Search

Dimensionality reduction methods are used routinely in statistics, pattern recognition, data mining, and machine learning to cope with high-dimensional spaces. Also in the case of high-dimensional multiobjective optimization problems, a reduction of the objective space can be beneficial both for search and decision making. New questions arise in this context, e.g., how to select a subset of objectives while preserving most of the problem structure. In this chapter, two different approaches to the task of objective reduction are developed, one based on assessing explicit conflicts, the other based on principal component analysis (PCA). Although both methods use different principles and preserve different properties of the underlying optimization problems, they can be effectively utilized either in an a posteriori scenario or during search. Here, we demonstrate the usability of the conflict-based approach in a decision-making scenario after the search and show how the principal-component-based approach can be integrated into an evolutionary multicriterion optimization (EMO) procedure. [ABSTRACT FROM AUTHOR]

We consider a class of optimization problems wherein the quality of candidate solutions is estimated by their performance on a number of tests. Classifier induction, function regression, and certain types of reinforcement learning, including problems often attacked with coevolutionary algorithms, can all be seen as members of this class. Traditional approaches to such test-based problems use a single objective function that aggregates the scores obtained on the tests. Recent work, by contrast, argues that useful finer-grained distinctions between candidate solutions are obtained when each test is treated as a separate objective, and that algorithms employing such multiobjective comparisons show favourable behaviour relative to those which do not. Unfortunately, the number of tests can be very large. Since it is well-known that high-dimensional multiobjective optimization problems are difficult to handle in practice, the question arises whether the multiobjective treatment of test-based problems is feasible. To begin addressing this question, we examine a method for reducing the number of dimensions without sacrificing the favorable properties of the multiobjective approach. Our method, which is a form of dimension extraction, finds underlying objectives implicit in test-based problems. Essentially, the method proceeds by placing the tests along the minimal number of coordinate axes that still preserve ordering information among the candidate solutions. Application of the method to the strategy set for several instances of the game of Nim suggest the technique has significant practical benefits: a type of compression of the set of objectives is observed in all tested instances. Surprisingly, we also find that the information contained in the arrangement of tests on the coordinate axes reveals important information about the structure of the underlying problem. [ABSTRACT FROM AUTHOR]

Model-based multiobjective optimization is one class of metaheuristics for solving multiobjective optimization problems, where a probabilistic model is built from the current distribution of the solutions and new candidate solutions are generated from the model. One main difficulty in model-based optimization is constructing a probabilistic model that is able to effectively capture the structure of the problems to enable efficient search. This chapter advocates a new type of probabilistic model that takes the regularity in the distribution of Pareto-optimal solutions into account. We compare our model to two other model-based multiobjective algorithms on a number of test problems to demonstrate that it is scalable to high-dimensional optimization problems with or without linkage linkage among the design variables. [ABSTRACT FROM AUTHOR]

This chapter considers a number of alternative methods for fitness assignment in evolutionary algorithms for multiobjective optimization. Most of the fitness assignment methods in the literature were designed to work for any number of objectives, in principle; but, in practice, some of the more popular methods (e.g. those in NSGA-II, IBEA and SPEA) do not perform well on problems with four or more objectives. We investigate why this is the case, considering two aspects of performance: convergence towards the Pareto front and drive towards a set of well spread solutions. The visualization of induced fitness surfaces is used to understand the effects of the different fitness assignment methods, and both Pareto- and non-Pareto-based methods are analysed. [ABSTRACT FROM AUTHOR]

