obtain competitive results (in terms of performance and interpretability) versus two selected state-of-the-art rule-based classification system, C4.5 and FURIA.

Quantum artificial intelligence exploits the interplay between artificial intelligence and quantum physics: on the one hand, a plethora of tools and ideas from artificial intelligence can be adopted to tackle intricate quantum problems; on the other hand, quantum computing could also bring unprecedented opportunities to enhance, speed up, or innovate artificial intelligence. Yet, quantum learning systems, similar to classical ones, may also suffer adversarial attacks: adding a tiny carefully-crafted perturbation to the legitimate input data would cause the systems to make incorrect predictions at a notably high confidence level. In this paper, we introduce the basic concepts and ideas of classical and quantum adversarial learning, as well as some recent advances along this line. First, we introduce the basics of both classical and quantum adversarial learning. Through concrete examples, involving classifications of phases of two-dimensional Ising model and three-dimensional chiral topological insulators, we reveal the vulnerability of classical machine learning phases of matter. In addition, we demonstrate the vulnerability of quantum classifiers with the example of classifying hand-written digit images. We theoretically elucidate the celebrated no free lunch theorem from the classical and quantum perspectives, and discuss the universality properties of adversarial attacks in quantum classifiers. Finally, we discuss the possible defense strategies. The study of adversarial learning in quantum artificial intelligence uncovers notable potential risks for quantum intelligence systems, which would have far-reaching consequences for the future interactions between the two areas.

Swarm intelligent algorithms focus on imitating the collective intelligence of a group of simple agents that can work together as a unit. Such algorithms have particularly significant impact in the fields like optimization and artificial intelligence (AI). This research article focus on a recently proposed swarm-based metaheuristic called the Artificial Bee Colony (ABC) algorithm and suggests modification to the algorithmic framework in order to enhance its performance. The proposed ABC variant shall be referred to as Migratory Multi-swarm Artificial Bee Colony (MiMSABC) algorithm. Different perturbation schemes of ABC function differently in varying landscapes. Hence to maintain the basic essence of all these schemes, MiMSABC deploys a multiple swarm populations that are characterized by different and unique perturbation strategies. The concept of reinitializing foragers around a depleted food source using a limiting parameter, as often used conventionally in ABC algorithms, has been avoided. Instead a performance based set of criteria has been introduced to thoroughly detect subpopulations that have shown limited progress to eke out the global optimum. Once failure is detected in a subpopulation provisions have been made so that constituent foragers can migrate to a better performing subpopulation, maintaining, however, a minimum number of members for successful functioning of a subpopulation. To evaluate the performance of the algorithm, we have conducted comparative study involving 8 algorithms for testing the problems on 25 benchmark functions set proposed in the Special Session on IEEE Congress on Evolutionary Competition 2005. Thorough a detailed analysis we have highlighted the statistical superiority of our proposed MiMSABC approach over a set of population based metaheuristics.

Economic dispatch (ED) is an important task in power system operation. It is able to decrease the operating cost, save energy resources, and reduce environmental load. In this paper, a multi-strategy ensemble biogeography-based optimization (MsEBBO) based method for ED problems is proposed. BBO is a population-based meta-heuristic algorithm inspired by the science of biogeography and mainly consists of three components: migration model, migration operator, and mutation operator. It has good local exploitation ability but lacks satisfactory global exploration ability. To keep a proper balance between exploration and exploitation, MsEBBO has three extensions to BBO’s three components according to the no free lunch theorem. First, a nonlinear migration model based on sinusoidal curve is employed. Second, a backup migration operator through adopting a backup strategy to combine perturb operator and blended operator is presented. This operator can make the entire population fully exchange or share information and thus further strengthen the exploitation ability. Finally, both differential mutation and Levy local search are embedded as mutation operator for MsEBBO using a similar backup strategy. Gaining from this mutation operator, MsEBBO can be accelerated to escape from local optima and perform efficient search within global range. Additionally, an effective repair technique is proposed to handle different constraints of ED problems. The performance of MsEBBO is tested on four ED problems with diverse complexities. Experimental results and comparisons with other recently reported ED solution methods confirm that MsEBBO is capable of yielding a good balance between exploration and exploitation, and obtaining competitive solution quality. Moreover, the sensitivity of MsEBBO to variations in population size is investigated as well.

