Supporting creative design in a visual evolutionary computing environment

Transcription

1 Advances in Engineering Software 35 (2004) Supporting creative design in a visual evolutionary computing environment Hong Liu a, *, Mingxi Tang b, John Hamilton Frazer b a School of Information Management, Shandong Normal University, Jinan City , People s Republic of China b Design Technology Research Center, School of Design, The Hong Kong Polytechnic University, Hong Kong, People s Republic of China Received 30 October 2002; revised 27 February 2004; accepted 30 March 2004 Abstract In product design, sketches and images are strong devices for stimulating creativity. This paper presents a novel visual evolutionary computing environment to provide support for creative design. The 2D sketches and 3D images can be generated by combination of evolutionary computing technology and visualization technology in this environment. A tree-based genetic algorithm is illustrated by a reading lamp design example, which uses genetic algorithm with binary mathematical expression tree to form 2D sketches and programming using Visual Cþþ6.0 and ACIS to generate 3D images. It shows that the approach is able to generate some creative solutions and demonstrates the potential of computational approach in creative design. q 2004 Elsevier Ltd. All rights reserved. Keywords: Creative design; Generic algorithm; Mathematical expression tree; Visualization 1. Introduction The quality of the product heavily lies in its design [1].In today s highly competitive market place, the strategy of developing a product is transformed from product-push type to market-pull model. Facing the intense changes in the market, a well-designed product should not only satisfy consumers physical requirements but also satisfy their psychological needs. Design must exhibit performance, not only in quality and productivity, but also in novel and goodlooking externality [2,3]. This requires that designers and engineers use various kinds of design knowledge and tools for supporting their creative design [4]. This paper presents a new way of using evolutionary computing technology and visualization technology to support creative design. Our goal is to give more opportunities to designers to be creative by unleashing the creative potential with computational environment. The approach to support creative design is to develop computational tools that can generate useful sketches and images * Corresponding author. Tel.: þ ; fax: þ address: lhsdcn@jn-public.sd.cninfo.net (H. Liu) /$ - see front matter q 2004 Elsevier Ltd. All rights reserved. doi: /j.advengsoft for simulating the mind s eye of designers in the creative design process. The remainder of this paper is organized as follows. Section 2 analyses the support of media and environment for creative design. Section 3 summarizes related work while Section 4 introduces tree-based generic algorithm. In Section 5, a reading lamp design example is presented for showing how to use the generic algorithm and mathematical expressions to generate 2D sketch shapes and 3D images. Section 6 summarizes the paper and gives an outlook for the future work. 2. The support of media and environment for creative design 2.1. Creativity in design Engineering design may be defined as a process of establishing requirements based on human needs, transforming them into performance specification and functions, which are then mapped and converted (subject to constraints) into design solution (using creativity, scientific principles and technical knowledge) that can be economically manufactured and produced. From the viewpoint of 转载

