1

Interactive Evolutionary Computation: Fusion ofthe Capabilities of EC Optimization and Human

EvaluationHideyuki TAKAGI

Dept. of Art and Information Design, Kyushu Institute of Design:

4-9-1, Shiobaru, Minami-ku, Fukuoka 815-8540, Japan

Tel&Fax: +81-92-553-4555, E-mail: takagi@kyushu-id.ac.jp, URL: http:// www.kyushu-id.ac.jp/ ∼takagi/

Abstract— We overview the research on interactive evolu-tionary computation (IEC). The IEC is an EC that optimizessystems based on human subjective evaluation. First, thedefinition and features of the IEC are described. Then, theoverview of the IEC research follows. The overview consistsof two parts: application and interface researches. The IECapplication fields in this survey include graphic art and ani-mation, 3-D CG lighting, music, editorial design, industrialdesign, face image generation, speech processing and syn-thesis, hearing aid fitting, virtual reality, media database re-trieval, data mining, image processing, control and robotics,food industry, geophysics, education, entertainment, and soon. The interface research to reduce human fatigue in thissurvey includes interfaces of fitness input and display basedon fitness prediction, accelerating EC convergence especiallyin early EC generations, combinations of interactive andnormal ECs, and active user intervention. Finally, we dis-cuss the IEC from the point of the future research directionof computational intelligence. The biggest feature of thispaper is to wholly survey them with 200 IEC papers ratherthan to carefully select representative papers to show thestatus quo of the IEC researches.

I. Introduction

There are two types of target systems for system op-timization: systems whose optimization performances arenumerically or at least quantitatively defined as evaluationfunctions and systems whose optimization indexes are hardto be specified. Most of engineering researches use severaloptimization methods based on minimizing error criteriaand focus on the former, which includes auto-control, pat-tern recognition, engineering design, and so on.However, consider to obtain the most favorite outputs

from human-machine interaction systems that create or re-trieve graphics or music, for example; such outputs mustbe subjectively evaluated, and it is hard or even impos-sible to design evaluation functions. Generally speaking,the best outputs of the systems that human five senses di-rectly accept their outputs such as images, acoustic sounds,and virtual reality are evaluated by only impression, pref-erence, emotion or intelligibility in users’ mind. There aremany such systems in not only artistic or aesthetic fieldsbut also engineering and education fields as shown in thispaper. To optimize such systems, system parameters orstructures must be optimized based on user’s subjectiveevaluation. Since we cannot use gradient information ofour psychological space in mind, we need another approachthat is different from conventional optimization methods.

Interactive Evolutionary Computation (IEC) is an op-timization method that adopts evolutionary computation(EC) among system optimization based on human subjec-tive evaluation. Simply speaking, it is an EC techniquewhose fitness function is replaced with a human user. Fig-ure 1 shows a general IEC system where a user sees orhears and evaluates system outputs and the EC optimizesthe target system to obtain preferred output based on theuser’s evaluation. In this sense, we can say that the IEC is atechnology that embeds human preference, intuition, emo-tion, psychological aspects, or more general term, KANSEIin to the target system.

t a

r

g

e

t

s

y

s

t

e

mf

(

p

1

,

p

2

,

.

.

.

,

p

n

)

s u

b

j

e

c

t

i

v

e

e

v

a

l

u

a

t

i

o

ni

n

t

e

r

a

c

t

i

v

e

E

C

m

y

g

o

a

l

i

s

.

.

.

s y

s

t

e

m

o

u

t

p

u

t

Fig. 1. General IEC system: system optimizationbased on subjectiveevaluation.

There are two definitions of IEC; narrow definition is“technology that EC optimizes the target systems basedon human subjective evaluation as fitness values for thesystem outputs,” and wide definition is “technology thatEC optimizes the target systems having an interface ofhuman-machine interaction.” Previous definition, “an ECtechnique whose fitness function is replaced with humanuser” is of the narrow one. We survey only IEC papers ofthe narrow definition in this paper because most of pastpapers that appeal their IEC aspect are categorized in thenarrow definition, though there are some exceptions [36],[37], [59], [104], and there are so many EC-based human-machine interaction papers that do not explicitly appealtheir IEC aspect. However, note that the discussion on hu-man interface in section IV is commonly useful for the IECresearches of wide definition.Finally, we explain terms used in this paper. EC is a

biologically inspired general computational concept andincludes genetic algorithms (GA), evolutionary strategy(ES), genetic programming (GP), and evolutionary pro-

gramming (EP). We use the term of interactive GA, ES,GP, and EP (IGA, IES, IGP, and IEP) to individually ex-plain each IEC research in this paper. The subjective eval-uation of the IEC is generally in n levels of rating, and theminimum evaluation scale level is two levels that means toselect and not select EC individuals. This type of the IECis called simulated breeding or user selection because of theanalogy with the selection of artificial mating. The term ofindividual is also frequently used in this paper. The EC isa population-based searching algorithm and outputs multi-ple candidate of system outputs. Each of them is called anindividual. See EC textbooks for other terms and conceptin detail.The technical framework of the IEC is described in sec-

tion II, and IEC survey follows in sections III and IV. Thebiggest feature of this paper is to wholly survey IEC re-searches as much as possible rather than to carefully selectrepresentative papers. This survey includes not only IECapplications but also IEC interface researches, which is an-other feature of this paper. Finally, we discuss the futureIEC researches from the point of computational intelligenceresearch.

II. Technical Frameworks and Features of IEC

IEC is a technology that human and EC cooperativelyoptimize target systems based on the mapping relation be-tween physical and psychological spaces. IEC users evalu-ate individuals according to the distance between the targetin their psychological spaces and the actual system outputs,and the EC searches the global optimum in a feature pa-rameter space according to the psychological distance (seeFigure 2).

t h

e

i

m

p

r

e

s

s

i

o

n

o

f

s

y

s

t

e

m

o

u

t

p

u

t

p1pn

p

a

r

a

m

e

t

e

r

v

e

c

t

o

r

o

f

t

h

e

t

a

r

g

e

t

s

y

s

t

e

m

p a

r

a

m

e

t

e

r

s

p

a

c

e

dt h

e

i

m

p

r

e

s

s

i

o

n

o

f

g

o

a

l

i

n

m

i

n

d

p s

y

c

h

o

l

o

g

i

c

a

l

s

p

a

c

e

f a

c

t

o

r

1

f a

c

t

o

r

2

f a

c

t

o

r

m

d

s u

b

j

e

c

t

i

v

e

e

v

a

l

u

a

t

i

o

n

Fig. 2. Psychological distance between target in our psychologicalspaces and actual system outputs becomes the fitness axis of afeature parameter space where EC searches the global optimumin an IEC system.

Conventional approaches for these human evaluationbased systems have frequently modeled the human eval-uation characteristics and embedded the substitute eval-uation model in optimization systems. Let this approachbe called analytical approach. The analytical approach iscommon approach in AI research, but it is hard to perfectlymake, for example, a personal preference model. On theother hand, the IEC is based on the synthetic approachthat directly embeds a human as a black box evaluator inthe optimization system and let a computer optimize thetarget system according to his or her evaluation [195], [208],

[210].

The IEC users sometimes do not evaluate phenotypesof EC individuals but system outputs specified by the ECindividuals. For example, they do directly evaluate filtercoefficients but the images or sounds processed by the fil-ters.

We cannot avoid the fluctuation of human preferencewhich is the target coordinate in a psychological space andevaluation which is a distance measure on the psychologicalspace. The requirement to the IEC is to find out the coordi-nates in a feature parameter space that are mapped to theneighborhood of the target in a psychological space evenif human’s subjective evaluation for same image or soundfluctuates according to time. Fortunately, it is said that ECsearch is robust for noise, and it is also reported that therewas few influence of the fluctuation through simulation us-ing actually measured human fluctuation characteristics ofsubjective evaluation [191], [193], [139].

The main reason is that the global optimum of the IECis very rough, because every system outputs that a hu-man user cannot distinguish are psychologically same. Theglobal optimum of the IEC is not a point but rather anarea from the normal EC optimization point of view. Moreactively speaking, wide IEC global optimum area is prefer-able. For example, suppose to find the next model of Toy-ota Camry or Honda Accord. The purpose is not to find outonly one point like as montage face image of a suspect butto find out several different car models whose impressionsare common to own impression of the past car models.

Remained IEC technical problems are to solve humanfatigue problem that is common to all human-machine in-teraction systems and to accelerate EC convergence witha small population size and a few generation numbers thatare IEC own problem and deeply related with the fatigueproblem. The EC population size is very limited by thenumber of individual images that are spatially displayedon a computer monitor at once or the human memory ca-pacity for time-sequentially displayed individual sounds ormovies. The number of EC search generation is limited byhuman fatigue as well, and 10 or 20 EC search generationsare maximum for normal use of the IEC search. Several re-searches to solve these problems are introduced in sectionIV in detail.

It is important not only to expand the IEC applicationfields but also to evaluate their effectiveness in each fieldand develop several IEC interfaces resulting less fatiguefor the practical use of the IEC. Since the IEC handles ahuman, subjective tests and statistical tests are essentialto evaluate its effectiveness [149]. We cannot expect todevelop a practical IEC technology without objective andquantitative evaluation even in the field of artistic creativ-ity that apparently seems far from quantitative evaluation.This is an author’s special remark to call IEC researchers’attention.

2

TABLE I

Statics of IEC papers by field and year. Papers that discuss more than one topic are counted in each corresponding

category.

1980s 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 total

graphic art & CG animation 2 3 2 4 5 5 2 2 4 8 3 403-D CG lighting design 1 3 1 5music 1 3 3 1 1 3 3 15editorial design 1 1 2 4industrial design 2 2 1 5 4 2 4 9 29face image generation 1 1 1 2 1 4 5 1 16speech processing & prosodic control 2 1 2 1 1 7hearing aids fitting 2 7 5 14virtual reality 1 1 2database retrieval 2 1 8 8 1 20knowledge acquisition & data mining 5 3 3 1 4 16image processing 1 2 3control & robotics 1 2 3 4 2 12food industry 1 1 2geophysics 1 2 3art education 2 2writing education 1 3 4games and therapy 1 1 1 3

soxial system 1 1

discrete fitness value input method 5 2 7prediction of fitness values 1 2 1 8 3 1 16interface for dynamic tasks 1 1 2 4acceleration of EC convergence 1 1 3 1 7combination of IEC and non-IEC 1 2 3active intervention 1 3 2 6

total 2 0 5 5 8 11 23 28 22 47 54 36 241

III. IEC Applications

A. IEC Research in 1990s.

The IEC research has been conducted in 1990s though itoriginates from the work of Dawkins in 1986 [30]. Staticsof IEC papers is shown in Table I. Roughly speaking, thereare two major research streams in early and after 1990s.

One is in Artificial Life, and researchers and artists wereinterested in self-grown computer graphics (CG) and musicaccording to increasing interest in the Artificial Life. Themajor IEC researches in early and mid 1990s are on thisline, especially researches on artistic image creation [12].Practical application field, industrial design, forms a partof this stream.

The second one comes from the increase of researcherswho are interested in humanized technology or human re-lated systems, and these researchers have applied the IECto engineering and other fields. Another feature of this lineis that the research on IEC interface and human fatigueincreased as well as application research. It is one featurethat recent IEC researches spread over practical fields suchas engineering or edutainment, while the early stage of theIEC research was biased in artistic applications [195], [196],[198], [199], [205], [208].

Main objective of this paper is to overview whole the IECresearches. This section surveys several IEC applicationsby sorting their fields in the order of artistic, engineering,and edutainment fields.

B. Graphic Art and CG Animation

First IEC research was the biomorph of Dawkins whosetheory of evolution, selfish gene, upset the tranquility of theSociety [30], [31]. He used L-system that mathematicallyexpresses recursive development process of plants or other,iterated subjective selection of the L-system outputs andmutation of genes that express the number and angles ofL-system branches, and created several insect-like 2-D CGforms.Many applications to creating CG followed the

biomorph; they are similar biomorph [182], plant CG basedon L-system [112], [137] (the main topic of the reference[137] is to avoid user interaction by trying to make a fitnessfunction for aesthetic plan image though it handles IEC.) ,2-D CG based on mathematical equations or cellar automa-ton rules [179], [180], [224], [181], [10], [11], [105], [236], [47],[228], [1], [229], [230], [167], [231], [240], [232], 3-D CG ren-dering [223], animal CG [44], and airplane drawing evolvedby wings and bodies [125], [126]. Math equation-based IECuses several methods to create CG: CG creation using non-linear math equations generated by GP, liner and nonlinearfractal transformation equations within the framework ofan iterated function system, and given differential and dy-namical equations. The parameters or structure of theseequations are modified by IEC.Artistic system, Mutator, has been applied to the wide

applications of animation, cartoon face drawing, commer-cial 3-D screen saver of Computer Artwork Ltd., music,and financial planning bedsides 2-D and 3-D CG [225].An aquarium that is scheduled to open in Nagasaki,

3

Japan in April, 2001 will have a virtual aquarium as anannex. One of the virtual aquarium projects is to let visi-tors who have no CG experience create their own 3-D fishCG using IGA and release the fishes in the virtual realityspace [79], [66]. The shape of the 3-D fish is expressed bymath equations and the parameters of the equations aremodified by the IGA. The project also includes to developthe internet version of IEC interface and allow visitors tocreate their own fishes through the internet before theyvisit the aquarium.The IEC was applied to generate 3-D shapes and tex-

ture that are expressed with Fourier transform in a virtualreality space [102]. The IGP was used to create 3-D CGand music source localization in a CAVE for virtual realitythough this research is mainly for virtual reality researchrather than artistic creation [29], [156], [32].One of most active IEC graphic artists would be Sims1.