This chapter provides a comparative discussion on natural and artificial systems. It focuses on multiobjective problems as related to the evolution of systems either naturally or artificially; yet, it should be viewed as relevant to other forms of adaptation. Research developments in areas such as evolutionary design, plant biology, robotics, A-life, biotechnology, and game theory are used to support the comparative discussion. A unified approach, namely multi-competence cybernetics (MCC) is suggested. This is followed by a discussion on the relevance of a Pareto approach to the study of nature. One outcome of the current MCC study is a suggested analogy between species and design concepts. Another resulting suggestion is that multi-fitness dynamic visualization of natural systems should be of a scientific value, and in particular for the pursuit of understanding of natural evolution by way of ∈dexthought experimentthought experiments. It is hoped, at best, that MCC would direct thinking into fruitful new observations on the multi-fitness aspects of natural adaptation. Alternatively, it is expected that such studies would allow a better understanding of the similarities and dissimilarities between the creation of natural and artificial systems by adaptive processes. [ABSTRACT FROM AUTHOR]

This chapter centres around the use of interactive evolutionary computation as a search and exploration tool for open-ended contexts in design. Such contexts are characterized by poor initial definition and uncertainty in terms of objectives, constraints and defining variable parameters. The objective of the research presented is the realization of ‘user-centric' intelligent systems, i.e., systems which can overcome initial lack of understanding and associated uncertainty, whilst also stimulating innovation and creativity through a high degree of human / machine interaction. Two application areas are used to illustrate how, through the adoption of bespoke visualization techniques, flexible representations, and machine learning agents that ‘observe' the evolutionary process, this objective can be achieved. [ABSTRACT FROM AUTHOR]

In optimization studies, often researchers are interested in finding one or more optimal or near-optimal solutions. In this chapter, we describe a systematic optimization-cum-analysis procedure which performs a task beyond simply finding optimal solutions, but first finds a set of near-Pareto-optimal solutions and then analyses them to unveil salient knowledge about properties which make a solution optimal. The proposed ‘innovization' task is explained and its working procedure is illustrated on a number of engineering design tasks. The variety of problems chosen in the chapter and the resulting innovations obtained for each problem amply demonstrate the usefulness of the proposed innovization task. The procedure is a by-product of performing a routine multiobjective optimization for a design task and in our opinion portrays an important process of knowledge discovery which may not be possible to achieve by other means. [ABSTRACT FROM AUTHOR]

In this chapter, we discuss the application of evolutionary multiobjective optimization (EMO) to association rule mining. Especially, we focus our attention on classification rule mining in a continuous feature space where the antecedent and consequent parts of each rule are an interval vector and a class label, respectively. First we explain evolutionary multiobjective classification rule mining techniques. Those techniques are roughly categorized into two approaches. In one approach, each classification rule is handled as an individual. An EMO algorithm is used to search for Pareto-optimal rules with respect to some rule evaluation criteria such as support and confidence. In the other approach, each rule set is handled as an individual. An EMO algorithm is used to search for Pareto-optimal rule sets with respect to some rule set evaluation criteria such as accuracy and complexity. Next we explain evolutionary multiobjective rule selection as a post-processing procedure in classification rule mining. Pareto-optimal rule sets are found from a large number of candidate classification rules, which are extracted from a database using an association rule mining technique. Then we examine the effectiveness of evolutionary multiobjective rule selection through computational experiments on some benchmark classification problems. Finally we examine the use of Pareto-optimal and near Pareto-optimal rules as candidate rules in evolutionary multiobjective rule selection. [ABSTRACT FROM AUTHOR]

The work presented in this chapter is concerned with the identification and modelling of nonlinear dynamical systems using multiobjective evolutionary algorithms (MOEAs). This problem involves the processes of structure selection, parameter estimation, model performance and model validation and defines a complex solution space. Evolutionary algorithms (EAs), in particular genetic programming (GP), are found to provide a way of evolving models to solve this identification and modelling problem, and their use is extended to encompass multiobjective functions. Multiobjective genetic programming (MOGP) is then applied to multiple conflicting objectives in order to yield a set of simple and valid human-understandable models which can reproduce the behaviour of a given unknown system. [ABSTRACT FROM AUTHOR]