2 262 H. Liu et al. / Advances in Engineering Software 35 (2004) cognitive science, design activity is a special problem solving activity. The product information usually is imprecise, uncertain and incomplete. Therefore, it is hard to solve design problem by general problem solving methods. Humans have a clear and unequivocal capacity to design. They appear to have the capacity to design at various levels, partly depending on need and depending on the designer. Gero classified design into (1) routine design, (2) non-routine design. Non-routine design is classified into innovative design and creative design [5]. Since, the early years of design automation, a number of computer-based design tools, methods, and methodologies have been developed to support problem solving and facilitate other work in routine design. At the same time, non-routine design has not been given due attention, and it is still poorly automated and provided with little information support. Creativity plays a central role in non-routine design. It is associated with a cognitive process that generates a design solution, which is novel or unusual and satisfies certain requirements. There are many definitions of creativity. In the present study, we have adopted one, based on commonly held beliefs about creativity; creativity is the process that leads to the creation of products that are novel and valuable [6]. Creativity is not a result of a one-shot affair but an outcome of continuous efforts of discovering and evaluating alternatives. In iteratively discovering and evaluating alternatives, a creative individual seeks a balance between usefulness and innovativeness that is necessary for a product to be creative. The product must be novel so that it is not a part of existing well-known solutions. On the other hand, if the product is not useful, or of little value, it cannot be regarded as creative. Following orderly rules leads to a design product that is useful, but not necessarily novel. To transcend the tradition, one needs to take a chaotic approach by breaking rules, which, however, has less chance to produce a useful product. Creativity is a human trait that is not easily converted into a computational tool. It is not realistic to simulate creativity by computational tools, but it is possible to stimulate designer by altering the underlying environment. Rather than to realize the creative design by computer, computer supported design system should be used to help designers to catch sudden inspiration. Thus, creativity could be enhanced by stimulating designers and by allowing them to explore innovative designs more easily Creative idea emerges in a special environment Creativity can occur in a variety of situations, going from artistic situations to situations of technological innovation. However, it is true that sudden inspiration is often stimulated via special media in a special environment. Most of researchers in the field of creativity agree that designers who are engaged in creative design tasks use external resources extensively [7 9]. Such external resources include a variety of physical and logical information, for instance, reading books, browsing photographic images, talking to other people, listening to music, looking at the sea or taking a walk in the mountains. Sketches and other forms of external representations produced in the course of design are also a type of external resources that designers depend on [10]. When designers discover a new or previously hidden association between a certain piece of information and what they want to design, the moment of creative brainwave emerges. Designers then apply the association to their design and produce a creative design. The particular useful information of activating creativity is visual images. Industrial designers, for instance, often have image albums that hold a large number of visual images that they have accumulated over the years. In the early phase of the design process, the designer browses the album to find images that help them generate new ideas. One story was introduced by Kumiyo Nakakoji [11]. While designing a chair, one designer browsed image in his image album seeking for some that would be useful for his design. Although, he did not have a clear goal in mind while browsing, he was vaguely thinking of objects that have the same functionality as a chair. When he saw a picture of flower, the image clicked the moment of creative insight. The round bowl-like shape of a chair emerged from his mind. When he was browsing images in his image album, he already had a vague understanding about his design: such as seat-able, comfortable, nice-looking and the typical shape of a chair, although he has no clear idea about his design. As Fig. 1 indicates, the picture of a flower makes this association between round bowl-like shape and ambiguous adumbration in his mind. This process depends on the designer s ability to discover this association but is stimulated by the image. In product design, visual expression, especially in the form of sketching, is a key activity in the process of originating new product ideas. In the early conceptual stage of the design process, it is typical for an engineer or architect to use various relatively unstructured forms of pictorial representation such as sketches. As the design develops, other more structured forms of pictorial representations, such as plans or sections, become a part of the process. The use of these forms of pictorial representation has long been considered to relate to creativity and innovation in design. Empirical evidence regarding these beliefs is, however, relatively sparse. This applies to both the general question of the role that pictorial representation plays, and the more specific issue of the cognitive processes involved in using such pictorial representations and how they might lead to creative and innovative problem solving.

3 H. Liu et al. / Advances in Engineering Software 35 (2004) Fig. 1. The use of visual images in a creative design process. Larkin and Simon [12] argued that expert reasoning used two forms of representations of a problem. One was sentential or conceptual representation of physical knowledge while the other was imaginal representations in the mind s eye [13] that could then be externally represented in the form of diagrams. Larkin and Simon suggest that such visual forms of representation lead to a more computationally efficient search for information relevant to solving problems because of the 2D, spatial structures of diagrams. That is, the diagrams allow the direct discovery of relevant spatial information for the solution of the problem. While the research on the relationship between imagination and perception was primarily concerned with the question of the functional equivalence between the two, imagination has been seen as essential part of creative problem solving. Imagination as such was not seen as essential to creativity but rather the insights that appeared to be supported by reinterpretations of images that is, creativity was associated with the emergence of new ways of seeing images and this occurred in the mind s eye. In this paper, we do not pay attention to analyze the ability of association and inspiration of human being in design. The purpose of our discussion is to illuminate that the visual representation and environment can indeed push designers generate new ideas and stimulate their design inspiration for creative design. 3. Related works Genetic algorithms are highly parallel mathematical algorithms that transform populations of individual mathematical objects (typically fixed length binary character strings) into new populations using operations patterned after (1) natural genetic operations such as sexual recombination (crossover) and (2) fitness proportionate reproduction (Darwinian survival of the fittest). Genetic algorithms begin with an initial population of individuals (typically randomly generated) and then iteratively (1) evaluate the individuals in the population for fitness with respect to the problem environment and (2) perform genetic operations on various individuals in the population to produce a new population [14]. John Holland presented the pioneering formulation of genetic algorithms and described how the evolutionary process in nature can be applied to artificial systems using the genetic algorithm operating on fixed length character strings in Adaptation in Natural and Artificial Systems [15]. In this work, Holland demonstrated that a wide variety of different problems in adaptive systems (including problems from economics, game theory, pattern recognition, optimization, and artificial intelligence) are susceptible to reformation in genetic terms so that they can potentially be solved by the highly parallel mathematical genetic Fig. 2. The hierarchical structure of a product tree.