Many software engineering problems have been addressed with search algorithms. Search algorithms usually depend on several parameters (e.g., population size and crossover rate in genetic algorithms), and the choice of these parameters can have an impact on the performance of the algorithm. It has been formally proven in the No Free Lunch theorem that it is impossible to tune a search algorithm such that it will have optimal settings for all possible problems. So, how to properly set the parameters of a search algorithm for a given software engineering problem? In this paper, we carry out the largest empirical analysis so far on parameter tuning in search-based software engineering. More than one million experiments were carried out and statistically analyzed in the context of test data generation for object-oriented software using the EvoSuite tool. Results show that tuning does indeed have impact on the performance of a search algorithm. But, at least in the context of test data generation, it does not seem easy to find good settings that significantly outperform the “default” values suggested in the literature. This has very practical value for both researchers (e.g., when different techniques are compared) and practitioners. Using “default” values is a reasonable and justified choice, whereas parameter tuning is a long and expensive process that might or might not pay off in the end.

According to no free lunch theorem, a single search technique cannot perform best under different conditions. The integration of principally similar search techniques is one of the option, that has been explored to effectively investigate the search area. In order to avoid stagnation at local minima and to enhance the search capability of particle swarm optimization, this paper proposes a technique that integrates predator prey optimization and anti-predatory particle swarm optimization. The integrated algorithm utilizes the capability of particle swarm optimization to find high quality solution, the capability of anti predatory particle swarm optimization to avoid search in worst solution regions and capability of predator prey optimization to escape from local minima under the effect of predator. The proposed algorithm has been implemented to solve economic dispatch problem. The different practical constraints such as ramp rate limits, prohibited operating zone(s) along with power balance constraint and generator limit are undertaken. The multiple fuel system is considered with transmission losses. The proposed algorithm is tested on various models of ED problems and compared with results reported in literature and found satisfactory.

Bus terminal assignment with the objective of maximizing public transportation service is known as bus terminal location problem (BTLP). We formulate the BTLP, a problem of concern in transportation industry, as a p-uncapacitated facility location problem (p-UFLP) with distance constraint. The p-UFLP being NP-hard (Krarup and Pruzan, 1990), we propose evolutionary algorithms for its solution. According to the No Free Lunch theorem and the good efficiency of the distinctive preserve recombination (DPX) operator, we design a new recombination operator for solving a BTLP by new evolutionary and memetic algorithms namely, genetic local search algorithms (GLS). We also define the potential objective function (POF) for the nodes and design a new mutation operator based on POF. To make the memetic algorithm faster, we estimate the variation of the objective function based on POF in the local search as part of an operator in memetic algorithms. Finally, we explore numerically the performance of nine proposed algorithms on over a thousand randomly generated problems and select the best two algorithms for further testing. The comparative studies show that our new hybrid algorithm composing the evolutionary algorithm with the GLS outperforms the multistart simulated annealing algorithm.

A sizable amount of research has been done to improve the mechanisms for knowledge extraction such as machine learning classification or regression. Quite unintuitively, the no free lunch (NFL) theorem states that all optimization problem strategies perform equally well when averaged over all possible problems. This fact seems to clash with the effort put forth toward better algorithms. This letter explores empirically the effect of the NFL theorem on some popular machine learning classification techniques over real-world data sets.