This chapter investigates the use of multiobjective techniques in genetic programming (GP) in order to evolve compact programs and to reduce the effects caused by bloating. The underlying approach considers the program size as a second, independent objective besides program functionality, and several studies have found this concept to be successful in reducing bloat. Based on one specific algorithm, we demonstrate the principle of multiobjective GP and show how to apply Pareto-based strategies to GP. This approach outperforms four classical strategies to reduce bloat with regard to both convergence speed and size of the produced programs on an even-parity problem. Additionally, we investigate the question of why the Pareto-based strategies can be more effective in reducing bloat than alternative strategies on several test problems. The analysis falsifies the hypothesis that the small but less functional individuals that are kept in the population act as building blocks building blocks for larger correct solutions. This leads to the conclusion that the advantages are probably due to the increased diversity in the population. [ABSTRACT FROM AUTHOR]

This chapter sets out a number of the popular areas in multiobjective supervised learning. It gives empirical examples of model complexity optimization and competing error terms, and presents the recent advances in multi-class receiver operating characteristic analysis enabled by multiobjective optimization. It concludes by highlighting some specific areas of interest/concern when dealing with multiobjective supervised learning problems, and sets out future areas of potential research. [ABSTRACT FROM AUTHOR]

This chapter identifies five distinct modes∈dexmodes of multiobjective optimization in which multiobjective optimization is used to solve practical optimization problems. Implications for the interpretation and analysis of the resulting Pareto front, and for decision making, are discussed, and each mode is illustrated using application examples taken from recent research. [ABSTRACT FROM AUTHOR]

Many real-world problems are multiobjective optimization problems, and evolutionary algorithms are quite successful on such problems. Since the task is to compute or approximate the Pareto front, multiobjective optimization problems are considered as more difficult than single-objective problems. One should not forget that the fitness vector with respect to more than one objective contains more information that in principle can direct the search of evolutionary algorithms. Therefore, it is possible that a single-objective problem can be solved more efficiently via a generalized multiobjective model of the problem. That this is indeed the case is proved by investigating the single-source shortest paths problem and the computation of minimum spanning trees. [ABSTRACT FROM AUTHOR]

Finding the native structure of a protein starting from its amino acid sequence remains one of the most challenging open problems in bioinformatics and molecular biology. The Protein Structure Prediction (PSP) problem has been tackled from many different directions. The common approach is to cast it in the form of a global single-objective optimization problem using energy functions to evaluate the physical state of the conformations. In this work we reformulate the PSP as a multiobjective optimization problem motivated by the fact that the folded state of a protein is a small ensemble of conformational structures. A 2-objective decomposition of the CHARMM energy function is proposed based on local and nonlocal interactions between atoms, supported by experimental evidence that these objectives are in fact conflicting. A new MOEA algorithm is used to search for (observed) Pareto-optimal sets of conformations with respect to the 2-objective formulation and tested on a large set of medium-size proteins (26-70 residues), with results demonstrating the effectiveness of this approach and providing different measures of protein complexity. Results also point to instances in which the CHARMM energy model suffers from low accuracy owing to the required trade-off between differing objectives in finding "good" conformations. [ABSTRACT FROM AUTHOR]

In this chapter, we discuss the use of multiobjective evolutionary algorithms (MOEAs) for solving single-objective optimization problems in dynamic environments. Specifically, we investigate the consideration of a second (artificial) objective, with the aim of maintaining greater population diversity and adaptability. The paper suggests and compares a number of alternative ways to express this second objective. An empirical comparison shows that the best alternatives are competitive with other evolutionary algorithm variants designed for handling dynamic environments. [ABSTRACT FROM AUTHOR]

In this chapter, we present a survey of constraint-handling techniques based on evolutionary multiobjective optimization concepts. We present some basic definitions required to make this chapter self-contained, and then introduce the way in which a global (single-objective) nonlinear optimization problem is transformed into an unconstrained multiobjective optimization problem. A taxonomy of methods is also proposed and each of them is briefly described. Some interesting findings regarding common features of the approaches analyzed are also discussed. [ABSTRACT FROM AUTHOR]