4 264 H. Liu et al. / Advances in Engineering Software 35 (2004) Tree-based genetic algorithm Fig. 3. A crossover operation. algorithm that simulates Darwinian evolutionary processes and naturally occurring genetic operations on chromosomes. Genetic algorithm has shown a great potential to work out several real-world problems in the point of optimization, but it is still quite far from realizing a system of matching the human performance, especially in creative applications such as architecture, art, music, and design. The optimization of existing designs is relatively common, with the creation of artistic images and artificial life growing rapidly. However, the creation of new designs seems to be a less common subject for research, with little literature in existence [16]. Some of the work was performed by Professor John Frazer, who spent many years developing evolutionary architecture systems with his students. He showed how evolution could generate many surprising and inspirational architectural forms, and how novel and useful structures could be evolved [17 19]. In Australia, the work of Professor John Gero and colleagues also investigated the use of evolution to generate new architectural forms. This work concentrates on the use of evolution of new floor plans for buildings, showing over many years of research how explorative evolution can create novel floor plans that satisfy many fuzzy constraints and objectives [20]. They even show how evolution can learn to create buildings in the style of well-known architects. Professor Celestino Soddu of Italy uses evolution to generate castles and 3D Picasso sculptures [21]. However, the development of evolutionary design tools is still at its early stage. So far, many genetic algorithms have been used and tested only in design problem solution with small scope. The research and development of design support tools using evolutionary computing technology are still in process and have huge potential for the development of new design technology. Solving a given problem with genetic algorithm starts with specifying a representation of the candidate solutions. Such candidate solutions are seen as phenotypes that can have very complex structures. The expression of standard generic algorithm has solved many problems successfully. However, when applying genetic algorithms to highly complex applications, some problems do arise. The most common is fixed length character strings present difficulties for some problems. For example, mathematical expressions may be arbitrary size and take a variety of forms. Thus, it would not be logical to code them as fixed length binary strings. John Koza, leader in genetic programming, pointed out Representation is a key issue in genetic algorithm work because genetic algorithms directly manipulate the coded representation of the problem and because the representation scheme can severely limit the window by which the system observes its world. Fixed length character strings present difficulties for some problems particularly problems where the desired solution is hierarchical and where the size and shape of the solution is unknown in advance. The structure of the individual mathematical objects that are manipulated by the genetic algorithm can be more complex than the fixed length character strings [22]. The application of a tree representation (and required genetic operators) for using genetic algorithms to generate programs was first described in 1985 by Cramer [23]. Based on Cramer s work, Koza [24] extended the framework by relaxing the fixed length character string restriction. This results in genetic programming, which allows flexible presentation of solutions as hierarchies of different functions in tree-like structures. A natural representation of genetic programming is that of parse trees of formal logical expressions describing a model or procedure. Crossover and mutation operators are adapted so that they work on trees (with varying sizes). In this paper, tree-like presentation presented in genetic programming is adopted and extended. Definition 1. A binary expression tree is a finite set of nodes that either is empty or consists of a root and two disjoint binary trees called the left sub-tree and the right sub-tree. Fig. 4. A mutation operation.

5 H. Liu et al. / Advances in Engineering Software 35 (2004) Fig. 5. A hierarchical structure of a reading lamp. Each node of the tree is either a terminal node (operand) or a primitive functional node (operator). Operands can be either variables or constants. Operator set includes standard operators (þ, 2,*,/,^), basic mathematic functions (such as sqrt ( ), exp( ), log( )), triangle functions (such as sin( ), cos( ), tan( ), asin( ), acos( ), atan( )), hyperbolic functions (such as sinh( ), consh( ), tanh( ), asinh( ), acosh( ), atanh( )) and so on. Here, we use the expression of mathematic functions in MATLAB. Definition 2. Feature F i is a tri-tuples ðf i ID; t i ; v i Þ; where F i ID is the name of feature F i ; t i is the type and v i is the value of feature F i. In which, value is broad sense and can be number, character string, array, function, expression, file and so on. Definition 3. Feature vector FV is defined as a vector FV ¼, F 1 ; F 2 ; ; F n.; where F i is a feature. Definition 4. Feature tree FT is defined as FT ¼ðD; RÞ; where D ¼ {FV i } domain ðfv i Þ (NIL), FV i is a feature vector and is a node on the feature tree, R ¼ {fri} is a set of relations and constraints among the nodes of the feature tree. Definition 5. Product tree PT is defined as PT ¼ (PD,PR), where PD ¼ {FT i } domain ðft i Þ {NIL}, FT i is a feature tree and is a node on the product tree, PR ¼ {pri} is a set of relations and constraints among the nodes of the product tree. From the above definition, we can discover that the expression of a product can be divided into two layers (see Fig. 2) and a multi-branch tree is formed. Genetic operations include crossover, mutation and selection. According to the above definition, the operations are described here. All of these operations take the tree as their operating object. Fig. 6. An example of the curve with three axis points and 19 curve points.