He used GP, evaluated CG created by the math equation,evolved the math equation by GP, and created fine CGart. The equations are used to calculate each pixel bypixel [179], [180] or create graphic movies by adding timevariable in the dynamic differential equations [180], [181].He created several graphic art including Panspermia andPrimordial Dance in 1991 and 1993 and also let visitorsinteract with his interactive art system at some art showsand exhibitions. His Galapagos is an L-system-based IECsystem that runs on 16 graphic workstations and a paral-lel machine, allows visitor create graphic art through theirinteraction, and is exhibited at multimedia museum, ICC(NTT InterCommunication Center), in Tokyo.Let’s see how to create IEC graphics using public-opened

SBART2 [228], [229], [230], [231], [232] as an example IECart system3. The SBART creates tree structures of mathequations using GP and calculates values at each pixel us-ing the math equation and (x, y) coordinate values of thepixel. The SBART assigns arithmetic operators such asthe four fundamental operators of arithmetic, a power op-erator, √, sin, cos, log, exp, min, and max to its non-terminated nodes and constant and variables (XY0 in Fig-ure 3) to its terminator nodes. Three values at each pixelare calculated using one generated math equation by as-suming that the constants are 3-D vectors consisting ofthree real numbers and variables are 3-D variables consist-ing of (x,y,0). The three calculated values are regarded asmembers of a (hue, lightness, saturation) vector and aretransformed to RGB values at each pixel. These three val-ues are normalized into [−1, 1] using the following saw-likefunction:

f(x) ={

x − 4m (4m − 1) ≤ x ≤ (4m + 1)−x+ 4m+ 2 (4m+ 1) ≤ x ≤ (4m + 3)

where m is an integer. Graphic movie is created by replac-ing the constant (x,y,0) with (x,y,t), where t is a time vari-able. Also, the functions of the SBART was expanded to

1See his works at http://genarts.com/karl/.2SBART is downloadable from URL:

http://www.intlab.soka.ac.jp/˜unemi3Reference [168] introduces 14 IEC-based 2-D CG, 3-D CG, and

music package with their URLs.

Y

X

0 X

Y

0

p

o

w a

b

s

c o

s

m

i

n

- 0

.

7

5

0

*

/

+

-

0.297

m

i

n

c o

s

c o

s

- 0

.

7

5

0

Y

X

0

Y

X

0 X

Y

0hypot

Fig. 3. SBART: simulated breeding-based 2-D CG system. GPevolves math equation that is applied to each pixel, and graphicscreated by the equations are evolutionary obtained.

create collage [231], [232]. A human user selects preferred2-D images from 20 displayed images at each generationand IGP creates next 20 offsprings. This operations areiterated until the user obtains a satisfactory image.

An interesting EC operator was proposed and used tocreated deformed offspring images [45]. Selected parentbitmap images are lined up and regarded as time sequentialsignal, and each vertex or pixel point of offspring image iscalculated from the time sequential points in time sequen-tial parent images by applying an auto-regressive model.

Besides previously mentioned Panspermia, PrimordialDance, and SBART, Mutator [223] and some works [240]at International Interactive Genetic Movie, the third web-based exhibition, are graphic movies that add a time pa-rameter to still image coordinates (x, y). Fractal movieswas created by the IEP using affine transformation thatlinearly maps a coordinate to another [1]. As the affinetransformation creates only a still image, meta-level trans-formation that transforms a set of affine transformationsto another fractal set of affine transformations was definedand used to create a graphic movie. The coefficient matrixof the affine transformation is renewed by the IEP, and auser rates 10 movies at each generation.

The IEC was used to create not only graphic movies butalso animation of concrete objects. Comical movement of adeformed line body was created by the interaction betweenan animator and the EC [233], [234]; walking motion tra-jectory of a line body was created by finding suitable com-bination of joint angles of arms and legs using the IEC[103]; animated pass-motions of two arms was generatedby combining a multi-objective GA and IGA [176], [177].

4

g

l

o

o

m

y

e

x

p

r

e

s

s

i

o

n

c h

e

e

r

f

u

l

e

x

p

r

e

s

s

i

o

n

h

e

r

o

i

n

e

e

x

p

r

e

s

s

i

o

ni

n

a

m

o

v

i

e

p

a

m

p

h

l

e

t

v i

l

l

a

i

n

e

x

p

r

e

s

s

i

o

n

i

n

a

m

o

v

i

e

p

a

m

p

h

l

e

t

b

y

h

a

n

d b

y

I

E

C

Fig. 4. 3-D CG lighting arts whose design motif is ”debut of a villainactress” made by an amateur with manual (left) and with IGA(right).

C. 3-D CG Lighting Design

Since 3-D CG is a simulation of a photograph, CG im-pression is dramatically changed by lighting conditionssuch as positions, lightness, or colors of lights as well asphotographs. Most of normal people can tell good CGlighting from bad ones, nevertheless they do not have CGlighting skill to realize the intended impression in mind.The role of lighting design support that deeply influencesthe CG quality have gradually become important nowa-days as CG have spread from professional to business andhobby areas. The IGA was applied to develop a 3-D CGlighting design support system [3], [5], [4], [6], [7].IGA-based and manual-based CG lighting designs were

compared. Experimental subjects were requested to adjustthe impression of CG woman by using 3 lights with 2 typesof an infinite and omnidirectional light, 16 levels of inten-sity, on/off, and coordinates in a 3-D space. Figure 4 showsthe comparison.Subjective tests and statistical tests for the comparison

showed that the IGA is not significantly useful for expe-rienced CG designers but for CG designers with little orno experience. Experienced designers’ explicit intentionsshortens their manual lighting design times, while ama-teurs’ designs, often accomplished by trial and error, takesmore time than that of the IGA.

D. Music

IEC was applied to generate melody [123], [124], [13],[16], [17], [18], [19], [20], [152] and rhythm of percussionpart [53].Sonomorphs is the first IEC-based music system, which

was motivated by the biomorph of Dawkins as you see from

G

e

n

J

a

m

B

a

s

s P

i

a

n

o D

r

u

m

P

r

o

g

r

e

s

s

i

o

n

p

h

r

a

s

ep

o

p

u

l

a

t

i

o

nm

e

a

s

u

r

e

p

o

p

u

l

a

t

i

o

n

M

I

D

I

s

e

q

u

e

n

c

e

g o

o

d

/

b

a

d

s o

l

o

Fig. 5. GenJam system that interactively generates jazz melody.

its naming, and it generates measure to phrase lengthsof melodies based on user-based selection [123], [124].Sonomorphs runs with MAX that is a popular object-oriented computer music language.Figure 5 is the IGA-based melody generator, GenJam,

that inputs rhythm sections and cord progress and gener-ates jazz melody. A user listens and evaluates the generatedmelody by tapping a good or bad key from moment to mo-ment during the melody is played. Audio CD made by theGenJam was put on the market [14]; the performance thatthe majority decision of audience was fed-back to the Gen-Jam to generate jazz music was conduced [15]; performerscan play with the GenJam real-timely [17], [18].It takes too much time and causes too much human fa-

tigue if a user gives one fitness after one melody is playedout. The GenJam allows a user to evaluate the gener-ated melody measure by measure instead of whole melodyby melody to avoid this problem [13]. This method thatshortens IEC time would be useful for other IEC tasks thathandle time-sequential signal.Vox Populi is a composition system that allows a user

to real-timely change the parameters of a fitness functioninstead of giving subjective fitness values [115]. A fitnessfunction that judges how each melody from soprano to basssatisfies each voice part range and harmony degree is setin the system. A user listens the melody generated by GAand controls the parameters of the fitness function throughfive user interfaces.Although it is not music, the IEC was applied to control

the parameters of FM sound synthesizer to realize a tonein mind [65].

E. Editorial Design

The IEC was applied to design HTML style sheets [114],[213]. As the internet advances and expands from businessto home, the number of people who open their private webpage as hobby is increasing. Many visual HTML editorsare found on the commercial market and at free & share-ware software sites, and designing homepage is nowadaysadopted at many elementary and junior high schools. Theimpression of web pages very depends on the combinationof background color, font type, size, and color, font of ti-tle level, link color, and so on. The IEC is a good tool tovisually optimize these combination.The IEC was applied to design a poster by optimizing the

combination of its layout, font size, and color [113], [136].A user expresses the impression of a poster in mind withthe degrees of 12 impression words after he or she inputs

5

sentences and specifies image file names used in the poster.Fuzzy logic rules describing layout knowledge of expertsin their system estimate the impressions of posters gener-ated by the GA with the degrees of 12 impression words.The distance between user’s and system’s impression de-gree vectors was used as a fitness value, and offspring posterlayouts are generated [113]. Same design is conducted forfont and color for a color poster design as well [136]. Theirfuzzy logic system has a learning function to absorb thepersonal difference of impression. The HTML style sheetcolor design of [213] is based on this system.As well as systems in section III-K, it is happy if the best

poster is obtained at once, otherwise the user iterates theEC search with changing the coordinate of target impres-sion on a psychological space, which corresponds the ECsearch in next generation.

F. Industrial Design

As we see that most of industrial design is designed usingCAD systems, it is easy for us to understand that industrialdesign is a good application task of IEC because industrialdesign is an expression form of design parameters such aslengths, angles, coordinates, color numbers and so on. Sev-eral IEC applications for industrial design includes car de-sign [72], [42], [43], fashion design [120], [121], [122], [74],[28], [75], [76], [77], [78], color design for knitwears or graph-ics [34], and shape development for new tools, such as scis-sors, saws, and any general 3-D object shapes [170], [171],[175], [87], [88]. There is a formation based 3-D form de-sign approach that encodes five transformation commandsof insert, delete, fold, lift, and pole-hole as genotype codeand interactively creates new 3-D forms [135].The IEC was also applied to civil engineering and archi-

tectural design. The designs of concrete arch dam [157],suspension bridges [40], houses [43], [46] are some of itsapplications. Graf et al.’s system displays bitmap imagesgenerated by warping and morphing operators that inter-polate the vertices and points on two parent bitmap imagesonto an offspring bitmap image [42], [43], [46]. As they usephoto images as initial EC parents, their car or architec-ture design has reality. This system seems to be widelyuseful for general design in industrial, architectural, andcivil engineering designs.Positional design of HyperCard on a screen and general

layout design being applicable to GUI design [109], [110],[111], [55] and floor planning [183], [35] can be categorizedin this section. Masui’s layout design makes a fitness func-tion adapt to user’s preference by teaching GP which lay-outs are good or bad [111].The IEC was applied to layout of the interior furniture.

The furniture layout may depend on personal preference,functional relations of furniture positions, the purpose ofthe room, or even FengShui very frequently used in Chinesesociety. An IGA-based interior layout system that the GAoptimizes the positions of furniture with fitness values con-sisting of objective evaluation based on expert knowledgeand subjective evaluation based on user’s preference waspresented [85].

G. Face Image Generation

Montage systems are the systems that combine partialimages of face photos and compose a face image. The mon-tage system used at Fukuoka Prefectural Police in author’sresident area combines the partial images of 50 people facesto compose criminal suspect’s face. Clear memory of par-tial face feature and explicit indication of partial face imageare necessary for conventional montage systems, but it isvery hard for general witnesses to memorize a suspect’sface even in detail. IEC-based montage systems optimizethe combination of partial face images and generate a faceimage based on witness’s total impression of a suspect inmind rather than memory of each facial part [25], [80], [81],[203]. While the montage systems use 2-D photos as orig-inal images, 3-D face expression whose data are obtainedby a 3-D digitizer was created using the IGA [64].Generating faces of line drawing was frequently used as

a research task of the IEC [8], [9], [118], [119], [130], [131],[184], [185], [54], [186]. One of such researches is auto-fitness assignment to accelerate EC convergence by esti-mating fitness values using Euclidian distance from indi-viduals [118], [119] and positional relationship with user-selected individuals [130], [131], [184], [185], [54], [186] (c.f.section IV-C).The expression of face photo image can be changed by

deformation that the pixel positions of face parts such aseyes, eyebrows, or mouse area are continuously mapped toother pixel positions. The IEC was used to generate thisface expression [101].

H. Speech Processing and Prosodic Control

The IEC was used to design a digital filter that audiblyreduces speech distortion. Concretely speaking, the GAoptimizes eight coefficients of FIR distortion reduction fil-ters [238], [239] or amplification levels of each frequencyband [226], [227] for distorted input speech whose powerin low frequency range where voice formants exit is sup-pressed, and a user hears the recovered speech, evaluate itsquality, and feeds-back his or her subjective fitness.The result of the subjective tests showed that the quality

of the recovered speech was significantly better than orig-inal distorted speech not only for the IGA users but alsofor other subjects as well [238], [239].The IEC can be also applied for speech synthesis. Speech

has phonetic and prosodic information, and voice impres-sion and naturality are controlled by prosodic parametersof pitch, amplitude, duration, and speed. Linguists havetried to make rules of the relationship between the impres-sion or naturality and the prosodic parameters in conven-tional speech synthesis research. Prosodic control is a goodtask of the IEC because the quality or naturality of syn-thesized speech is evaluated by only humans. The IEC wasused to modify prosodic parameters to change voice im-pressions to reflect five impressions such as peace, anger,or joy [172], [173], [174].

6

C

o

n

v

e

n

t

i

o

n

a

l P

r

o

p

o

s

e

d

h

e

a

r

i

n

g

a

i

d

EC

h

e

a

r

i

n

g

a

i

d

Fig. 6. Conventional hearing aid fitting (left) and IEC Fitting (right).

I. Hearing Aids Fitting

Seeing an aging society come, digital hearing aids us-ing digital signal processing have spread and solved severalproblems that conventional analog type of hearing aids can-not overcome. However, hearing aid fitting that maximizesthe performance of hearing aids to a wide variety of usershas not yet established as well as natural sound quality.The essential reason is that only the user can evaluate thehearing quality; no one can perceive how other personshear.