This chapter reviews a line of coevolutionary algorithm research that reframes coevolution as a form of multiobjective optimization. This shift in perspective towards coevolution as a form of multiobjective optimization has led researchers to new algorithmic and analytical formulations that have advanced the state of the art in coevolutionary algorithm research. We review relevant literature and discuss the basic concepts and issues involved in the application of multiobjective optimization to coevolutionary algorithms. [ABSTRACT FROM AUTHOR]

This book explores some emerging techniques for problem solving of a general nature, based on the tools of EMO. In this introduction, we provide background material to support the reader's journey through the succeeding chapters. Given here are a basic introduction to optimization problems, and an introductory treatment of evolutionary computation, with thoughts on why this method is so successful; we then discuss multiobjective problems, providing definitions that some future chapters rely on, covering some of the key concepts behind multiobjective optimization. These show how optimization can be carried out separately from subjective factors, even when there are multiple and conflicting ends to the optimization process. This leads to a set of trade-off solutions none of which is inherently better than any other. Both the process of multiobjective optimization, and the set of trade-offs resulting from it, are ripe areas for innovation — for new techniques for problem solving. We briefly preview how the chapters of this book exploit these concepts, and indicate the connections between them. [ABSTRACT FROM AUTHOR]

This chapter examines the possibilities and challenges that lie ahead for evolutionary music and art. Evolutionary computing methods have enabled new modes of creative expression in the art made by humans. One day, it may be possible for computers to make art autonomously. The idea of machines making art leads to the question: what do we mean by ‘making art' and how do we recognise and acknowledge artistic creativity in general? Two broad categories of human-machine creativity are defined: firstly, machines that make art like, and for, humans; and secondly, machines that make ‘art' that is recognised as creative and novel by other machines or agents. Both these categories are examined from an evolutionary computing perspective. Finding ‘good' art involves searching a phase-space of possibilities beyond astronomical proportions, which makes evolutionary algorithms potentially suitable candidates. However, the problem of developing artistically creative programs is not simply a search problem. The multiple roles of interaction, environment, physics and physicality are examined in the context of generating aesthetic output. A number of ‘open problems' are proposed as grand challenges of investigation for evolutionary music and art. For each problem, the impetus and background are discussed. The paper also looks at theoretical issues that might limit prospects for art made by machines, in particular the role of embodiment, physicality and morphological computation in agent-based and evolutionary models. Finally, the paper looks at artistic challenges for evolutionary music and art systems. [ABSTRACT FROM AUTHOR]

A novel approach to the production of evolutionary art is presented. This approach is based on the promotion of an arms race between an adaptive classifier and an evolutionary computation system. An artificial neural network is trained to discriminate among images previously created by the evolutionary engine and famous paintings. Once training is over, evolutionary computation is used to generate images that the neural network classifies as paintings. The images created throughout the evolutionary run are added to the training set and the process is repeated. This iterative process leads to the refinement of the classifier and forces the evolutionary algorithm to explore new paths. The experimental results attained across iterations are presented and analyzed. Validation tests were conducted in order to assess the changes induced by the refinement of the classifier and to identify the types of images that are difficult to classify. Taken as a whole, the experimental results show the soundness and potential of the proposed approach. [ABSTRACT FROM AUTHOR]

Following the ground breaking work of Sims and Latham there was a flurry of activity in interactive artificial evolution of images. However, the move towards non-interactive evolution of images that arises by invoking fitness functions to serve in place of users in order to guide simulated evolution proceeded haltingly and unevenly. If evolutionary computational models for image evolution are indeed inspired by nature, then it is natural to consider image evolution in the broader co-evolutionary context. This chapter briefly surveys the role co-evolutionary methods have played in evolutionary computation and then examines some of the instances where it has been applied to evolutionary art. The paucity of examples leads to a discussion of the challenges faced, and the difficulties encountered, when trying to use co-evolutionary methods both in evolutionary art and artificial creativity. [ABSTRACT FROM AUTHOR]