6 266 H. Liu et al. / Advances in Engineering Software 35 (2004) (2) Mutation. The mutation operation is used to enhance the diversity of trees in the new generation thus opening up new areas of solution space. It works by selecting a random node in a single parent and removing the sub-tree below it. A randomly generated sub-tree then replaces the removed sub-tree. A mutation operation is shown as Fig. 4. (3) Selection. For general design, we can get the requirement from designer and transfer it into objective function. Then, the fitness value can be gained by calculating the similar degree between the objective and individual by a formula. However, for creative design, it has no standards to form an objective function. It is hard to calculate the fitness values by a formula. In our system, we use the method of interaction with designer to get fitness values. The range of fitness values is from 1 to 1. After an evolutionary procedure, the fitness values appointed by designer are recorded in the knowledge base for reuse. Next time, when the same situation appears, the system will access them from the knowledge base. Fig. 7. Fitting a mathematical expression to curve points. (1) Crossover. The primary reproductive operation is the crossover operation. The purpose of this is to create two new trees that contain genetic information about the problem solution inherited from two successful parents. A crossover node is randomly selected in each parent tree. The sub-tree below this node in the first parent tree is then swapped with the sub-tree below the crossover node in the other parent, thus creating two new offspring. A crossover operation is shown as Fig. 3. Many explorative systems use human input to help guide evolution. Artists and designers can completely take over the role of fitness function [25,26]. Because evolution is guided by human selectors, the evolutionary algorithm does not have to be complex. Evolution is used more as a continuous novelty generator, not as an optimizer. This method gives the designer the authority to select their favorite designs and thus guide system to be evolved toward the promising designs. Artificial selection can be a useful means for dealing with ill-defined selection criteria, particularly user-centered concerns. For clarity, we will present the performing procedure of the generic algorithm together with a design example in Section 5. Fig. 8. Two parent trees with one crossover node.

7 H. Liu et al. / Advances in Engineering Software 35 (2004) Fig. 9. The results of a crossover operation. 5. A reading lamp figuration design example A reading lamp design example is presented in this section for showing how to use tree-based generic algorithm to generate 2D sketches and 3D images in design process. Fig. 5 shows a hierarchical structure of a reading lamp based on the functional components, which can be classified as: Lamp cover Light Lamp holder Bottom A tree-based genetic algorithm is used on two layers: the first is on the feature layer, and the second is on the component layer. At the feature layer, the execution of genetic algorithm is going to generate some new component shapes while at the second layer the generated outcome will be some afresh combinations of the components. Here, we take lamp holder generation as an example for showing the execution of the genetic algorithm on feature layer. Step 1. Initialize the population of chromosomes. The populations are generated by randomly selecting nodes in the set of operands and the set of operators to form a mathematical expression. We use the stack to check whether such a mathematical expression has properly balanced parentheses. Then, using parsing algorithm, the mathematical expression is read as a string of characters and the binary mathematical expression tree is constructed according to the rules of operator precedence. For extracting the features of successful design from outside, we also generate some chromosomes from the product design database and build mathematical expression trees by the following approach. (a) (b) Create a scanned image file. An image can be brought into the computer using a scanner or a digital camera and is saved as a JPEG image file. Scanners can offer a more satisfying resolution, which will be important if the digitized data must be very accurate. Digital cameras may also be used, although accuracy will be degraded. Open a scanned image file. Selecting Open command from the menu and shows the scanned image on the screen. The color of the scanned image will be filtered. Fig. 10. The lamp holders correspond to generated curves in Fig. 9.