The IEC has potential for automatically tuning the pa-rameters of the signal processing according to the user’shearing perception, so that IEC Fitting was developed (seeFigure 6 ) [138], [143], [200], [201], [209]. This approachis essentially different from those of conventional hearingaid fitting and auto-fitting, and it tunes the hearing aidparameters based on how each of hearing impaired personshears without previously measured their hearing charac-teristics, basically. So far, hearing aid engineers or otolo-gists measure the hearing characteristics, such as with anaudiogram, and adjust the hearing aid parameters to com-pensate the difference of the characteristics between theimpaired person and normal hearing persons. The prob-lems of this conventional approach are: sound source formeasurement is limited only pure tone or band pass noise,measurable hearing characteristics are just few parts of ahuman auditory system, full measurement takes long time,and the fitting location is limited to only where the fittingexpert works. On the contrary, the IEC Fitting can opti-mize hearing aid parameters based on user’s comprehensivehearing from auditory peripheral to central nerves with anydaily sound at anywhere without previous measurement ofthe hearing characteristics basically. Thanks to its soundsource free characteristics, the IEC Fitting was evaluatedwith speech, speech with multi-talker noise, and music andshown that its users’ satisfaction and intelligibility werehigher than those of conventional fitting [144], [145], [146],[201], [204], [148]. Comparison of two fitting methods hasalso analyzed [38], [39].

An IEC Fitting system for one type of a hearing aid onthe market was constructed on a note PC and is on fieldtest with actual users [237], [147], [39].

A

r

m

w

r

e

s

t

l

i

n

g

:

h

u

m

a

n

v

s

.

r

o

b

o

t

c o

n

t

r

o

l

l

e

r

f u

z

z

y

c

o

n

t

r

o

l

r

u

l

e

s

f

o

r

V

R

I E

C

s u

b

j

e

c

t

i

v

e

e

v

a

l

u

a

t

i

o

n

f

o

r

V

R

m

o

d

i

f

y

i

n

g

r

u

l

e

s

Fig. 7. VR of arm wrestling. Fuzzy logic controller evolved based onhuman perception of VR.

J. Virtual Reality

Which factor gives us virtual reality (VR) ? The IECmay solve this problem. Imagine the VR control of an armwrestling robot fighting against a human (see a photo inFigure 7). Whether a human fighter feels VR in action ofthe arm wrestling robot or mechanical swinging dependson the control rules.We obtained the control rules to win using a classifier

system in the first stage. When the arm wrestling robotpushes a human fighter forward, the control rule of the clas-sifier system is awarded, and 20,000 rules were obtained.Then the 20,000 classifier rules were complied into 8 fuzzyrules to help our analysis. This compilation was conductedby using fuzzy knowledge acquisition techniques based onthe GA [68], [69].The final stage is that the IEC modifies the fuzzy con-

trol rules for winning to those for VR. Since only a humanfighter can evaluate how he or she feels as if the opponentwere a human, we can use the IEC to modify the parame-ters of the fuzzy control rules. It is expected that the VRfactors of the force perception may be explicated by analyz-ing the difference between fuzzy control rules for winningand those for VR.

K. Database Retrieval

Suppose that we wish to retrieve an image or music froma huge database or the internet. In most case, especiallythe case of images, the media that we want to retrieve is nota specific one but one that is suitable for a certain purposeor that is preferable. Retrieval with keywords can rarelybe used for this purpose, and we often do not know whatimage or music are in the huge database or on the internet.The IEC is applicable to contents-based media retrieval

and can find out image, music or other media that matchto certain purpose or user’s preference based on humanevaluation for retrieved ones.The EC searches media on their feature space, while hu-

man evaluates them on a psychological one in mind [178],[26], [94], [95], [96], [97], [27], [28] such as Figure 2 or anartificially constructed psychological one [132], [133], [202],[206] for intelligibility. A mapping method from the psy-chological space of the latter to the feature one is neededto find the best matched media to the target point on thepsychological space. However, the dimension number of

7

NN

E

C

s

e

a

r

c

h

f e

a

t

u

r

e

s

p

a

c

e

p s

y

c

h

o

l

o

g

i

c

a

l

s

p

a

c

e

l i g

h

t

bright

c h

e

e

r

f

u

l

f r

e

q

u

e

n

c

y

a

m

p

l

i

t

u

d

e

d

i

s

t

r

i

b

u

t

i

o

np

i

t

c

h

m

y

i

m

p

r

e

s

s

i

o

n

i

s

.

.

.

Fig. 8. Media database can be retrieved by reverse-mapping from thepsychological space of a user to the physical space of the mediain the database after obtaining the opposite mapping relation.

the psychological space is generally so far less than thatof the feature space that it is difficult to map in this di-rection. On the other hand, the mapping in the oppositedirection is easier. Neural networks (NN), fuzzy reasoning,or other nonlinear mapping methods are suitable to mapfrom m-D space to n-D one where (m ≥ n). We may beable to use the trained such nonlinear mapping system forthis direction, and GA was used for the inverse mapping(see Figure 8). Of course, there are several points that cor-respond to one point on the psychological space due to oneversus multiple mapping. This means hat multiple imagesor music which give us same impression can be retrieved,which is desirable for media retrieval.Sometimes user must feel that the retrieved media does

not match with with the target impression. There are twouser choices in this case: to begin again at the beginningor to try again by changing the target coordinate on thepsychological space. The IEC is used for the latter.Several papers applied the IEC to image database re-

trieval. Some of them are: interactive image browsing usingpipeline-type GA mentioned later [82], [83], [84], interac-tive image retrieval using wavelet image feature [26], [94],[95], [96], [97], [27], [28], image retrieval that uses color, tex-ture, position of meshed image areas as features and thatassigns bias fitness values to undisplayed images while userrates displayed images [178], image material retrieval formulti-media title design [117], and national flag retrieval[108].The image material retrieval system for multi-media title

design uses the IEC to find out user’s preferable images andtime-sequentially lines them up. This system uses MessyGA to increase the redundancy of gene codes, which resultsto reduce the load of interactive user-selection [117].The idea of a media converter was proposed. It combines

IEC-based image and MIDI file retrieval systems and uni-fies their psychological spaces into a common one, whichcan realize the retrieval of image features–psychologicalspace–music features direction and its opposite direction[206], [134]. In other words, computers can show us an im-age of same impression when we input the music, and viceversa.As the EC usually searches a feature space of the spatial

distribution information such as color and gray levels, it isimpossible to retrieve media in semantic level. For exam-

2

,

3

0

0

q

u

e

s

t

i

o

n

n

a

i

r

e

f

o

ro

r

a

l

c

a

r

e

g

o

o

d

s

1

6

i

m

a

g

e

w

o

r

d

s*

2

,

3

0

0

16goods attributes

inductive learningtool

i m a

g

e

w

o

r

d

s

v

s

.

g

o

o

d

a

t

t

r

i

b

u

t

e

s

r

e

l

a

t

i

o

n

t

r

e

e

G

A

r e

l

a

t

i

o

n

t

r

e

e

#

1

r e

l

a

t

i

o

n

t

r

e

e

#

2

r e

l

a

t

i

o

n

t

r

e

e

#

3

r e

l

a

t

i

o

n

t

r

e

e

#

4

r e

l

a

t

i

o

n

t

r

e

e

#

n

G

A

c

r

e

a

t

e

s

v

a

r

i

a

t

i

o

n

o

fi

m

a

g

e

-

a

t

t

r

i

b

u

t

e

r

e

l

a

t

i

o

n

t

r

e

e

s

H

u

m

a

n

c

h

o

o

s

e

s

s

i

m

p

l

e

r

a

n

d

b

e

t

t

e

rr

e

l

a

t

i

o

n

s

o

f

i

m

a

g

e

-

a

t

t

r

i

b

u

t

e

r

e

l

a

t

i

o

n

s

.

Fig. 9. Reliable rules with fewer number of attributes that charac-terize each oral care product are determined from questionnairedata using the IEC and a marketing expert.

ple, it is hard to retrieve an image of a smiling girl in thedark with happy impression. If the semantic level is lim-ited to basic shapes such as triangles, squares, or circles,they may be able to be used as image features for imageretrieval in the semantic level. There was an IEC-basedimage retrieval that handles the semantic by limiting theretrieved target to only such simple images [70], [71].

L. Knowledge Acquisition and Data Mining

When a new version of product is planned to develop,firstly the customer target and product concept are decidedaccording to its market research, and then several attributeparameters of the product are adjusted to match to the newconcept. The problem is how to match the parameters.Their relationship rules may be obtained from past marketresearch, but noise in questionnaire survey from consumersis inevitable.The IEC was applied to acquire marketing knowledge of

oral care product from the questionnaire data [60], [214],[215], [216], [217], [218], [219], [61], [220], [221]. Their ap-proach consists of: (1) basically applying inductive learningto obtain the relationship rules from the data, (2) increas-ing diversity of possible attribute parameters in the rulesusing genetic operations, (3) expert’s choice of better rulesthrough the IEC process, and (4) finally determining re-liable decision making rules or trees with fewer numberof attributes that characterize each oral care product (seeFigure 9).The inductive learning and the IEC with an expert are

applied to 2,300 questionnaires with 16 image questionssuch as “better taste and smell” and “usable for all family.”They reported that their obtained rules have dramaticallysmall number of product attributes and small size of rules,and the obtained rules show the marketing strategy of acertain company.Same approach was applied to the questionnaire data

for beer product [62], [63] and clinical data [56], [222]. Therules obtained by inductive learning were evaluated by theexperts of marketing or clinical medicine and were polishedup by a simulate breeding.

8

The IEC was applied to another type of data miningbased on 2-D mapping. GP generates nonlinear functionsthat inputs all attribute parameters of target data and out-puts two coordinate values, (v1, v2), i.e. this function mapsdata from an m-D to 2-D space. The multiple data distri-butions on the 2-D space generated by the GP are dis-played to a user, and the user selects visually meaningfuldistributions of the data. The point of this research isthe meaningful; the meaningful distribution implies thatthe mapping function generated by the GP may have somekind of meaningful knowledge. This selected ones evolve,and the offspring distribution maps are generated. It isreported that this method acquits proper knowledge [235].Combination of normal GP and the IGP was applied

to obtain knowledge that classifies eight kinds of damagedata of stainless steel [188]. Unlike the normal IEC, humaninteraction was not conducted in every generation but inevery 100th generation to speed up the IGP. The GP gen-erates rules classifying the types of damage by inputting 17parameters of damaged data in their experiment. It wasreported that fitness values were drastically changed at the100th or 200th generation and the rules obtained by theGP showed robustness for noisy data.

M. Image Processing

Image enhancement for medical images is helpful andnecessary for medical doctors to detect diseased parts easilyand correctly. The performance of the image filters formedical image enhancement is evaluated by only humans,especially medical doctors, and the best enhanced imagemight depend on doctors’ visual preference. This is a goodIEC task.The IGP was applied to design image filters that enhance

MRI (magnetic resonance image) and echo-cardiographicimages [166]. The GP creates math equations that de-scribes the image filters based on the human’s visual dis-tinguishability.The IEC was applied to detect the outline of plants in

an image. Image processing that segments plants part anddetects their outlines is necessary to obtain the growth in-formation such as shape, growth rate, or leave color fromtele-monitored plant images. To obtain the best image pro-cessing filters, the GA optimizes their coefficients to bestmatch a filtered image with the objective plant image thata user traces on an original image using a drawing user in-terface. The combination of a fitness function showing thematching degree at each pixel and human visual evaluationis fed-back to the GA [153].Another IEC approach is to determine the sequence of

image filtering. Due to the wide spread of digital cameras,image scanners, PCs, and the internet, the possibility thatamateurs work with images has increased. Most amateursuse retouching software that prepares several image filters.Generally speaking, different orders of image filtering re-sults in differently retouched images, i.e. an image filteredby A, B, and C is different from an image filtered by C, B,and A. Although it may be difficult for amateurs to decideon a filtering sequence, it is not difficult for them to evalu-

ate which image is better or preferable. Simulated breedingwas used to auto-generate the sequence of image filtering[116].