A recent direction of hyper-heuristics is the automated design of intelligent systems with the aim of reducing the man hours needed to implement such systems. One of the design decisions that often has to be made when developing intelligent systems is the low-level construction heuristic to use. These are usually rules of thumb derived based on human intuition. Generally a heuristic is derived for a particular domain. However, according to the no free lunch theorem different low-level heuristics will be effective for different problem instances. Deriving low-level heuristics for problem instances will be time consuming and hence we examine the automatic induction of low-level heuristics using hyper-heuristics. We investigate this for classical artificial intelligence. At the inception of the field of artificial intelligence search methods to solve problems were generally uninformed, such as the depth first and breadth first searches, and did not take any domain specific knowledge into consideration. As the field matured domain specific knowledge in the form of heuristics were used to guide the search, thereby reducing the search space. Search methods using heuristics to guide the search became known as informed searches, such as the best-first search, hill-climbing and the A* algorithm. Heuristics used by these searches are problem specific rules of thumb created by humans. This study investigates the use of a generative hyper-heuristic to derive these heuristics. The hyper-heuristic employs genetic programming to evolve the heuristics. The approach was tested on two classical artificial intelligence problems, namely, the 8-puzzle problem and Towers of Hanoi. The genetic programming system was able to evolve heuristics that produced solutions for 20 8-puzzle problems and 5 instances of Towers of Hanoi. Furthermore, the heuristics induced were able to produce solutions to the instances of the 8-puzzle problem which could not be solved using the A* algorithm with the number of tiles out of place heuristic and at least one admissible heuristic was evolved for all 25 problems.

Representation is a central issue in evolutionary computation. The no free lunch theorem demonstrates that there is no intrinsic advantage in a particular algorithm when considered against complete spaces of problems. The corollary is that algorithms should be fitted to the problems they are solving. Choice of representation is the primary point in the design of an evolutionary algorithm where the designer can incorporate domain knowledge and, in effect, choose the adaptive landscape he is searching. This tutorial will introduces representations for level design, agents for playing mathematical games, and for the design of non-player characters. Time permitting, other examples of representation will be included.

It is our great pleasure to welcome you to the GECCO Workshop on Evolutionary Computation Software Systems (EvoSoft). The evolution of so many different metaheuristic optimization algorithms results from the fact that no single method can outperform all others for all possible problems. As postulated in the No Free Lunch Theorem, a general-purpose and universal optimization strategy is impossible. The only way how one strategy can outperform another is to be more specialized to the structure of the tackled problem. Consequently it always takes qualified algorithm experts to select and tune a metaheuristic algorithm for a specific application. Choosing an appropriate method for a certain problem is not trivial, as problem characteristics may change remarkably for different instances and the performance of a metaheuristic may vary considerably for different parameter settings. Therefore the choice of a well suited method and according parameter values is a crucial aspect when applying metaheuristics. Because of these issues it is not advisable to implement only one specific metaheuristic algorithm for a specific problem, as it cannot be determined in advance whether the chosen method is suitable for the tackled problem or not. Thus soundly engineered, reusable, flexible, user-friendly and interoperable software systems are required which offer a broad spectrum of different metaheuristic algorithms and support researchers in adapting and comparing these methods for different problems in order to bridge the gap between theoretical research and practical applications. However, the development of such systems is both, time consuming and complex. Surprisingly the chance to join forces in the development of evolutionary computation software systems is not yet seized sufficiently within the evolutionary computation research community. Instead of reusing existing software and continuing their work on the shoulders of others, many researchers implement individual and highly specialized applications. Often these systems are not even publically released which hinders comparability and replicability of research results. The EvoSoft Workshop should highlight the importance of software engineering and software quality in the domain of evolutionary computation and should enable researchers to exchange their ideas on how to develop and apply generic and reusable software systems. It should provide a common platform to present open and freely available frameworks, to discuss novel approaches in their development, to identify cooperation potentials and synergies between research groups, and to define common standards.

No model is perfect and evaluation can be especially challenging in the operational public safety and security environment. Overall model accuracy rarely is the best measure, particularly with infrequent or rare events. Therefore, the nature, cost, and consequences of errors should be considered, especially within the context of the operational requirements and objectives. The use of training and test samples, as well as the confusion matrix to evaluate public safety and security models is discussed.