Artistic image filters are evolved using genetic programming. The system uses automatic image analysis during fitness evaluation. Multi-objective optimization permits multiple feature tests to be applied independently. One unique fitness test is Ralph's bell curve model of aesthetics. This model is based on an empirical evaluation of hundreds of fine art works, in which paintings have been found to exhibit a bell curve distribution of color gradient. We found that this test is very useful for automatically evolving non-photorealistic filters that tend to produce images with painterly, balanced and harmonious characteristics. The genetic programming language uses a variety of image processing functions of varying complexity, including a higher-level paint stroke operator. The filter language is designed so that components can be combined together in complex and unexpected ways. Experiments resulted in a surprising variety of interesting "artistic filters", which tend to function more like higher-level artistic processes than low-level image filters. Furthermore, a correlation was found between an image having a good aesthetic score, and its application of the paint operator. [ABSTRACT FROM AUTHOR]

Artists have always learned from nature. A new generation of artists is adapting the very processes of life to create exciting new works. But art is more than the creation of objects. It is also a progression of ideas with a history and a correspondence to the larger culture. The goal of this chapter is to take a step back from the details of the technology and the consideration of specific works, and to view evolutionary art in the broader context of all art. This kind of multidisciplinary discussion requires one to be multilingual, and this chapter will use the language of scientists, humanists, artists, and philosophers. While doing so we will quickly visit complexity science, postmodernism in the arts, and the conflict between the cultures of the humanities and the sciences. With this as a backdrop, I will introduce a new approach I call complexism. Complexism is the application of a scientific understanding of complex systems to the subject matter of the arts and humanities. We will see that the significance of evolutionary art is that it takes complexism as both its method and content. Evolutionary art is a new kind of dynamic iconography: the iconography of complexism. And complexism offers nothing less than the reconciliation of the sciences and the humanities through a higher synthesis of the modern and the postmodern. To a certain extent this chapter participates in the modernist tradition of the art manifesto. The art manifesto is a form of speculative writing where the artist-author posits a new revolutionary creative direction for a group of artists who share a set of common interests, as well as a new worldview that offers a radical break with the past. Writers of such manifestos have included Marinetti, Kandinsky, Schwitters, Moholy-Nagy, Gropius, Breton, and others[1]. Like other manifestos, this chapter includes forward-looking assertions about work not yet started let alone completed. I have tried to identify the more speculative parts of this chapter as being part of this complexist manifesto. [ABSTRACT FROM AUTHOR]

This paper explores the application of ecosystem simulation to the production of works of generative electronic art. The aim is to demonstrate that virtual ecosystems are capable of producing outcomes that are rich, complex and interesting aesthetically. A number of artworks that employ virtual ecosystems are surveyed. The author argues that the most interesting works of generative art exhibit four basic properties: coherence and unity; multi-scaled temporal complexity; autonomous production of novelty; responsiveness to perturbation. The virtual ecosystem is assessed for its suitability as a medium for constructing generative art in light of these desirable properties. It is concluded that the ecosystem's strengths lie in its exhibition of multi-scaled complexity and its autonomous production of novelty. Whilst an artist may manipulate a simulation to retain visual and sonic coherence, the software also possesses an implicit coherence inherent in its ability to self-organize. Under some circumstances it appears that the weakness of the virtual ecosystem as an artistic medium lies in its unpredictable response to perturbation. Consequently, the paper also explores virtual ecosystems' susceptibility to external control and describes methods that have been employed to adjust the responsiveness of art works that employ them. [ABSTRACT FROM AUTHOR]