N. Control and Robotics

The IEC applications to control have increased recentlybesides the arm wrestling robot in section III-J.The first engineering application of the IEC was the GA

control of a six legged insect-style robot in 1992 [100]. Itseach leg has two neurons for swing and elevation respec-tively, and the leg movement is determined by the oscilla-tion between the two neurons. Then, the locomotion of therobot is determined by the connection parameters amongoscillators of the six legs. The GA optimizes the first oscil-lator neurons based on human observation of each leg, i.e.IGA, and the latter control among legs is evolved based ona fitness function of how long the robot goes forward orbackward with less totter.Similar approach to NN control of eight legged robot

introduced the IEC and three devices and dramatically de-creased the number of GP generations from one millionof their simulation to 200 of their hard system [48]. An-other IEC approach was the obstacle avoidance control ofa Khepera miniature robot [33]. The interaction betweenthe GA and user’s visual evaluation was used in their firstevolving phase, but the interaction between the GA anda human evaluation model obtained by the observation ofthe user evaluation in the first phase was used in their sec-ond evolving phase to reduce human evaluation fatigue (seesection IV-C).The IEC was used to control a Lego robot to realize its

interesting locomotion that children prefer. The connectionweighs of the NN that inputs robot sensor information andoutputs locomotion control values are evolved according tothe selection of better locomotion robots by children [154],[106], [107]. It can say that the IEC realizes a program-mingless program that is the best for children games ortoys. It is surprising to know that only nine populationsize of moving robots generated the control NNs that real-ized interesting movement of robots after only five or sevengenerations in their experiment [107].The IEC was applied to generate humanity movements

of robots. One application was robot arm path planningnot to give a human a fright when a human and a robot docooperative works such as handing goods [89]. While theevaluation measure for the arm path planning of industrialrobots is efficiency such as the shortest distance or time, theminimum energy, or the biggest stability, different measureis needed for that of consumer robots. Humans must fearmore or less when a lump of iron, robot arm, approachestoward the humans. The objective of this research is tofind the best tracking path and speed of the robot armthat minimizes the fright feeling using the IEC. This typeof approach should become important for care robot, petrobots, or other consumer robots that human-friendlinessis required rather than efficiency.As we can guess human emotion from his or her body

gesture, it is possible to express such emotion in the move-

9

ment of autonomous robots. This trial was conducted tooptimize the consequent vales of fuzzy control rules for therobot movement using the IES . A user observes the differ-ent movements of autonomous robots and chooses bettermovements whose caused motional emotions are the clos-est to the emotion that the user looks for, such as happy,angry, or sad [73].The IEC is considered to use to control an NN controller

that inputs throttle angle and vehicle speed and outputsthe air-fuel ratio to take account of user’s preference forcomfortableness to ride in [67]. As the first step of theirresearch, they did not use the IEC for their simulation andan experiment with a real engine but used an EC and ahuman model of the NN that learned the relationship be-tween machine data and user’s preference instead of thereal user. They are planing to use the IEC in their nextresearch step [67].When we design or acquire control rules, sometimes their

intelligibility to explain them to other people has higherpriority rather than their best performance. The IEC wasapplied to obtain such fuzzy control rules for car park-ing with the combination of a fitness function for objectiveevaluation of the control performance and human evalua-tion for the intelligibility of the obtained rules [2], [150],[151]. This research might be categorized in section III-L.It was proposed to allow a user to directly indicate which

control rules are better to increase the fitness during a nor-mal fuzzy classifier system runs. Their trial to acquire fuzzylogic rules that control a robot to go to take a flag andcome back without collision with other robots was failedin their first simulation. Then, they observed the partiallysucceeded rules to reach to the flag in the 21st generationand indicated their classifier system to search in the neigh-borhood of the fired fuzzy control rules. This human in-tervention resulted to find out the successful robot controlrules at the 23rd generation [37].Recent cool topic on robotics in Japan is a spread of

robots from industry to home. The size of Honada’s AIMOis determined taking account of the cooperation with a hu-man at home, Sony’s AIBO that 5,000 of its first modelwere sold out within only 17 minutes through internet orderin spite of its extremely expensive price of 250,000 YEN em-phasizes its entertainment aspect as a pet robot, and sealand cat robots with fur of Mechanical Engineering Labora-tory, MITI, aim to make users’ feeling of cuteness or mentalhealing run high, too. The IEC technique will be in muchdemand for generating cute, natural, and eased motions ofconsumer robots whose sales talk is not working efficiencybut cuteness or safety.

O. Food Industry

The IES was applied to coffee blending. Before the ap-plication, the performance of the IES was evaluated withthree toy tasks [51]: (1) to find out the mixture ratio ofthree liquid primary colors of cyan-blue, magenta-red, andyellow and with clean water, which results the color ofCherry Brandy, (2) to generate a square whose size andcolor are same with those of the target square by optimiz-

ing the five parameters of length, width, red, green, andblue, (3) to generate a polygon whose shape is specified byline segments that radially spread with equal angles fromthe origin coordinate and match to the target polygon.Then, the IES was applied to coffee blending with the

cooperation of three professional coffee tasters of a coffeeroasting company in Berlin. The task is to find out the mix-ture ratio of five kinds of coffee beans whose coffee taste issame as that of the target coffee. The IES found our the ra-tio that the professional coffee tasters could not distinguish[52].

P. Geophysics

The IEC has been applied to geophysics. The mantleconvection has already modeled with physical characteris-tics such as earth revolution. If this mathematical model iscorrect and correct parameters of the initial earth inside sit-uation are inputted to the model, the current undergroundsituation can be estimated. However, as nobody knows thepast situation of earth inside, the modification of the ini-tial situation parameters and inputting them to the modelwere iterated by try and error according to the observationof its simulation results. This task is categorized to a re-verse problem because the model input is estimated fromthe model output. If this model simulation goes well, wemay estimate where are mines.Expert knowledge of geophysics and geology and really

measured geophysical data are indispensable to the appro-priateness of this simulation. So far, the initial values of theconventional simulation were determined by trial and errorthough the experts evaluated the simulation result. TheIGA has been applied to this point; the GA determines theinitial parameters of the simulation, and the expert evalu-ates the validity of the simulation results [24], [21], [22].This approach was applied to estimate the existence of

heavyweight mass. When gravity from A point to B pointon the earth’s surface is measured, ununiform gravity graphis obtained if there is heavy rock or mines underground.The task is to estimate the depth, gravity, and size of therock or mines. Although it is mathematically impossibleto determine unique values of the three variables from the1-D measured gravity data, the geophysical knowledge andexperience would help to narrow the solution candidatesdown. Here, the IEC with a geophysicist is applied to esti-mate the combination of the depth of its location, gravity,and size of the rock [23].

Q. Art Education

Both of artistic sense and skill which are essential for artcreation are usually trained through sketch and formingtraining. However, the skill training needs long time assame as the training of playing music instruments. We canseparate these two trainings and focus on education thatpolishes up artistic sense by introducing the IEC.An IEC-based educational system that consists of a 3-D

CG formingmodel and the IEC changing the model param-eters to create several 3-D shapes fitting to the given motifis being developing [127], [128], [129]. This system aims

10

the education on especially artistic sense rather than artis-tic skill training and let the artistic skillless students polishtheir artistic sense, while the artistic sense has been pol-ished up through artistic skill training in traditional artis-tic practice as mentioned. Figure 10 shows an experimentalscene of creating 3-D green peppers on a big screen withviewing from several angles. Now this system has installedto note PCs and can be brought to any classes and schoolsof artistic education. We are going to connect the internetversion of the IEC interface and realize a remote educationsystem for the artistic sense training.

Fig. 10. IEC-based 3-D CG forming system for art education.

R. Writing Education

Since the IEC displays several individuals, sometimesthey give a user hints or inspiration; it implies that theIEC is useful to stimulate human creativity. This inspiredcreativity obtained during the IEC iteration is more im-portant for education than what the EC outputs, such asimages or music.This characteristic of the IEC was applied to develop

a writing support system for children [90], [91], [92], [93].Composing a story is more important and difficult thanwriting sentences, and children in the lower classes are fre-quently puzzled. Their system displays 24 pictures and letsa child choose two sets of a four-picture sequence; this se-quence is the writing story. Their system then creates sev-eral four-picture sequences using the selected parents’ four-picture sequences and GA operations. The child choosestwo better four-picture sequences in the next generation.This iteration is repeated until a satisfactory story is writ-ten.

S. Games and Therapy

Similar to educational applications, edutainment orgames are a good application field of the IEC, because itis much easier for children to select better ones than givedetail instruction or write programs. This selection corre-sponds to an award of reinforcement learning, and the IECcan train the control rules or mechanism of a target systemsimilar to reinforcement learning.Besides the Lego robot in section III-N, some of IEC-

based games are an artificial life survival game, painting

graphics based on NN evolution, Artificial Painter [236],[155], and drawing faces [154].It is reported that the drawing face software is used for

not only edutainment but also mental therapy. Italian ther-apists joined and started a project to apply the software toencourage mental diseased children to understand face ex-pression [154].

T. Social Systems

New type of the IEC was proposed for social systems; allEC operations are conducted by humans. Since the EC is acomputational model of biological evolution, it has been acommon sense for almost EC researchers that the EC runsin a computer. Even IEC applications in previous sectionspasses only evaluation part from a computer to humans.The new IEC that was applied to two social systems.

Knowledge Free Exchange is a forum on web to obtainvaluable idea, and its process is based on the IGA [86].Texts of questions and answers (chromosomes) consistingof words of natural language (genes) compose its knowl-edge database (population). Forum participants observeand combine these submitted answers and fill out new an-swer (crossover), or are inspired by other answers and mod-ify their fill out quit new idea (mutation). Each answer isevaluated by the number of participants who are interestedin the answer (human evaluation of fitness). New answeror opinions (offsprings) are generated through this process.Teamwork for the Quality Education is a course work

project that was proposed during curriculum reform for en-gineering education, that was based on team competitionin academics, service and design, and summer-job replace-ment, and that was tested in Spring and Fall semestersin 1997 at General Engineering Dept. of Univ. of Illi-nois [41]. Although the original reference [41] did not ex-plicitly describe the relationship between tht project rulesand the GA process, the reference [86] pointed out that;teams (chromosomes) consisting students (genes) competebased on the evaluation to each team (human evaluationof fitness) in each semester (generation); first team mem-bers are collected through the draft (initialization), and theteam competition (selection) is conducted with team mem-ber swap (crossover) and imitation of other team activity(mutation) at the beginning of next semester.

U. IEC of Wide Definition

We are focusing the IEC researches of the narrow def-inition in this paper (see section I), but let’s see some ofthose of the wide definition in this section. To distinguishboth definitions of the IEC, we temporally use the termsof narrow-IEC and wide-IEC only in this section.The most wide-IEC researches would be multi-objective

optimization research introducing the GA. Most of multi-objective optimization papers whose titles include the termof interactive are the wide-IEC researches. Some of GA-based multi-objective optimization, such as the researchof multi-objective evaluation consisting of user’s subjectiveevaluation in part [212] and that letting a user subjectively

11

give weights to multiple fitness values [187], should be cat-egorized in the narrow-IEC.The wide-IEC was applied to a nurse scheduling support

system [59] that might be in the narrow-IEC. Although theGA makes a nurse scheduling table based on the give fitnessfunction, sometimes a head nurse may find inconvenient orunacceptable scheduling results because it is difficult to de-sign the fitness function to fully satisfy user’s specifications.In this case, new constraints of nurse scheduling that thehead nurse finds are added to the fitness function, and theGA search is restarted. This is, this system optimizes thescheduling based on the interaction of a human and GA,but subjective evaluation is not directly used as a GA fit-ness value but used to correct the fitness function.Computational biology is a new approach in pharmaceu-

tical industry, and the GA is used for a simulation to gen-erate a new molecular system by combining biomolecules.The GA searches the most tight bond of the biomolecules,i.e. the combination that minimizes the free energy ofthe obtained molecular system, by changing the combina-tion of six parameters such as rotation of the molecularin a (x, y, z) space. After several GA generations, a hu-man generates a new individual molecular system by man-ually combining two biomolecules in a virtual reality space,CAVE, and adds it as a new elite [99]. Although this is nota narrow-IEC application, it is an interesting GA-basedhuman-machine interaction system in a new field.The GA takes longer time to solve a TSP (traveling sales-

man problem) according to its size. There is a proposal toallow a user to interactively divide a big TSP into somesmall TSPs and visually combine the solutions of the smallTSPs to obtain the solution of the final big TSP to reduceGA calculation time [104].The interactive evolutionary design environment that

Parmee et al. developed with British Aerospace is a designsystem through the interaction of a design team and thesystem with a built-in GA [158], [159], [160], [161], [162],[163], [164], [165].

IV. Non-Applicational Researches

A. Remained Technical Problems

The biggest remaining problem of the IEC is reducinghuman fatigue. Since a human user cooperates with a tire-less computer and evaluates EC individuals, the IEC pro-cess cannot be continued for many generations. This is thebiggest problem prohibiting the practical use of the IEC.The secondary problem derived from the fatigue prob-

lem is that the IEC has to search the goal with a smallerpopulation size within fewer number of searching genera-tions than a normal EC search due to the limitation ofthe individuals displayed in a monitor at once and limita-tion of human capacity to memorize the individuals time-sequentially displayed. This limitation results poor andslower EC convergence.The third problem is that the IEC has to time-

sequentially display individuals when the outputs from thetarget system are sound or movie. Since the IEC user is

forced to compare the current displayed sounds or movieswith the past ones in his or her memory and evaluate them,user’s psychological fatigue must increase. Same situationsometimes happens even if the outputs are still images thatare able to be usually displayed spatially; if the imagesare big and detailed ones, they should be displayed eachby each or each a few by a few; If the population size isincreased to solve the above secondary problem, the indi-vidual images must be displayed each set by set due tothe maximum number of images displayed on a monitorat once. The third problem is how to solve this fatigueproblem caused by time-sequential display.The good news is that many IEC tasks do not require

a large number of generations to achieve satisfactory re-sults. Due to the good initial convergence of EC, un-like gradient methods, the human fatigue problem may befewer than gradient searches. Task characteristics of sub-jective searches and numerical/combinational optimizationare quite different, and the former does not have the exactoptimum point. This is why it is sufficient for the humanevaluated task to reach to an optimum area rather thanone point, which searches out a satisfactory solution withfewer searching generations.Nevertheless, we must solve the fatigue problem to make

the IEC fit for practical use. Most of non-applicationalresearches are related with the IEC interface research tosolve this fatigue problem.

B. Discrete Fitness Value Input Method

Psychological fatigue is deeply influenced by the ease ofevaluating the outputs of the IEC and feeding back theevaluation values to the EC. For example, as we cannotexactly distinguish the difference between 62 and 63 pointsin 100 levels rating, to determine 62 or 63 points of oursubjective evaluation to individuals causes psychologicalfatigue.It is expected that users can daringly evaluate the EC in-

dividuals and therefore reduce their psychological fatiguewhen rough rating, such as five or seven levels rating, isused instead of high order level rating. Such psychologi-cal discrete input method that distinguishes from 100 or200 levels rating is proposed [189], [190], [191], [192], [193],[139], [197]. Since the rougher level rating results higherlevel of quantization noise, the EC convergence may be-come worse. We evaluated the total performance of theproposed method by taking into account both the advan-tages and disadvantages.The subjective test and statistical test showed that the

proposed method significantly reduces human fatigue. Asimulation experiment showed that the worse convergencebecomes significant when the EC search reaches several 10sor 100s of generations. It showed that the poorer conver-gence in practical IEC generations, such as less than thefirst 10 or 20 generations, is not problem. This simulationresult supports the result of the subjective test [190], [193],[139].