This paper introduces a new method for learning algorithm evaluation and selection, with empirical results based on classification. The empirical study has been conducted among 8 algorithms/classifiers with 100 different classification problems. We evaluate the algorithms' performance in terms of a variety of accuracy and complexity measures. Consistent with the No Free Lunch theorem, we do not expect to identify the single algorithm that performs best on all datasets. Rather, we aim to determine the characteristics of datasets that lend themselves to superior modelling by certain learning algorithms. Our empirical results are used to generate rules, using the rule-based learning algorithm C5.0, to describe which types of algorithms are suited to solving which types of classification problems. Most of the rules are generated with a high confidence rating.

The ‘no free lunch theorem’ claims that for the set of all problems no algorithm performs better than random search and, thus, selection can be advantageous only on a limited set of problems. In this paper we investigate how the topological structure of the environment influences algorithmic efficiency. We study the performance of algorithms, using selective learning, reinforcement learning, and their combinations, in random, scale-free, and scale-free small world (SFSW) environments. The learning problem is to search for novel, not-yet-found information. We ran our experiments on a large news site and on its downloaded portion. Controlled experiments were performed on this downloaded portion: we modified the topology, but preserved the publication time of the news. Our empirical results show that the selective learning is the most efficient in SFSW topology. In non-small world topologies, however, the combination of the selective and reinforcement learning algorithms performs the best.

PurposeThis paper addresses issues dealing with genetic algorithm (GA) convergence and the implications of the No Free Lunch Theorem which states that no single algorithm outperforms all others for all possible problem landscapes. In view of this, the authors propose that it is necessary for a GA to have the ability to classify the problem landscape before effective parameter adaptation may occur.Design/methodology/approachThe new hybrid intelligent system for landscape classification is proposed. This system facilitates intelligent operator selection and parameter tuning during run time in order to achieve maximum convergence. This work introduces two adaptive crossover techniques, the runtime adaptation of crossover probability and the participation level of multiple crossover operators in order to refine the quality of the search and to regulate the trade‐off between local and global search respectively. In addition, a Rule‐Based reasoning system (RS) is presented which can be utilised to analyse the problem landscape and provide a supervisory element to a GA. This RS is capable of instigating change by utilising the analysis in order to counteract premature convergence, for various classes of problems.FindingsResults are presented which show that the application of this Rule‐Based system and the adaptive crossover techniques proposed in this paper significantly improve performance for a suite of relatively complex test problems.Originality/valueThis work demonstrates the effectiveness of landscape classification and consequent rule‐based reasoning for GAs, particularly for problems with a difficult path to the optimal. Moreover, both adaptive crossover techniques proposed present improved performance over the traditional static parameter GA.

The problem of knowledge acquisition in animals is considered from the point of view of cybernetics. We show that all typesof animal behavior can be consistently explained on the basis of innate behavior programs and the creation of new behaviorprograms is logically inconsistent. The hypothesis that all animal behavior is completely innate is proposed. As a possiblephysical implementation of the storage of congenital programs, we considered quantum entanglement of biologically importantmolecules. Experiments to test hypotheses are proposed.

Machine learning is a vast area of research that is primarily concerned with finding patterns in empirical data. We restrict our attention to a limited number of core concepts that are most relevant for quantum learning algorithms. We discuss the importance of the data-driven approach, compared with the formal modeling of traditional artificial intelligence. We outline how to build high-dimensional feature spaces that will correspond to quantum states in quantum computing. An algorithm's ability to generalize beyond training data is intrinsically linked with its complexity—this is the foundation of the Vapnik-Chervonenkis theory, which we briefly explain. There is no free lunch, however: a single algorithm will not work in every learning scenario, which leads to combinations of learners and ensembles. Dependence on the dimensions and volume of the data defines computational complexity for learning methods, which often do not scale well.