8 268 H. Liu et al. / Advances in Engineering Software 35 (2004) Fig. 11. One parent tree and a sub-tree. (c) (d) (e) Create two or three axis points. The program uses axis points to define the coordinates of the scanned images. If the scanned image has the same scale in both the horizontal and vertical directions, then, only two axis points are needed. Once the axis points are defined, the status bar automatically shows the graph coordinates of the cursor as it is moved around. In addition, grid lines can then be shown. Create points for the curve. There may be one, two or more curves in a worksheet. Selected curve is described using at least two points (see Fig. 6). Fit a mathematical expression to curve points. Choose a mathematical expression template, edit the equation and adjust the equation coefficients to improve the fitness through the points (see Fig. 7). (f) Use parsing algorithm, the mathematical expression is read and a binary mathematical expression tree is constructed according to the rules of operator precedence. Step 2. Get the fitness for each individual of the population via interaction with designer. The population with high fitness will be shown in 3D form first. Designers can change the fitness value after they see the 3D images. Step 3. Form a new population according to each individual s fitness. Step 4. Perform crossover and mutation operations on the population. Fig. 8 shows two binary mathematical expression trees. Their mathematical expressions are (1.2þsin(8x))x^2cos(x) and x(1 2 x)(1.5 þ (cos(8x)), respectively. Fig. 12. The result of a mutation operation.

9 H. Liu et al. / Advances in Engineering Software 35 (2004) Fig. 13. A reading lamp tree with one crossover point A. (1) Crossover operation. A crossover node is randomly selected in each parent tree. The sub-tree below this node on one parent tree is then swapped with the subtree below the crossover node on the other parent, thus generating two new offspring. If the new tree cannot pass the syntax check or its mathematical expression cannot form a normal curve, it will die. Taking the two trees in Fig. 8 as parent, after the crossover operations by nodes A, we get a pair of children (see Fig. 9). Fig. 10 shows generated lamp holders in 3D form correspond to generated curves in Fig. 9. (2) Mutation operation. The mutation operation works by selecting a random node in a single parent and removing the sub-tree below it. A randomly generated sub-tree then replaces the removed sub-tree. The offspring will die if it cannot pass the syntax check or it cannot form a normal curve. One parent tree and a sub-tree are shown in Fig. 11. After a mutation operation, a generated child sketch is shown as the right side in Fig. 12. Step 5. If the procedure is not stopped by the designer, go to step 2. This process of selection and crossover, with infrequent mutation, continues for several generations until it is stopped by the designers. Then the amending design will be done by designers with human wisdom. Fig. 14. A crossover operation on a cover shape feature.

10 270 H. Liu et al. / Advances in Engineering Software 35 (2004) Fig. 15. Some generated reading lamps. Next phase, the similar operations are performed on product tree. Here, we only show a crossover operation of the second phase. Fig. 13 is a reading lamp product tree. After a crossover operation on node A (two parent use the same crossover point-cover shape feature), we get the outcome in Fig. 14. When crossover operations happened on different feature nodes, children will change their related features (shape, size or color). Generally, these operations cannot produce surprised result because they are the recombination of existing components and are constrained by many factors. The mutation operator works by selecting a random node in a single parent and removing the sub-tree below it. In general, a new sub-tree will be gotten from outsideother design group or a public component base [27] and then it replaces the removed sub-tree. Designers embellish the generated images by using computer operations, such as rotating, rending, lighting, coloring and so on. Then, we can get some reading lamp images as shown in Fig Conclusions With this insight into enabling creativity by evolution, we create a framework for explorative supporting creative design by evolutionary computing technology. For only a part of generated mathematical expressions can be expressed by curves and generate useful shapes, generated shapes in this system are relative simple and limited. Although looking simple, the framework employs a feasible and useful approach in a visual evolutionary computing environment. This environment is used to stimulate the imagination of designers and activate their eye in mind. It will give the designers concrete help for extending their design spaces. The work described in this paper is a part of the continuing project done by the Design Technology Research Centre (DTRC) in the School of Design at the Hong Kong Polytechnic University [28]. There is still much work to be done before the full potential power of the system can be realized. Our current work is the development of an integrated computer-aided design environment. Evolutionary computation, artificial intelligence, integrated and interactive system techniques, and virtual reality are employed for the implementation of this environment. Acknowledgements This project is funded by the Research Fellow Matching Fund Scheme 2001 (No. G.YY.34, No. G.YY.35) of the Hong Kong Polytechnic University, and supported by National Natural Science Foundation of China (No and No ) and Natural Science Foundation of Shandong Province (No. Y2003G01).