12

C. Prediction of Fitness Values

Some methods that predict the fitness values of individ-uals for fast IEC convergence as a solution of the secondproblem in section IV-A were proposed. If the IEC hasa prediction function, it can increase searching capabilityby using big population size being equal to that of normalEC but do not increase user’s fatigue by display only a fewnumber of individuals that are predicted to have higherfitness values. Two prediction methods that learn user’sevaluation characteristics and predict fitness values wereproposed; one uses distances in an EC searching space,and another uses an NN.One of typical distance-based approaches is to prepare a

function that predicts the fitness values of any individualsbased on the Euclidian distance from the individuals that auser gives his or her subjective fitness values, to apply thefunction to 200 individuals, and to display the only best10 of the 200 individuals to the user for subjective eval-uation [118], [119]. One of improvement of this approachis to cluster new individuals around each past evaluatedindividual and to predict the fitness values of the new in-dividuals using the distance from the evaluated individualinside the cluster [98].Another distance-based prediction approach is to use a

rule-based system. There are two IEC evaluations; ratingall individuals and simulated breeding or user-selection.The former has higher searching capability but causeshigher user fatigue, while the latter has opposite charac-teristics. A compromise proposal is to allow a user to se-lect better ones and to let a rule-based system or functionassign bias fitness values to unselected individuals in steadof 0 value. One simple idea is to assign a same bias fitnessvalues to the unselected individuals; another is to measurethe Euclidian distances among chromosomes for them [54];the other is to use approximate reasoning to predict theirbias fitness values from the selected individuals [130], [131].Further improvement for the IEC-based cartoon face draw-ing system is to measure the distances not in a phenotypespace, face parameter space, but in a psychological spaceconstructed by the multidimensional scaling method [184],[185], [186].The robot control mentioned in section III-N obtains ob-

stacle avoidance behavior through two phases; IEC phasewhere the fitness prediction model is formed during the IECprocess comes first, and non-IEC phase that uses the modelfollows. User interacts to the candidate robot behavior byshining light to robot photo sensors in the IEC phase whenthe robot takes undesired behavior. Then, the IEC makesrules of user preference model to robot behavior. NormalEC training phase follows the IEC phase without real userbut with the user model, which shortens the human inter-action time and therefore reduces the human fatigue.Some researches not only just predict new fitness values

from the user’s fitness values but also use the predicted val-ues for the improvement of the display interface or screen-ing. One proposal was to display individuals roughly inthe order of human evaluation and allow IEC users be ableto evaluate them by comparing neighbor individuals using

Euclidian distance and/or an NN, which is expected to re-duce human fatigue [194], [138], [140], [141], [142], [197].Or, when similar evaluated individuals are grouped anddisplayed, it is also expected to roughly and easily eval-uate individuals. Its application to screening was to tryto make an NN learn nonmusic melodies and to eliminatethe nonmusic melody from the individuals using the NN toreduce human fatigue [16].There are many remained research points on fitness pre-

diction. One of them is to construct a distance measurethat is same as human evaluation scale. Euclidian distance-based fitness prediction ignores the difference of evaluationdegrees among searching parameters. For example, the de-gree that the shapes of eyebrows and a mouth influence tothe face impression is much bigger than that of hair andnose, and prediction performance depends on whether suchdifferences are reflected into the distance measure. The sec-ond point comes from that the IEC user’s evaluation scaleis not absolute but relative in each generation, i.e. thefitness values in far past generations and in near past gen-erations may be different even if the evaluation target aresame, and these difference becomes noise in learning user’sprediction characteristics. The simulation evaluation of theabove display method in order of predicted fitness valuesshowed the significant prediction performance, but theirsubjective test using human subjects has not shown thesignificant effectiveness in reducing human fatigue [142].Also, it was reported that the sNN creening of nonmusicmelody was very tough [16]. Further research and deepreconsideration of the prediction target or parameters areneeded.

D. Interface for Dynamic Tasks

The fatigue problem becomes serious when an IEC usercompares and evaluates sounds or movies which are spa-tially impossible to be displayed as mentioned as the thirdproblem in section IV-A. Fatigue reduction methods usinga speech processing task was investigated.The first investigation was on the number of displayed

sound sources. When one sound source is used, it is easierfor a user to compare the difference of several processingeffects, but the user becomes boring and tired, while op-posite effect happens according to increase the number ofsound sources. Subjective tests were conducted and com-pared the three cases on operability and EC convergence:(1) one sound source is used for all individuals in all gen-erations, (2) one of 20 sources are randomly assigned toany individual, and (3) sound sources are changed at eachgeneration, and same source is used for all individual insame generation. The subjective tests showed that the (3)is significantly better than others [226], [227].The second investigation was on the effect of explicit

display of an elite individual to reduce forcing the userto compare the displayed sound and past sounds in hisor her memory. Two display interfaces were evaluated bysubjective tests; one indicates which is the elite individualand allows an IEC user to play the sound whenever he orshe wants to compare, and another is a conventional one.

13

The subjective test results showed that this improvementof an display interface is significantly effective [226], [227].The third investigation was comparison of two GUIs; one

GUI has only 1 play button and forces the user to evaluateonly the sound that the IEC GUI displays [237], and an-other has 20 ones and allows an IEC user to choose any of20 individual sounds for evaluation. The former providessimple user operation because the next sound is automat-ically played as soon as the user inputs fitness rating andit is no need for the user to choose next individual to playand compare it, but user cannot choose any individuals.The latter’s operability is good from the free comparisonpoint of view, but the number of mouse clicking for choos-ing individual to play is increased by one. This trade-offwas evaluated through subjective test and it was shownthat the former is effective [227].The GenJam in section III-D adopts real-time evaluation

of melodies to shorten total evaluation time rather thanevaluation after whole one melody is played [13]. Whilethe comparison of a displayed melody with other past wholemelodies takes long time and is tiresome, but this real-timeevaluation phrase by phrase is easier. This idea can be usedfor other IEC tasks of time-sequential display such as VRcontrol rule acquisition of an arm wrestling robot in sectionIII-J.

E. Acceleration of EC Convergence

Acceleration of EC convergence significantly reduces hu-man fatigue. Although any fast EC search methods are ap-plicable, methods whose convergence in early generationsthat is especially fast are useful for the IEC. As previouslymentioned, the practical evaluation generation of the IECis less than first 10 or 20 generations. Quickening meth-ods that work later, such as second order convergence ofgradient methods near global minimum, are not suitable.A new elitist method by approximating the searching

surface using a convex curve was proposed [57] and ap-plied to the IEC [141], [58], [197]. Evaluation tests withseven benchmark functions showed the significance of a fastconvergence especially in early EC generations. This ideais useful for any EC tasks whose fitness evaluation takeslonger time than the calculation time of the convex curve(note that the geophysical simulation in section III-P takes30 minutes to one hour in certain condition, and fast con-vergence in early generation is much required even if thequickening method takes time.) However, subjective testshave not shown that this fast convergence did not resultsignificantly redaction of human fatigue, though the taskis only one task of drawing faces [141].The human evaluation time in the IEC is extremely long

from a computer point of view. The idea to make thecomputer do something during this waiting time results toshorten total IEC time. The idea of the pipeline-type GAis based on this idea [82], [83], [84].

F. Combination of IEC and Non-IEC

It was proposed that the parameters of fitness functionare tuned by simple reinforcement learning during interac-

tion phase to reduce the IEC user fatigue [61], [220]. Itrealizes a system that iteratively conducts IEC and GAsearches.The damage detection system of stainless steel in section

III-L is combination system, too. It uses the GP usuallyand the IGP at every 100th generation [188].

G. Active Intervention

Two active user interventions into EC search were pro-posed: on-line knowledge embedding and Visualized IEC.The role of normal IEC users is just to evaluate the dis-played sounds or images from a computer, and optimiza-tion search itself is conducted by the EC. Researches thatallow the EC users to actively interfere the EC search toaccelerate the EC or IEC convergence is the topic of thissection. Active intervention that motivates the users andfast EC convergence resulted by the active intervention di-rectly reduce human fatigue.On-line knowledge embedding is a method that provides

a mechanism for accepting searching ideas, hints, or inten-tions during the IEC operation and allows the user to ac-tively participate in the EC search and, hopefully, improveits convergence. For example, when a user perceives that acertain facial feature of an individual montage image willimprove the EC search, the genes corresponding to the fa-cial feature are masked not to be crossovered and mutated[25], [81], [203]. This masking means that the dimensionalnumber of the searching space is reduced and we may beable to accelerate the EC search. On the other hand, ithas negative aspects such as we have to pay attention tonot only to whole impression but also partial impressions,or user’s operation to specify the masked feature increases.This trade-off was evaluated using a montage system withsubjective tests. The result showed that (1) the unit op-eration time increased 30 %, nevertheless the total conver-gence time became faster than the increased time, and (2)the obtained montage image with the on-line knowledgeembedding method was significantly closer to the targetface image [81], [203].The Visualized EC or Visualized IEC is a method that

visualizes the EC search landscape by mapping the individ-uals with their fitness values from an n-D searching spaceto a 2-D space, helps to let the user estimate rough posi-tion of the global optimum on the 2-D space, and uses theposition as a new elite individual [49], [207], [50]. The Visu-alized EC or IEC combines different advantages of the ECand human searching techniques; the EC directly and sys-tematically searches the original n-D gene space based onEC operators, which is better than human searching tech-niques, while humans have an excellent capacity to graspan entire distribution of individuals in the 2-D space at amacroscopic level, which cannot be interpreted by an EC.The good news of this method is that we can expect lowrisk, high return; the EC convergence is dramatically im-proved in the best case, and only one of many individualsdoes not work well in the worst case.The performances of a normal GA and the Visualized GA

were compared using benchmark functions. Self-organizing

14

map was used for the 2-D mapping. The experimentalresults showed that the convergence of the Visualized GAwith 20 population size corresponds to that of normal GAwith 100 to 1,000 population size [49], [207], [50].

C

GE

C

subjective evaluation

xn

e

w

s e

l

e

c

tM

-

D 2-D

M

a

p

p

i

n

g

E

C

s u

b

j

e

c

t

i

v

e

e

v

a

l

u

a

t

i

o

n

C

G

Fig. 11. Normal IEC (upper) and Visualized IEC (lower). VisualizedIEC maps individuals in n-D EC searching space to 2-D spaceand visualizes the distribution of fitness values.

H. Theoretical Research

There was a trial to make a probabilistic model for theIEC process with stachatics Mealy automata, which wasmotivated by the GA modeling with the Markov chain[169]. Although this research seems to be in the early stageof the research, this is a quite unique IEC research whileall other IEC researches are on either IEC applications orinterfaces.

V. Conclusion

We overviewed how IEC technology have been spread tothe wide variety of fields, what problems remain, and whatkinds of struggles to solve the problems and to make thetechnology practical have been conducted.Let’s look back over the history of computational intel-

ligence, especially that of so-called Soft Computing [211].The seeds of the NN, fuzzy systems, and evolutionary com-putation were sown in 1960s, and they were widely butindependently researched in 1980s; as these research activ-ities had increased in 1980s, interest in fusing them has alsobeen rapidly increasing at the end of 1980s, and these coop-erative technologies widely and practically spread into com-mercial products and industrial systems during the 1990s.Roughly speaking,1980s was the decade of each computa-tional intelligence technology, and 1990s was the decade ofthe cooperative technologies of computational intelligencetechnologies.Then, which direction do the computational intelligence

researches go to ? One of possible research directions would

be a humanized technology that computational intelligencetechnologies and humans are cooperatively combined as theresearch on human factors or KANSEI increase since 1990s.One of the realization ways of the humanized technologyis the IEC that combines the human evaluation capabilityand computer’s optimization capability.The IEC will become more important in design, adjust-

ment, and creation that are directly related with humanessential KANSEI factors such as preference, emotion, feel-ing, and so on. It is deeply expected that the IEC is furtherdeveloped and becomes of some help toward humanizedtechnology and user friendly technology.

Acknowledgements

This survey was completed with the help of several peo-ple who provide information of IEC papers, send the pa-pers, comment this article, help to make a database, andpolish up this paper. Especially, author is grateful to Prof.sTakao Takao and Tatsuo Unemi for their offers of paperinformation and comments to this paper and to my stu-dent, Kei Ohnishi, for his assist to complete author’s IECdatabase.

References

[1] P. J. Angeline, “Evolving fractal movies,” in 1st Annual Conf.on enetic Programming, (Stanford, CA, USA), pp. 503–511,July 1996.

[2] T. Anzai and T. Onisawa, “Acquirement of understandablefuzzy rules,” in 15th Fuzzy System Symposium, (Osaka, Japan),pp. 467–470, June 1999. (in Japanese).

[3] K. Aoki, H. Takagi, and N. Fujimura, “Interactive GA-baseddesign support system for lighting design in computer graph-ics,” in Int. Conf. on Soft Computing (IIZUKA’96), (Iizuka,Fukuoka, Japan), pp. 533–536, World Scientific, Septem-ber/October 1996.

[4] K. Aoki and H. Takagi, “3-D CG lighting with an interactiveGA,” in 1st Int. Conf. on Conventional and Knowledge-basedIntelligent Electronic Systems (KES’97), (Adelaide, Australia),pp. 296–301, May 1997.

[5] K. Aoki and H. Takag, “3-D CG lighting design support systemusing interactive GA,” in IEICE A-15-30, pp. 16–1–1608, Mar.1997. (in Japanese).

[6] K. Aoki and H. Takagi, “Lighting design support on 3-D CG,”in 13th Fuzzy System Symposium, (Toyama, Japan), pp. 311–314, June 1997. (in Japanese).