It is our great pleasure to welcome you to the GECCO Workshop on Evolutionary Computation Software Systems (EvoSoft). The evolution of so many different metaheuristic optimization algorithms results from the fact that no single method can outperform all others for all possible problems. As postulated in the No Free Lunch Theorem, a general-purpose and universal optimization strategy is impossible. The only way how one strategy can outperform another is to be more specialized to the structure of the tackled problem. Consequently it always takes qualified algorithm experts to select and tune a metaheuristic algorithm for a specific application. Choosing an appropriate method for a certain problem is not trivial, as problem characteristics may change remarkably for different instances and the performance of a metaheuristic may vary considerably for different parameter settings. Therefore the choice of a well suited method and according parameter values is a crucial aspect when applying metaheuristics. Because of these issues it is not advisable to implement only one specific metaheuristic algorithm for a specific problem, as it cannot be determined in advance whether the chosen method is suitable for the tackled problem or not. Thus soundly engineered, reusable, flexible, user-friendly and interoperable software systems are required which offer a broad spectrum of different metaheuristic algorithms and support researchers in adapting and comparing these methods for different problems in order to bridge the gap between theoretical research and practical applications. However, the development of such systems is both, time consuming and complex. Surprisingly the chance to join forces in the development of evolutionary computation software systems is not yet seized sufficiently within the evolutionary computation research community. Instead of reusing existing software and continuing their work on the shoulders of others, many researchers implement individual and highly specialized applications. Often these systems are not even publically released which hinders comparability and replicability of research results. The EvoSoft Workshop should highlight the importance of software engineering and software quality in the domain of evolutionary computation and should enable researchers to exchange their ideas on how to develop and apply generic and reusable software systems. It should provide a common platform to present open and freely available frameworks, to discuss novel approaches in their development, to identify cooperation potentials and synergies between research groups, and to define common standards.

Different learning algorithms have different accuracies. The no free lunch theorem asserts that no single learning algorithm always achieves the best performance in any domain. They can be combined to attain higher accuracy. Data fusion is the process of fusing multiple records representing the same real-world object into a single, consistent, and clean representation. Fusion of data for improving prediction accuracy and reliability is an important problem in machine learning.

In this study, we have compared six state-of-the-art Multiobjective Evolutionary Algorithms (MOEAs) designed specifically for many-objective optimization problems under a number of carefully crafted benchmark problems. Using the performance metrics ensemble, we aim at providing a comprehensive measure and more importantly revealing insight pertaining to specific problem characteristics that the underlying MOEA could perform the best. The experimental results confirm the finding from the No Free Lunch theorem: any algorithm's elevated performance over one class of problems is exactly paid for in loss over another class. In addition, the experimental results show that the performance of MOEA to solve many-objective optimization problems depends on two distinct aspects: the ability of MOEA to tackle the specific characteristics of the problem and the ability of MOEA to handle high-dimensional objective space.

This is the first of two papers that use off-training set (OTS) error to investigate the assumption-free relationship between learning algorithms. This first paper discusses the senses in which there are no a priori distinctions between learning algorithms. (The second paper discusses the senses in which there are such distinctions.) In this first paper it is shown, loosely speaking, that for any two algorithms A and B, there are “as many” targets (or priors over targets) for which A has lower expected OTS error than B as vice versa, for loss functions like zero-one loss. In particular, this is true if A is cross-validation and B is “anti-cross-validation” (choose the learning algorithm with largest cross-validation error). This paper ends with a discussion of the implications of these results for computational learning theory. It is shown that one cannot say: if empirical misclassification rate is low, the Vapnik-Chervonenkis dimension of your generalizer is small, and the training set is large, then with high probability your OTS error is small. Other implications for “membership queries” algorithms and “punting” algorithms are also discussed.