11 H. Liu et al. / Advances in Engineering Software 35 (2004) References [1] Magrab EB. Integrated product and process design and development, the product realization process. Boca Raton: CRC Press; [2] Roy R, Riedel J. Design and innovation in successful product competition. Technovation 1997;17(10): [3] McMeekin A, Green K, Tomlinson M, Walsh V, Innovation by demand: an interdisciplinary approach to the study of demand and its role in innovation, Manchester: Manchester University Press; [4] Walsh V, Roy R, Bruce M, Potter S. Winning by design: technology, product design and international competitiveness. Oxford: Blackwell; [5] Gero JS. Computational models of innovative and creative design processes. Technol Forecast Social Change 2000;64(2 3): [6] Akin Ö, Akin C. On the process of creativity in puzzles, inventions, and designs. Autom Constr 1998;7(2 3): [7] Dartnall T, editor. Artificial intelligence and creativity. Dordrecht: Kluwer Academic Publisher; [8] Gero JS, Maher ML, editors. Creativity and knowledge-based creative design. Hillsdale, NJ: Lawrence Erlbaum Associations Inc.; [9] Hill W, Stead L, Rosenstein M, Furnas G. Recommending and evaluating choices in a virtual community of use, human factors in computing systems. CHI 95 Conference Proceedings (Denver, CO), New York: ACM; p [10] Won PH. The comparison between visual thinking using computer and conventional media in the concept generation stages of design. Autom Constr 2001;10(3): [11] Kumiyo N, Yasuhiro Y, Masao O. Computational support for collective creativity. Knowledge-Based Syst 2000;13(7 8): [12] Larkin JH, Simon HA. Why a diagram is (sometimes) worth words. Cognit Sci 1987;11: [13] Kosslyn SM, Koenig O. Wet mind: the new cognitive neuroscience. New York: The Free Press; [14] Koza JR. Genetically breeding populations of computer programs to solve problems in artificial intelligence. In: Proceedings of the Second International Conference on Tools for AI, Herndon, Virginia, November 6 9, 1990, Los Alamitos, CA: IEEE Computer Society Press; p [15] Holland JH, Adaptation in natural and artificial systems, Ann Arbor, MI: University of Michigan Press; [16] Bentley PJ. Generic evolutionary design of solid objects using a genetic algorithm. PhD Thesis, Division of computing and control systems. Department of Engineering, University of Huddersfield; [17] Frazer JH. An evolutionary architecture. London: Architectural Association Publications; [18] Frazer JH, Frazer JM, Liu XY, Tang MX, Janssen P. Generative and evolutionary techniques for building envelope design. In: Proceedings of Generative Art 2002, Fifth International Conference GA2002, Milan, December 2002; [19] Liu H, Tang MX, Frazer JH. Supporting evolution in a multi-agent cooperative design environment. J Adv Eng Software 2002;33(6): [20] Gero JS, Kazakov V. An exploration-based evolutionary model of generative design process. Microcomput Civ Eng 1996;11: [21] Soddu C. Recreating the city s identity with a morphogenetic urban design. 17th International Conference on Making Cities Livable, Freiburn-im-Bresgau, Germany; p [22] Koza JR. Evolution and co-evolution of computer programs to control independent-acting agents. In: Meyer J-A, Wilson SW, editors. From Animals to Animats: Proceedings of the First International Conference on Simulation of Adaptive Behavior, Paris, September 24 28, Cambridge, MA: The MIT Press; p [23] Cramer NL. A representation for the adaptive generation of simple sequential programs. In: Proceedings of an International Conference on Genetic Algorithms and their Applications, Carnegie-Mellon University; p [24] Koza JR. Genetic programming: on the programming of computers by means of natural evolution. Cambridge, MA: MIT Press; [25] Kim HS, Cho SB. Application of interactive genetic algorithm to fashion design. Eng Applic Artif Intell 2000;13(6): [26] Todd L. The mutation and growth of art by computers. In: Bentley PJ, editor. Evolutionary design by computers. San Francisco, CA: Morgan Kaufman Publishers Inc.; [27] Bentley PJ. Aspects of evolutionary design by computers. In: Roy R, Furuhashi T, Chawdhry PK, editors. Advances in Soft Computing. London, UK: Springer; p [28] Liu H, Tang MX, Frazer JH. Supporting learning in a shared design environment. J Adv Eng Software 2001;32(4):