[7] K. Aoki and H. Takagi, “Interactive GA-based design supportsystem for lighting design in 3-D computer graphics,”Trans. ofIEICE, vol. J81-DII, no. 7, pp. 1601–1608, 1998. (in Japanese).

[8] E. Baker, “Summary: Evolving line drawings,” in 5th Int. Conf.on Genetic Algorithm (ICGA’93), (Urbana-Champaign, IL,USA), p. 627, CA, USA: Morgan Kaufmann Publisher, 1993.

[9] E. Baker and M. Seltzer, “Evolving line drawings,” in GraphicsInterface ’94 (W. A. Davis and B. Joe, eds.), (Banff, Alberta,Canada), pp. 91–100, CA, USA: Morgan Kaufmann Publisher,May 1994.

[10] S. Baluja, D. Pomerleau, and T. Jochem, “Simulating user’spreferences: Towards automated artificial evolution for com-puter generated images,” Tech. Rep. CMU-CS-93–198, CMUComputer Science Technical Reports, Oct. 1993.

[11] S. Baluja, D. Pomerleau, and T. Jochem, “Towards automatedartificial evolution for computer-generated images,” Connec-tion Science, vol. 6, no. 2&3, pp. 325–354, 1994.

[12] W. Banzhaf, “Interactive evolution,” in Handbook of Evolu-tionary Computation (T. Back, D. Fogel, and Z. Michalewicz,eds.), ch. C2.10, pp. 1–5, Oxford University Press, 1997.

[13] J. A. Biles, “GenJam: A genetic algorithm for generating jazzsolos,” in Int. Computer Music Conf. (ICMC’94), (Aarhus,Denmark), pp. 131–137, 1994.

[14] J. A. Biles, “Al Biles Virtual Quintet, genjam.” Compact diskDRK-144, 1995.

15

[15] J. A. Biles and W. Eign, “GenJam populi: Training an IGAvia audience-mediated performance,” in Int. Computer MusicConf. (ICMC’95), (Banff, Canada), pp. 347–348, 1995.

[16] J. A. Biles, P. G. Anderson, and L. W. Loggi, “Neural networkfitness functions for a musical IGA,” in IIA’96/SOCO’96. Int.ICSC Symposia on Intelligent Industrial Automation And SoftComputing, pp. B39–44, 1996.

[17] J. A. Biles, “Interactive GenJam: Integrating real-time per-formance with a genetic algorithm,” in Int. Computer MusicConf. (ICMC’98), (Ann Arbor, MI, USA), pp. 232–235, Oct.1998.

[18] J. A. Biles, “Life with GenJam: Interacting with a musicalIGA,” in IEEE Int. Conf. on System, Man, and Cybernetics(SMC’99), (Tokyo, Japan), pp. 652–656, Oct. 1999. vol.3.

[19] J. A. Biles, “GenJam in perspective: A tentative taxon-omy for genetic algorithm music and art systems,” in Ge-netic and Evolutionary Computation Conf. Workshop Program(GECCO2000), (Las Vegas, NV, USA), pp. 133–135, July 2000.

[20] J. A. Biles, “GenJam: Evolution of a jazz improviser,” in Cre-ative Evolutionary Systems (P. J. Bentley and D. W. Corne,eds.), Morgan Kaufman, 2000.

[21] F. Boschetti and L. Moresi, “Comparison between interac-tive (subjective) and traditional (numerical) inversion by ge-netic algorithms,” in Congress on Evolutionary Computation(CEC2000), (La Jolla, CA, USA), pp. 522–528, July 2000.

[22] F. Boschetti and L. Moresi, “Interactive inversion in geo-sciences,” Geophysics, (in press).

[23] F. Boschetti and H. Takagi, “Visualization of ec landscapeto accelerate ec conversion and evaluation of its effect,” inCongress on Evolutionary Computation (CEC2001), (Seoul,South Korea), May 2001. (submitting).

[24] F. Boschetti, L. Moresi, and K. Covin, “Interactive inversionin geological applications,” in Int. Conf. on Knowledge-basedIntelligent Information Engineering Systems (KES’99), (Ade-laide, Australia), pp. 276–279, Aug./Sept. 1999.

[25] C. Caldwell and V. S. Johnston, “Tracking a criminal suspectthrough “face-space” with a genetic algorithm,” in 4th Int.Conf. on Genetic Algorithm (ICGA’91), (San Diego, CA, US),pp. 416–421, CA, USA: Morgan Kaufmann Publishers, July1991.

[26] S.-B. Cho, “Applying interactive genetic algorithm to content-based image retrieval: A preliminary result,” in Workshopon Interactive Evolutionary Computation, (Fukuoka, Japan),pp. 19–24, Mar. 1998. (in Japanese).

[27] S.-B. Cho and J.-Y. Lee, “Emotional image retrieval withinteractive evolutionary computation,” in Advances in SoftComputing–Engineering Design and Manufacturing (R. Roy,T. Furuhashi, and P. Chawdhry, eds.), pp. 57–66, Springer-Verlag, London, 1999.

[28] S.-B. Cho, “A human-oriented interface for emotional mediamanipulation with an a-life technology,” Three DimensionalImages, vol. 13, no. 3, pp. 71–76, 1999.

[29] S. Das, T. Franguiadakis, M. Papka, T. A. DeFanti, et al.,“A genetic programming application in virtual reality,” in 1stIEEE Conf. on Evolutionary Computation (ICEC’94), vol. 1,(Orlando, FL, USA), pp. 480–484, IEEE, June 1994.

[30] R. Dawkins, The Blind Watchmaker. Essex: Longman, 1986.[31] R. Dawkins, “The evolution of evolvability,” in Artificial Life

(C. G. Langton, ed.), pp. 201–220, Addison-Wesley, 1989.[32] T. Disz, M. E. Papka, and R. Stevens, “UbiWorld: An en-

vironment integrating virtual reality, supercomputing and de-sign,” in 6th Heterogeneous Computing Workshop (HCW’97)(D. Hensgen, ed.), (Geneva, Switzerland), pp. 46–57, IEEEComputer Society Press, Apr. 1997.

[33] G. Dozier, “Evolving robot behavior via interactive evolution-ary computation: From real-world to simulation,” in 16th ACMSymposium on Applied Computing (SAC2001), (Los Angeles,USA), Mar. 2001.

[34] I. K. C. Eckert and M. Stacey, “Interactive generative systemsfor conceptual design: an empirical perspective,” (AI EDAM)Artificial Intelligence for Engineering Design, Analysis andManufacturing, vol. 13, no. 4, pp. 303–320, 1999.

[35] H. Esbensen, “EXPLORER: An interactive floorplanner fordesign space exploration,” in European Design AutomationConf. (EURO-DAC’96) with EURO-VHDL’96 and Exhibition,(Geneva, Switzerland), pp. 356–361, Sept. 1996.

[36] B. Filipic and D. Juricic, “An interactive genetic algorithmfor controller parameter optimization,” in Int. Conf. on Artifi-

cial Neural Nets and Genetic Algorithms, (Innsbruck,Austria),pp. 458–62, Apr. 1993.

[37] M. Fujii and T. Furuhashi, “A study of interactive between hu-mans and systems using fuzzy classifier system,” in Workshopon Interactive Evolutionary Computation, (Fukuoka, Japan),pp. 37–41, Mar. 1998. (in Japanese).

[38] S. Fujii, H. Takagi, M. Ohsaki, and Y. Arai, “Knowledge acqui-sition for compensation for hearing impairment based on IECFitting,” in Technical Report of ASJ, H–99–121, (Kumamoto,Japan), Dec. 1999. (in Japanese).

[39] S. Fujii, H. Takagi, M. Ohsaki, M. Watanabe, and S. Sakamoto,“Evaluation and analysis of IEC Fitting,” in 7th WesternPacific Regional Acoustics Conf. (WESTPRAC VII), (Ku-mamoto, Japan), pp. 369–372, Oct. 2000.

[40] H. Furuta, K. Maeda, and E. Watanabe, “Application of ge-netic algorithm to aesthetic design of bridge structures,” Mi-crocomputers in Civil Engineering, vol. 10, no. 6, pp. 415–421,1995.

[41] D. E. Goldberg,W. B. Hall, L. Krussow, E. Lee, and A. Walker,“Teamwork for a quality education: Low-cost, effective ed-ucational reform, through department-wide competition ofteams,” IlliGAL Report 98005, University of Illinois, Urbana-Champaign, Feb. 1998.

[42] J. Graf and W. Banzhaf, “Interactive evolution of images,” in4th Annual Conf. on Evolutionary Programming (J. R. Mc-Donnell, R. G. Reynolds, and D. B. Fogel, eds.), (San Diego,CA, USA), pp. 53–65, MIT Press, Mar. 1995.

[43] J. Graf, “Interactive evolutionary algorithms in design,” in Int.Conf. on Artificial Neural Nets and Genetic Algorithms, (Ales,France), pp. 227–230, Apr. 1995.

[44] J. Graf and W. Banzhaf, “Interactive evolution for simulatednatural evolution,” in Artificial Evolution. European Conf.(AE’95). Selected Papers. (J. M. Alliot, E. Lutton, E. Ronald,M. Schoenauer, et al., eds.), pp. 259–272, Berlin, Germany:Springer-Verlag, Sept. 1995.

[45] J. Graf andW. Banzhaf, “An expansionoperator for interactiveevolution,” in IEEE Int. Conf. on Evolutionary Computation(ICEC’95), vol. 2, (Perth, WA, Australia), pp. 798–802, IEEE,Nov./Dec. 1995.

[46] J. Graf, “Interactive evolution: A prototyping framework forengineering,” Microcomputers in Civil Engineering, vol. 11,no. 3, pp. 185–193, 1996.

[47] G. R. Greenfield, “New directions for evolving expressions,” in1st Annual Conf. of BRIDGES: Mathematical Connections inArt, Music, and Science, (Winfield, Kansas, USA), July 1998.

[48] F. Gruau and K. Quatramarau, “Cellular encoding for interac-tive evolutionary robotics,” Cognitive Science Research Paper425, University of Sussex, July 1996.

[49] N. Hayashida and H. Takagi, “Introduction of an active selec-tion to the interactive evolutionarycomputation by usingmulti-dimentional data visualization,” in SOFT Kyushu ChapterMeeting, (Saga, Japan), pp. 17–20, Dec. 1999. (in Japanese).

[50] N. Hayashida and H. Takagi, “Visualized IEC: Interactive evo-lutionary computation with multidimensional data visualiza-tion,” in Industrial Electronics, Control and Instrumentation(IECON2000), (Nagoya, Japan), pp. 2738–2743, Oct. 2000.

[51] M. Herdy, “Evolution strategies with subjective selection,” inInt. Conf. on Evolutionary Computation (ICEC/PPSN IV),(Berlin, Germany), pp. 22–26, 1996.

[52] M. Herdy, “Evolutionary optimization based on subjective se-lection – evolving blends of coffee,” in 5th European Congresson Intelligent Techniques and Soft Computing EUFIT’97,(Aachen), pp. 640–644, Sept. 1997.

[53] D. Horowitz, “Generating rhythms with genetic algorithms,” inInt. Computer Music Conf. (ICMC’94), (Aarhus, Denmark),pp. 142–143, 1994.

[54] F.-C. Hsu and J.-S. Chen, “A study on multi criteria decision-making model: Interactive genetic algorithms approach,” inIEEE Int. Conf. on System, Man, and Cybernetics (SMC’99),(Tokyo, Japan), pp. 634–639, Oct. 1999. vol.3.

[55] T. Igarashi, S. Matsuoka, and T. Masui, “Adaptive recogni-tion of implicit structures in human-organized layouts,” in 11thIEEE Int. Symposium on Visual Languages (V. Haarslev, ed.),(Darmstadt, Germany), pp. 258–66, IEEE Comput. Soc. Press,Sept. 1995.

[56] M. Inada and T. Terano, “Analyzing clinical data using interac-tive evolutionary computation and inductive learning,” in 26th

16

Intelligent System Symposium of SICE, (Yokohama, Japan),Mar. 1999. (in Japanese).

[57] T. Ingu, H. Takagi, and M. Ohsaki, “Improvement of inter-face for interactive genetic algorithms -proposal for fast GAconvergence-,” in 13th Fuzzy System Symposium, (Toyama,Japan), pp. 859–862, June 1997. (in Japanese).

[58] T. Ingu and H. Takagi, “Accelerating a GA convergence by fit-ting a single-peak function,” in IEEE Int. Conf. on Fuzzy Sys-tems (FUZZ-IEEE’99), (Seoul, South Korea), pp. 1415–1420,Aug. 1999.

[59] T. Inoue, T. Furuhashi, M. Fujii, et al., “Development of nursescheduling support system using interactive ea,” in IEEE Int.Conf. on Systems, Man, and Cybernetics (SMC’99), (Tokyo,Japan), pp. 533–537, October 1999. vol.5.

[60] Y. Ishino, K. Yoshinaga, and T. Terano, “Data analyses usingsimulated breeding and inductive learning techniques,” in 21stIntelligent System Symposium, (Tokyo, Japan), pp. 1–6, May1995. (in Japanese).

[61] Y. Ishino and T. Terano, “Marketing data analysis using simu-lated breeding and inductive learning techniques,” J. of JapanSociety for Artificial Intelligence, vol. 12, no. 1, pp. 121–131,1997. (in Japanese).

[62] Y. Ishino, K. Hori, and S. Nakasuka, “Interactive knowledgeacquisition for concept development of consumer products,”in 3rd Int. Conf. on Knowledge-Based Intelligent Informa-tion Engineering Systems (KES’99), (Adelaide, SA, Australia),pp. 272–275, Aug./Sept. 1999.