The evolution of so many different metaheuristic optimization algorithms results from the fact that no single method can outperform all others for all possible problems. As postulated in the No Free Lunch Theorem, a general-purpose and universal optimization strategy is impossible. The only way how one strategy can outperform another is to be more specialized to the structure of the tackled problem. Consequently it always takes qualified algorithm experts to select and tune a metaheuristic algorithm for a concrete application. Choosing an appropriate method for a certain problem is not trivial, as problem characteristics may change remarkably for different instances and the performance of a metaheuristic may vary considerably for different parameter settings. Therefore the choice of a well suited method an according parameter values is a crucial aspect when applying metaheuristics. Because of these issues it is not advisable to implement only one specific metaheuristic algorithm for a specific problem, as it cannot be determined in advance whether the chosen method is suitable for the tackled problem or not. Thus soundly engineered, reusable, flexible, user-friendly and interoperable software systems are required which offer a broad spectrum of different metaheuristic algorithms and support researchers in adapting and comparing these methods for different problems in order to bridge the gap between theoretical research and practical applications. However, the development of such systems is both, time consuming and complex. Surprisingly the chance to join forces in the development of evolutionary computation software systems is not yet seized sufficiently within the evolutionary computation research community. Instead of reusing existing software and continuing their work on the shoulders of others, many researchers in the evolutionary computation domain implement individual and highly specialized applications. Often these systems are not even publically released which hinders comparability and replicability of research results. This workshop should highlight the importance of software engineering and software quality in the domain of evolutionary computation and should enable researchers to exchange their ideas on how to develop and apply generic and reusable software systems. It should provide a common platform to present open and freely available frameworks, to discuss novel approaches in their development, to identify cooperation potentials and synergies between research groups, and to define common standards. Many thanks go to the authors, reviewers, and Program Committee members whose efforts made this workshop possible in the first place. We genuinely hope you enjoy the EvoSoft Workshop and that it will boost joint efforts in the development of evolutionary computation software systems.

Given the posterior probability estimates of 14 classifiers on 38 datasets, we plot two-dimensional maps of classifiers and datasets using principal component analysis (PCA) and Isomap. The similarity between classifiers indicate correlation (or diversity) between them and can be used in deciding whether to include both in an ensemble. Similarly, datasets which are too similar need not both be used in a general comparison experiment. The results show that (i) most of the datasets (approximately two third) we used are similar to each other, (ii) multilayer perceptrons and k-nearest neighbor variants are more similar to each other than support vector machine and decision tree variants. (iii) the number of classes and the sample size has an effect on similarity. Publisher's Version Author Pre-Print

The complexity of fitness landscapes, e.g., in terms of multimodalities and presence of plateaus, typical of many realworld optimization problems coupled with the limitations imposed by the No Free Lunch Theorem, are strong mitigating factors for domain-specific Memetic Computing approaches. More specifically, since the No Free Lunch Theorem proves that the performance of each algorithm over all the possible problems is the same, there is no longer a reason to discuss which algorithm is universally better or worse. Thus, instead of trying to propose universally applicable algorithms, researchers in recent years have started to propose algorithms which are tailored specific to the problems in hand. In this context, the Memetic Computing paradigm offers the possibility of flexibly designing domain-specific optimization algorithms by integrating and coordinating algorithmic components capable of dealing with difficulties specifically related to the decision space and fitness landscape of a given problem. This thematic issue deals with a set of problem difficulties which are of great interest in an industrial context, i.e., uncertainties in the fitness function, and gathers novel Memetic Computing approaches which attempt to solve this set of problems. It addresses those optimization problems characterized by a complex fitness landscape and uncertain environments. An optimization problem contains uncertainties

Conservation of information (COI) popularized by the no free lunch theorem is a great leveler of search algorithms, showing that on average no search outperforms any other. Yet in practice some searches appear to outperform others. In consequence, some have questioned the significance of COI to the performance of search algorithms. An underlying foundation of COI is Bernoulli's Principle of Insufficient Reason1(PrOIR) which imposes of a uniform distribution on a search space in the absence of all prior knowledge about the search target or the search space structure. The assumption is conserved under mapping. If the probability of finding a target in a search space is p, then the problem of finding the target in any subset of the search space is p. More generally, all some-to-many mappings of a uniform search space result in a new search space where the chance of doing better than p is 50–50. Consequently the chance of doing worse is 50–50. This result can be viewed as a confirming property of COI. To properly assess the significance of the COI for search, one must completely identify the precise sources of information that affect search performance. This discussion leads to resolution of the seeming conflict between COI and the observation that some search algorithms perform well on a large class of problems.