A software application called "Musical Gene Pool" is described. It was designed to evolve non-linear music from an initial random soup of sounds, which play continuously. Most evolutionary music systems to date require the user to select for musical aspects in a piecemeal fashion, whereas this system is experienced as continuous music throughout the entire process, as follows: a human listener gives fitness rewards after sounds (organisms) emerge from the gene pool, take turns playing, and return back to the pool. Organisms start out unicellular (one sound), but as the listener selectively rewards random sequences deemed more musical than others, some organisms join up to form larger, multicellular organisms - which become phrases or extended musical gestures. Genetic operators of splitting, death, replication, and mutation occur in the gene pool among rewarded organisms. This results in gradual evolution of structure as the music continues to play. This emerges in response to the listener's own internal emerging musical language, based on accumulated musical memory. This structure is liquid - continually able to flow and rearrange to allow serendipity. While there is a limit to organism length (duration of phrases), it is proposed that the interactive scheme could be adjusted to evolve increasingly larger organisms, and hence, longer musical passages. These would essentially be mobile chunks of linear music with self-similarity in their structures - revealing the histories of their evolution. [ABSTRACT FROM AUTHOR]

My art comprises three levels: basic, methodical, and superordinate. The basic level refers to the production of individual works of art. At the methodical level, I am concerned with the development of artistic processes, in this case, inspired by evolutionary concepts. Finally, at the superordinate level, social sculpture, "Health Art", is considered. Between 1995 and 2003 I developed and used several pixel-based evolutionary art processes. These artistic approaches resort to biologically inspired concepts, such as population, variation, and selection. The individuals were pixel images. In the reproduction phase two (or more) individuals were selected as parents and the images were recombined by the exchange of image parts (Regions-of-Interest). Variation (mutation) was performed by means of image processing operations such as the rotation of image parts with random but constrained parameters. Parents and offspring were evaluated by the aesthetic preferences of the artist and the best individuals built the next generation. A unique and more complex evolution art process was developed in 2004, with no direct correspondence to known evolutionary algorithms like GA, ES, or GP. It uses some additional evolutionary concepts, such as a global pool of images and multi-sexual recombination, together with ontogenetic concepts, such as spores or fruits, and other concepts such as image templates. In a narrow sense, there is no change of generations of image individuals. Selected individuals are directly inserted into the global image pool, being also involved in a second reproduction process where they are transformed by mathematical symmetry operations. The resulting images are also inserted into the global image pool, while all the images in this pool build the basis for recombination in the next generation. Future plans, about file formats, other image reproduction operators, image databases, and aesthetic preference modelling, are also discussed. [ABSTRACT FROM AUTHOR]

We present how we have considered the artificial ant paradigm as a tool for the generation of music and painting. From an aesthetic perspective, we are interested in demonstrating that swarm intelligence and self-organization can lead to spatio-temporal structures that can reach an artistic dimension. In our case, the use of artificial pheromones can lead to the creation of melodies thanks to a cooperative behavior of the ant-agents but also to the emergence of abstract paintings thanks to competitive behaviors within the artificial colonies. The user's point of view is also taken into account through interactive genetic algorithms. [ABSTRACT FROM AUTHOR]

Creative evolutionary systems are often concerned with producing images of high artistic quality. A key challenge to such a system is to be human-competitive by producing the same quality. Then, mimicking existing human artists could be seen as a canonical benchmark, not unlike the Turing test for intelligence. This chapter discusses two applications aimed at evolving images in the styles of two well-known Dutch painters: Mondriaan and Escher. For both cases we have an evaluation criterion based on "style-fidelity" as perceived and judged by the users. In other words, here we have a target style, which makes the (subjective) selection less free than in applications solely aiming at nice images. Technically, the Mondriaan evolver is less difficult, given that his most popular style "simply" uses horizontal and vertical lines, and primary colours to fill the resulting rectangles. The Escher evolver project is more challenging. First, because Escher's style is less simple to capture. Second, the system is tested in vivo, in a real museum, posing requirements on the interface. We describe how to meet the style challenge based on the mathematical system behind Escher's tiling. Designing a suitable representation and the corresponding variation operators based on this system specifies an appropriate search space guaranteeing the Escher style to some extent and leaving enough freedom for the selection. As for the second objective, we describe two versions that differ in the way the images are presented to, respectively evaluated by the visitors. The experiences gained during a six-month exhibition period in the City Museum in The Hague, The Netherlands, are discussed from the visitors' perspective as well as from the algorithmic point of view and are illustrated with some "evolved Escher's". [ABSTRACT FROM AUTHOR]