[63] Y. Ishino, K. Hori, and S. Nakasuka, “Strategic concept forma-tion of consumer goods based on knowledge acquisition fromquestionnaire data,” in IEEE Int. Conf. on Systems, Man,and Cybernetics (SMC’99), (Tokyo, Japan), pp. 533–537, Oct.1999. vol.5.

[64] K. Isono, M. Oda, and S. Akamatsu, “A facial image generat-ing system based on a genetic algorithm-gene rating subjectiveimpression,” Transactions of the Institute of Electronics, In-formation and Communication Engineers D-II, vol. J82D-II,no. 3, pp. 483–493, 1999. (in Japanese).

[65] M. Iwai, “Adjustment of FM operator tones by genetic algo-rithm,” in Summer Program Symposium on Amusement andComputer, July 1994. (in Japanese).

[66] T. Iwasaki, A. Kimura, Y. Todoroki, Y. Hirose, H. Takagi, andT. Takeda, “Interactive virtual aquarium (1st report),” in 5thAnnual Conf. of the Virtual Reality Society of Japan, pp. 141–144, Sept. 2000. (in Japanese).

[67] I. Kamihira, M. Yamaguchi, and H. Kita, “Online adaptationof vehicles by means of an evolutionary control system,” inIEEE Int. Conf. on Systems, Man, and CyberneticsSMC’99),(Tokyo, Japan), pp. 553–558, Oct. 1999.

[68] S. Kamohara, H. Takagi, and T. Takeda, “Control rule acqui-sition for arm wrestling in virtual reality,” in 12th Fuzzy Sys-tem Symposium, (Tokyo, Japan), pp. 487–490, June 1996. (inJapanese).

[69] S. Kamohara, H. Takagi, and T. Takeda, “Control rule acqui-sition for an arm wrestling robot,” in IEEE Int. Conf. on Sys-tem, Man, Cybernetics (SMC’97), vol. 5, (Orlando, FL, USA),pp. 4227–4231, Oct. 1997.

[70] S. Kato, “Feature selection in flexible retrieval,” Transactionsof the Institute of Electronics, Information and Communica-tion Engineers D-II, vol. J80D-II, no. 2, pp. 598–606, 1997. (inJapanese).

[71] S. Kato and S. Iisaku, “An image retrieval method based on agenetic algorithm,” in Twelfth Int. Conf. on Information Net-working (ICOIN-12), (Tokyo, Japan), pp. 333–336, Jan. 1998.

[72] N. Katsuyama and H. Yamakawa, “A study on heredity andevolution of designs considering KANSEI by using genetic al-gorithms,” in Japan Society of Mechanical Engineering 3rd An-nual Meeting of Design Enginering & System, (Tokyo, Japan),pp. 43–47, May 1993. (in Japanese).

[73] Y. Katurada and Y. Maeda, “Support system for automaticallyemotional generation using interactive EC,” in 16th Fuzzy Sys-tem Symposium, (Akita, Japan), pp. 297–300, Sept. 2000. (inJapanese).

[74] H.-S. Kim and S.-B. Cho, “Development of an IGA-based fash-ion design aid system with domain specific knowledge,” in Proc.of IEEE SMC’99, (Tokyo, Japan), pp. III–663–III–668, Oct.1999.

[75] H.-S. Kim and S.-B. Cho, “Genetic algorithm with knowledge-based encoding for interactive fashion design,” in Lecture Notes

in Artificial Intelligence, Springer Valg, 2000. no. 1886, pp.404-414.

[76] H.-S. Kim and S.-B. Cho, “An emotion-based fashion designaid system using interactivity,” the Korea Information ScienceSociety (B), vol. 27, no. 9, pp. 942–951, 2000. (in Korean).

[77] H.-S. Kim and S.-B. Cho, “Knowledge-based encoding in in-teractive genetic algorithm for a fashion design aid system,” inProc. of GECCO, Poster Papers, p. 757, July 2000.

[78] H.-S. Kim and S.-B. Cho, “Application of interactive geneticalgorithm to fashion design,” Engineering Applications of Ar-tificial Intelligence, vol. 13, no. 6, pp. 635–644, 2000.

[79] A. Kimura, T. Iwasaki, Y. Kitajima, H. Takagi, and T. Takeda,“Development of fish shape generator based on interactive evo-lutionary computation,” in Joint Technical Meeting of KyushuChapters of Electric Related Societies, (Fukuoka, Japan),p. 525, September 2000. (in Japanese).

[80] K. Kishi and H. Takagi, “Developingmontage system for studyon interactive evolutionary computation,” in SOFT 3rd Work-shop on Evaluation of Heart and Mind, (Gamagoori, Aichi,Japan), pp. 15–18, Nov. 1998. (in Japanese).

[81] K. Kishi and H. Takagi, “Evaluation of on-line knowledgeembedding for interactive evolutionary computation,” in 15thFuzzy System Symposium, (Osaka, Japan), pp. 379–380, June1999. (in Japanese).

[82] A. Kitamoto and M. Takagi, “Learning criteria for similarity-based image retrieval using simulated breeding by pipeline-typegenetic algorithms,” in Technical Report of IEICE, HIP96-4,pp. 17–22, June 1996. (in Japanese).

[83] A. Kitamoto and M. Takagi, “Interactive image browsing usingpipeline-type genetic algorithm,” in Workshop on InteractiveEvolutionary Computation, (Fukuoka, Japan), pp. 31–36, Mar.1998. (in Japanese).

[84] A. Kitamoto and M. Takagi, “Interactive image browsing us-ing queue-based genetic algorithm,” J. of Japanese Society forArtificial Intelligence, vol. 13, no. 5, pp. 728–738, 1998. (inJapanese).

[85] M. Korenaga and M. Hagiwara, “An interior layout supportsystem with interactive evolutionary computation,” Transac-tions of the Information Processing Society of Japan, vol. 41,no. 11, pp. 3152–3160, 2000. (in Japanese).

[86] A. Kosorukoff and D. Goldberg, “Genetic algorithms for socialinnovation and creativity,” IlliGAL Report 2001005, Universityof Illinois, Urbana-Champaign, Jan. 2001.

[87] J. Kotani and M. Hagiwara, “An evolutionary design-support-system with tree structure representation,” in 16th Fuzzy Sys-tem Symposium, (Akita, Japan), pp. 327–330, Sept. 2000. (inJapanese).

[88] J. Kotani and M. Hagiwara, “An evolutionary design-support-system with structural representation,” in IEEE Int. Conf.on Industrial Electronics, Control and Instrumentation 2000(IECON2000), (Nagoya, Japan), pp. 672–677, Oct. 2000.

[89] N. Kubota, K. Watanabe, and F. Kojima, “Trajectory gen-eration of human-friendly robots using interactive genetic al-gorithm,” in 15th Fuzzy System Symposium, (Osaka, Japan),pp. 355–358, June 1999. (in Japanese).

[90] K. Kuriyama and T. Terano, “Interactive story compositionsupport by genetic algorithms,” in World Conf. on Artifi-cial Intelligence in Education. (AI-ED’97), (Kobe, Japan),pp. 615–617, Aug. 1997.

[91] K. Kuriyama, T. Terano, and M. Numao, “Authoring sup-port by interactive genetic algorithm and case based reason-ing,” in Workshop on Interactive Evolutionary Computation,(Fukuoka, Japan), pp. 13–18, Mar. 1998. (in Japanese).

[92] K. Kuriyama, T. Terano, and M. Numao, “Authoring supportby interactive genetic algorithm and case base retrieval,” inIEEE 2nd Int. Conf. on Conventional and Knowledge-basedIntelligent Electronic Systems (KES’98), (Adelade, SA, Aus-tralia), pp. pp.485–488, Apr. 1998.

[93] K. Kuriyama, T. Terano, and M. Numao, “Authoring supportby interactive genetic algorithm and case base retrieval,” inAsia Fuzzy System Symposium, (Masan, Korea), pp. pp.485–488, June 1998.

[94] J.-Y. Lee and S.-B. Cho, “Interactive genetic algorithm forcontent-based image retrieval,” in Asia Fuzzy System Sympo-sium (AFSS’98), (Masan, Korea), pp. pp.479–484, June 1998.

[95] J.-Y. Lee and S.-B. Cho, “Interactive genetic algorithm withwavelet coefficients for emotional image retrieval,” in Proc. Int.

17

Conf. on Soft Computing, (Iizuka, Fukuoka, Japan), pp. 829–832, Oct. 1998.

[96] J.-Y. Lee and S.-B. Cho, “Application of interactive genetic al-gorithm to image retrieval based on emotion,” the Korea Infor-mation Science Society (B), vol. 26, no. 3, pp. 422–430, 1999.(in Korean).

[97] J.-Y. Lee and S.-B. Cho, “Incorporating human preference intocontent-based image retrieval using interactive genetic algo-rithm,” in Proc. of GECCO’99, Poster Papers, p. 1788, July1999.

[98] J.-Y. Lee and S.-B. Cho, “Sparse fitness evaluation for reduc-ing user burden in interactive genetic algorithm,” in Proc. ofFUZZ-IEEE’99, pp. II–998–II1003, Aug. 1999.

[99] D. Levine, M. Facello, P. Hallstrom, et al., “Stalk: an inter-active system for virtual molecular docking,” IEEE Computa-tional Science and Engineering, vol. 4, no. 2, pp. 55–65, 1997.

[100] M. A. Lewis, A. H. Fagg, and A. Solidum, “Genetic program-ming approach to the construction of a neural network for con-trol of a walking robot,” in IEEE Int. Conf. on Robotics andAutomation, vol. 3, (Nice, France), pp. 2618–2623, May 1992.

[101] I. S. Lim, “Evolving facial expressions,” in IEEE Int. Conf. onEvolutionary Computation (ICEC’95), (Perth,WA, Australia),pp. 515–520, Nov./Dec. 1995. vol.2.

[102] I. S. Lim, “Evolutionary art in virtual world,” in 2nd EU-ROGRAPHICS Workshop on Virtual Environments, (MonteCarlo, Monaco), 1995.

[103] I. S. Lim and D. Thalmann, “Pro-actively interactive evolutionfor computer animation,” in Eurographics Workshop on An-imation and Simulation (CAS’99), (Milan, Italy), pp. 45–52,Sept. 1999.

[104] S. J. Louis and R. Tang, “Interactive genetic algorithms forthe traveling salesman problem,” in Genetic and Evolution-ary Computation Conf. (ICGA-99), (Orlando, FL, USA),pp. 1043–1048, July 1999. vol.1.

[105] H. H. Lund, L. Pagliarini, and P. Miglino, “Artistic design withGA & NN,” in 1st Nordic Workshop on Genetic Algorithmsand Their Applications (1NWGA), (Vaasa, Finland), pp. 97–105, Jan. 1995.

[106] H. H. Lund and O. Miglino, “Evolving and breedingrobots,” in 1st European Workshop on Evolutionary Robotics(EboRobot98), (Berlin, Germany), pp. 192–210, Springer-Verlag, Apr. 1998.

[107] H. H. Lund, O. Miglino, L. Pagliarini, A. Billard, andA. Ijspeert, “Evolutionary robotics – a childrs’s game,” in IEEEInt. Conf. on Evolutionary Computation (ICEC’98), pp. 154–158, May 1998.

[108] J. Martinez and S. Marchand, “Towards intelligent retrieval inimage databases,” in Int. Workshop on Multi-Media DatabaseManagement Systems, (Dayton, OH, USA), pp. 38–45, Aug.1998.

[109] T. Masui, “Interactive graphic object layout with genetic algo-rithms,” in Human Interface, 41-6, pp. 41–48, Mar. 1992. (inJapanese).

[110] T. Masui, “Graphic object layout with interactive genetic algo-rithms,” in IEEE Workshop on Visual Languages, (Los Alami-tos, CA, USA), pp. 74–80, 1992.

[111] T. Masui, “Evolutionary learning of graph layout constraintsfrom examples,” in UIST’94. 7th Annual Symposium on UserInterface Software and Technology, (Marina del Rey, CA,USA), pp. 103–8, ACM, Nov. 1994.

[112] J. McCormack, “Interactive evolution of L-system grammarsfor computer graphics modelling,” in Complex Systems: fromBiology to Computation (D. G. Green and T. Bossomaier, eds.),pp. 118–130, Amsterdam, Netherlands: IOS Press, Aug./Sept.1993.

[113] T. Miyazaki and M. Hagiwara, “A poster creating support sys-tem to reflect Kansei,” Transactions of the Information Pro-cessing Society of Japan, vol. 38, no. 10, pp. 1928–1936, 1997.(in Japanese).

[114] N. Monmarche, G. Nocent, M. Slimane, G. Venturini, andP. Santini, “Imagine: a tool for generating HTML style sheetswith an interactive genetic algorithm based on genes frequen-cies,” in IEEE Int. Conf. on System, Man, and Cybernetics(SMC’99), (Tokyo, Japan), pp. 640–645, Oct. 1999. vol.3.

[115] A. Moroni, J. Manzolli, F. von Zuben, and R. Gudwin, “Evo-lutionary computation applied to algorithmic composition,” inCongress on Evolutionary Computation (CEC’99), (Washing-ton, DC, USA), pp. 807–811, July 1999. vol.2.

[116] T. Mutoh, N. Komagata, and K. Ueda, “An experimental studyfor automatically generating image filter sequence by usingsimulated breeding,” in Workshop on Interactive Evolution-ary Computation, (Fukuoka, Japan), pp. 7–12, Mar. 1998. (inJapanese).

[117] S. Nabeta and T. Terano, “Design support system about time-varying multimedia titles by interactive genetic algorithm,” in12th Annual Conf. of JSAI, (Yokohama, Japan), pp. 489–490,June 1998. (in Japanese).