This paper shows that a constraint on universal Turing machines is necessary for Legg's and Hutter's formal measure of intelligence to be unbiased. It also explores the relation of the No Free Lunch Theorem to formal measures of intelligence.

As was shown recently [19], prior knowledge has a significant importance in machine learning from the no free lunch theorem viewpoint. One of the type of prior information for classification task is knowledge on the data. Here we will propose another type of prior knowledge, for which a distance from decision boundary to selected data samples (detractors) is maximised. Support Vector Machines (SVMs) is a widely used algorithm for data classification. Detractors will be incorporated into SVMs by weighting the samples. For the reason, that standard C-SVM sample weights are not suitable for maximising the distance to selected points, additional SVM weights will be proposed.We will show that detractors can enhance the classification quality for areas with lack of training samples and for time series classification.We will demonstrate that incorporating detractors to stock price predictive models can lead to increased investment profits.

The no free lunch (NFL) theorem tells us that without any structural assumptions on an optimization problem, no algorithm can perform better on average than blind search. Although for most such impossibility theorems, the proofs are lengthy, difficult, and often not at all intuitive, we can get a feel for the NFL theorem by considering the proverbial needle in a haystack problem. Clearly, in this instance, no algorithm has any better chance of finding the optimum than blind search. The title also refers to the human-machine interface (HMI). Traditionally, this term has had the narrow meaning of facilitating communication between humans and computing machines, such as via the graphical user interface or head-up displays. But as technology advances, this communication interface will also evolve. Certainly, in the future we can see voice, fuzzy, and natural language inputs and outputs. Here, however, we are using the term in an even broader sense: to demarcate the dividing line between what humans do and what machines do in problem solving or optimization. The twin purposes of the article are to explore the implications of NFL and to address the proper allocation of natural and computational intelligence in optimization problem solving.

Recent developments in the theory of evolutionary computation offer evidence and proof that overturns several conventionally held beliefs. In particular, the no free lunch theorem and other related theorems show that there can be no best evolutionary algorithm, and that no particular variation operator or selection mechanism provides a general advantage over another choice. Furthermore, the fundamental nature of the notion of schema processing is called into question by recent theory that shows that the schema theorem does not hold when schema fitness is stochastic. Moreover, the analysis that underlies schema theory, namely the k- armed bandit analysis, does not generate a sampling plan that yields an optimal allocation of trials, as has been suggested in the literature for almost 25 years. The importance of these new findings is discussed in the context of future progress in the field of evolutionary computation.© (1999) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

This paper provides insight into combining stochastic and deterministic search methods using evolutionary algorithms (EAs) such as evolutionary programming, evolutionary strategies, and genetic algorithms integrated with depth-first search with backtracking, branch and bound, and best-first search algorithms such as A*. An important view of such an integration focuses on both theoretical analysis and experimental evaluation. Included in the discussion is the constraining impact of the “No Free Lunch Theorem” on combined search performance. Also, a variety of combinations of EAs and deterministic combined search methods are proposed along with expert system components. Discussion of anticipated results of such architectures speculates about existence of high performance integrated search environments. A particular successful specific NPC problem environment is presented that employs search metrics for short-term performance evaluation and combined search algorithm selection.

This note discusses the recent paper "Some technical remarks on the proof of the no free lunch theorem" by Koppen (2000). In that paper, some technical issues related to the formal proof of the no free lunch (NFL) theorem for search were given by Wolpert and Macready (1995, 1997). The present authors explore the issues raised in that paper including the presentation of a simpler version of the NFL proof in accord with a suggestion made explicitly by Koppen (2000) and implicitly by Wolpert and Macready (1997). They also includes the correction of an incorrect claim made by Koppen (2000) of a limitation of the NFL theorem. Finally, some thoughts on future research directions for research into algorithm performance are given.