Witnesses and victims of serious crime are normally requested to construct a picture of the criminal's face. These pictures are known as facial composites and are typically produced by a witness recalling details of the face and then selecting individual facial features: hair, eyes, nose, mouth, etc. While composites remain an important tool for the apprehension of criminals, research has suggested that, even under favorable conditions, they are rarely recognized. In the current chapter, we present a new method called EvoFIT whereby users select complete faces and a composite is "evolved" using a Genetic Algorithm. While considerable development was required to tune the new approach, research indicates that EvoFIT now produces more identifiable composites than those produced from the traditional "feature" systems. Novel applications of the technology are also discussed. [ABSTRACT FROM AUTHOR]

We present the computational design tool Genr8 and six different architectural projects making extensive use of Genr8. Genr8 is based on ideas from Evolutionary Computation (EC) and Artificial Life and it produces surfaces using an organic growth algorithm inspired by how plants grow. These algorithms have been implemented as an architect's design tool and the chapter provides an illustration of the possibilities that the tool provides. [ABSTRACT FROM AUTHOR]

Evolutionary Design Evolutionary design can take many forms. In this chapter, we describe how different evolutionary techniques — such as genetic programming and evolution strategies — can be applied to a wide variety of nature-inspired designs. We will show how techniques of interactive evolutionary breeding can facilitate the creative processes of design. As practical examples we demonstrate how to use implicit surface modeling to create virtual sculptures, and furniture designs through evolutionary breeding. Rather than creating variations of blueprints through an evolutionary process, we then focus on the evolution of ‘design programs'. That is, instead of a static description (blueprint) of an object, we evolve recipes or algorithms to build objects. This leads to a much wider repertoire of variability on the designer's side and can be implemented in a straightforward manner using genetic programming. Starting with a simple breeding approach of fractals, we give examples of how to — either automatically or interactively — evolve growth programs for plants with particular characteristics, which we illustrate using a garden of artificial flowers. We use evolvable Lindenmayer systems (L-systems) to capture growth processes. The evolution of choreographic swarm interactions leads to new ways of ‘swarm programming', where changes in control parameters result in emergent agent behaviours. Swarm grammars, as we will show, combine swarming agents with developmental programs as an extension of L-systems. We demonstrate how to use this technique to generate virtual paintings on 2D and 3D canvases. These SwarmArt implementations have also been exhibited in various museums as interactive computer installations, which we will use to describe how to integrate music and sound generation into evolutionary swarm systems. [ABSTRACT FROM AUTHOR]

We describe an evolutionary approach to one of the most challenging problems in computer music: modeling the knowledge applied by a musician when performing a score of a piece in order to produce an expressive performance of the piece. We extract a set of acoustic features from jazz recordings, thereby providing a symbolic representation of the musician's expressive performance. By applying an evolutionary algorithm to the symbolic representation, we obtain an interpretable expressive performance computational model. We use the model to generate automatically performances with the timing and energy expressiveness of a human saxophonist. [ABSTRACT FROM AUTHOR]

Sound synthesis is a natural domain in which to apply evolutionary computation (EC). The EC concepts of the genome, the phenotype, and the fitness function map naturally to the synthesis concepts of control parameters, output sound, and comparison with a desired sound. More importantly, sound synthesis can be a very unintuitive technique, since changes in input parameters can give rise, via non-linearities and interactions among parameters, to unexpected changes in output sounds. The novice synthesizer user and the simple hill-climbing search algorithm will both fail to produce a desired sound in this context, whereas an EC technique is well-suited to the task. In this chapter we introduce and provide motivation for the application of EC to sound synthesis, surveying previous work in this area. We focus on the problem of automatically matching a target sound using a given synthesizer. The ability to mimic a given sound can be used in several ways to augment interactive sound synthesis applications. We report on several sets of experiments run to determine the best EC algorithms, parameters, and fitness functions for this problem. [ABSTRACT FROM AUTHOR]