[118] M. Nagao, M. Yamada, K. Suzuki, and A. Ohuchi, “Evalua-tion of the image retrieval system using interactive evolution-ary computation,” in Workshop on Interactive EvolutionaryComputation, (Fukuoka, Japan), pp. 25–30, Mar. 1998. (inJapanese).

[119] M. Nagao, M. Yamamoto, K. Suzuki, and A. Ohuchi, “Eval-uation of the image retrieval system using interactive geneticalgorithm,” J. of Japanese Society for Artificial Intelligence,vol. 13, no. 5, pp. 720–727, 1998. (in Japanese).

[120] Y. Nakanishi, “Applying evolutionary systems to design aidsystem,” in ALIFE V, Poster Presentations, pp. PP–25, 147–154, May 1996.

[121] Y. Nakanishi, “Capturing preference into a function using in-teractions with a manual evolutionary design aid system,” inLate Breaking Papers at Genetic Programming, pp. 133–140,July 1996.

[122] Y. Nakanishi, “Control and evaluation of interactive evolution-ary system with a preference function: applying genetic pro-gramming to design aid system,” J. of Japanese Society forArtificial Intelligence, vol. 13, no. 5, pp. 704–711, 1998. (inJapanese).

[123] G. L. Nelson, “Sonomorphs: An application of genetic algo-rithms to growth and development of musical organisms,” in4th Biennial Art & Technology Symposium, (New London, CT,USA), pp. 155–169, Mar. 1993.

[124] G. L. Nelson, “Further adventures of the sonomorphs,” in 5thBiennial Art & Technology Symposium, (New London, CT,USA), pp. 51–64, Mar. 1995.

[125] T. C. Nguyen and T. S. Huang, “Evolvable modeling: Designadaptation through hierarchical evolution for 3-D model-basedvision.”TCGA File No.02852, (unpublished tech. report), 1993.

[126] T. C. Nguyen and T. S. Huang, “Evolvable 3D modeling formodel-based object recognition systems,” in Advances in GP,ch. 22, pp. 459–475, Cambridge, MA, USA: MIT Press, 1994.

[127] H. Nishino, H. Takagi, and K. Utsumiya, “A digital prototypingsystem for designing novel 3D geometries,” in 6th Int. Conf. onVirtual Systems and MultiMedia (VSMM2000), (Ogaki, Gifu,Japan), pp. 473–482, Oct. 2000.

[128] H. Nishino, H. Takagi, and K. Utsumiya, “3D modeling supportmechanism by using interactive evolutionary computation,” inTech. Report of IEICE, Vol.100, No.461, IE2000-78, pp. 1–8,Nov. 2000. (in Japanese).

[129] H. Nishino, H. Takagi, S.-B. Cho, and K. Utsumiya, “A 3Dmodeling system for creative design,” in The 15th Int. Conf.on Information Networking (ICOIN-15), (Beppu, Japan),Jan./Feb. 2001.

[130] K. Nishio, M. Murakami, E. Mizutani, and N. Honda, “Efficientfuzzy fitness assignment strategies in an interactive genetic al-gorithm for cartoon face search,” in 6th Int. Fuzzy SystemsAssociation World Congress (IFSA’95), vol. 1, (Sao Paulo),pp. 173–176, July 1995.

[131] K. Nishio, M. Murakami, E. Mizutani, and N. Honda, “Fuzzyfitness assignment in an interface genetic algorithm for a car-toon face search,” in Genetic Algorithm and Fuzzy Logic Sys-tems, Soft Computing Perspectives (E. Sanchez, T. Shibata,and L. A. Zadeh, eds.), vol. 7 of Advances in Fuzzy Systems -Applications and Theory, pp. 175–192, World Scientific, 1997.

[132] T. Noda, H. Takagi, and M. Zhang, “Construction of a psy-chological factor space for an IGA-based image retrieval sys-tem and subjective test for the system,” in 15th Fuzzy Sys-tem Symposium, (Osaka, Japan), pp. 683–684, June 1999. (inJapanese).

[133] T. Noda, D. Zhao, H. Takagi, and A. Teraoka, “Media databaseretrieval and conversion system based on impression,” in SOFTKyushu Chapter Meeting, (Saga, Japan), pp. 1–4, Dec. 1999.(in Japanese).

[134] T. Noda, D. Zhao, and H. Takagi, “Music database retrievaland media conversion system based on impression,” in 6th Int.

18

Conf. on Soft Computing (IIZUKA2000), (Iizuka, Fukuoka,Japan), pp. 151–156, Oct. 2000.

[135] U.-M. O’Reilly and G. Ramachandran, “A preliminary investi-gation of evolution as a form design strategy,” in Artificial LifeVI (C. Adami, R. Belew, H. Kitano, and C. Taylor, eds.), (LosAngeles, CA, UAS), pp. 443–447, MIT Press, June 1998.

[136] T. Obata and M. Hagiwara, “A color poster creating supportsystem to reflect Kansei,”Transactions of the Information Pro-cessing Society of Japan, vol. 41, no. 3, pp. 701–710, 2000. (inJapanese).

[137] G. Ochoa, “On genetic algorithms and Lindenmayer systems,”in 5th Int. Conf. on Parallel Problem Solving from Nature(PPSN V), (Amsterdam, Netherlands), pp. 335–344, Sept.1998.

[138] M. Ohsaki and H. Takagi, “Applicationof interactive evolution-ary computation to digital hearing aids fitting,” in 14th FuzzySystem Symposium, (Gifu, Japan), pp. 193–194, June 1998. (inJapanese).

[139] M. Ohsaki, H. Takagi, and K. Ohya, “An input method usingdiscrete fitness values for interactive GA,” J. of Intelligent andFuzzy Systems, vol. 6, pp. 131–145, 1998.

[140] M. Ohsaki and H. Takagi, “Reduction of the fatigue of hu-man interactive EC operators-improvement of present interfaceby prediction of evaluation order,” J. of Japanese Society forArtificial Intelligence, vol. 13, no. 5, pp. 712–719, 1998. (inJapanese).

[141] M. Ohsaki, H. Takagi, and T. Ingu, “Methods to reduce thehuman burden of interactive evolutionary computation,” inAsia Fuzzy System Symposium (AFSS’98), (Masan, Korea),pp. 495–500, June 1998.

[142] M. Ohsaki and H. Takagi, “Improvement of presenting inter-face by predicting the evaluation order to reduce the burdenof human interactive EC operators,” in IEEE Int. Conf. onSystem, Man, Cybernetics (SMC’98), (San Diego, CA, USA),pp. 1284–1289, Oct. 1998.

[143] M. Ohsaki and H. Takagi, “Applicationof interactive evolution-ary computation to optimal tuning of digital hearing aids,” inInt. Conf. on Soft Computing (IIZUKA’98), (Iizuka, Fukuoka,Japan), pp. 849–852, World Scientific, Oct. 1998.

[144] M. Ohsaki, T. Tsumura, H. Takagi, and M. Shimada, “Evalua-tion of IEC fitting system on speech hearing,” in Annual Meet-ing of Acoustic Society of Japan, 1-2-19, pp. 361–362, Mar.1999. (in Japanese).

[145] M. Ohsaki, T. Tsumura, H. Takagi, and M. Shimada, “Applica-tion of IEC fitting system to music hearing,” in Annual Meetingof Acoustic Society of Japan, 1-2-20, pp. 363–364, Mar. 1999.(in Japanese).

[146] M. Ohsaki and H. Takagi, “Evaluation of an interactive EC-based fitting system for digital hearing aids,” in 15th FuzzySystem Symposium, (Osaka, Japan), pp. 381–384, June 1999.(in Japanese).

[147] M. Ohsaki, H. Takagi, M. Watanabe, and S. Sakamoto, “De-velopment and evaluation of an IEC-based hearing aids fittingsystem for practical use – application to a conventional hearingaid,” in Annual Meeting of Acoustic Society of Japan, 2-10-1,pp. 331–332, Mar. 2000. (in Japanese).

[148] M. Ohsaki and H. Takagi, “Design and development of an IEC-based hearing aids fitting system,” in 4th Asia Fuzzy SystemSymposium (AFSS’00), (Tsukuba, Japan), pp. 543–548, June2000.

[149] M. Ohsaki and H. Takagi, “Human interface of interactive evo-lutionary computation and its evaluation,” in Genetic Algo-rithms 4 (H. Kitano, ed.), ch. 13, pp. 397–438, Sangyo Pub-lisher, Aug. 30, 2000. (in Japanese).

[150] T. Onisawa and T. Anzai, “Acquisition of intelligible fuzzyrules,” in IEEE Int. Conf. on Systems, Man, and Cybernet-ics (SMC’99), (Tokyo, Japan), pp. 268–273, Oct. 1999. vol.5.

[151] T. Onisawa and T. Anzai, “Acquisitionof intelligible fuzzy ruleswith interaction,” in 16th Fuzzy System Symposium, (Akita,Japan), pp. 451–454, Sept. 2000. (in Japanese).

[152] T. Onisawa, W. Takizawa, and M. Unehara, “Composition ofmelody reflecting user’s feeling,” in IEEE Int. Conf. on Indus-trial Electronics, Control and Instrumentation (IECON2000),(Nagoya, Japan), pp. 2738–2743, Oct. 2000.

[153] K. Otoba, K. Tanaka, and M. Hitafuji, “Image processing andinteractive selection with Java based on genetic algorithms,” in3rd IFAC/CIGR Workshop on Artificial Intelligence in Agri-culture, (Makuhari, Japan), pp. 83–88, Apr. 1998.

[154] L. Pagliarini, H. H. Lund, O. Miglino, and D. Parisi, “ArtificialLife: A new way to build educational and therapeutic games,”in ALIFE V, pp. 152–156, MIT Press, 1996.

[155] L. Pagliarini, A. Dolan, F. Menczer, and H. H. Lund, “AL-ife meets web: Lessons learned,” in 1st Int. Conf. on VirtualWorld (VW’98), (Berlin, Germany), pp. 156–167, Springer-Verlag, July 1998.

[156] M. E. Papka and R. Stevens, “Ubiworld: An environment in-tegrating virtual reality, supercomputing and design,” in 5thIEEE Int. Symposium on High Performance Distributed Com-puting, (Syracuse, NY, USA), pp. 306–307, IEEE Comput. Soc.Press, Aug. 1996.

[157] I. C. Parmee, “The concrete arch dam: An evolutionary modelof the design process,” in Int. Conf. on Artificial Neural Netsand Genetic Algorithms, (Innsbruck, Austria), pp. 544–551,Apr. 1993.

[158] I. C. Parmee and C. R. Bonham, “The maintenance of searchdiversity for effective design space decomposition using clusteroriented genetic algorithms (COGAs) and multi-agent strate-gies (GAANT),” inAdaptive Computing in Engineering Designand Control (I. C. Parmee, ed.), (Plymouth, UK), pp. 128–138,1996.

[159] I. C. Parmee, “Strategies for the integration of evolutionary al-gorithms (COGAs) for the identification of high-performanceregions design spaces,” in Evolutionary Algorithms in Engi-neering Applications (D. Dasgupta and Z. Michelewicz, eds.),pp. 453–478, Berlin, Germany: Springer-Varlag, 1996.

[160] I. C. Parmee and C. R. Bonham, “Supporting innovative andcreative design using interactive designer/evolutionary com-puting strategies,” in Computation Models of Creative DesignConf. (J. Gero and M. Maher, eds.), (Sydney, Australia), 1998.

[161] I. C. Parmee and C. R. Bonham, “Cluster-oriented geneticalgorithms to support interactive designer/evolutionary com-puting systems,” in Congress on Evolutionary Computation(CEC’99), vol. 1, (Washington, DC, USA), pp. 546–553, July1999.

[162] I. C. Parmee, “Exploring the design potential of evolutionarysearch, exploration and optimization,” in Evolutionary Designby Computer, (San Farncisco, CA,USA), Morgan Kaufmann,1999.

[163] I. C. Parmee and C. R. Bonham, “Towards the sup-port of innovative conceptual design through interactive de-signer/evolutionary completing strategies,” (AI EDAM) Arti-ficial Intelligence for Engineering Design, Analysis and Man-ufacturing, vol. 14, no. 1, pp. 3–16, 2000.

[164] I. C. Parmee, A. H. Watson, D. Cvetkovic, and C. R. Bonham,“Multiobjective satisfaction within an interactive evolutionarydesign environment,” Evolutionary Computation, vol. 8, no. 2,pp. 197–222, 2000.

[165] I. C. Parmee, “Developing the framework for an interactiveevolutionarydesign system,” in 5th Joint Conf. on InformationSciences (JCIS2000), (Atlantic City, NJ, USA), pp. 991–994,Feb./Mar. 2000.

[166] R. Poli and S. Cagnoni, “Genetic programming with user-driven selection: Experiments on the evolution of algorithmsfor image enhancement,” in 2nd Annual Conf. on Genetic Pro-gramming, pp. 269–277, 1997.

[167] F. Raynal, E. Lutton, P. Collet, and M. Schoenauer, “Manipu-lation of non-linear ifs attractors using genetic programming,”in Congress on Evolutionary Computation (CEC’99), (Wash-ington, DC, USA), pp. 1171–1177, July 1999. vol.2.

[168] A. Rowbottom, “Evolutionary art and form,” in EvolutionaryDesign by Computers (P. Bentley, ed.), ch. 11, pp. 260–277,CA, USA: Morgan Kaufmann Publishers, 1999.

[169] G. Rudolph, “On interactive evolutionary algorithms andstochastic Mealy automata,” in 4th Parallel Problem Solvingfrom Nature - PPSN IV. Int. Conf. on Evolutionary Compu-tation, (Berlin, Germany), pp. 218–226, Sept. 1996.

[170] T. Sato and M. Hagiwara, “Tool creating support system usingevolutionary techniques,” in 15th Fuzzy System Symposium,(Osaka, Japan), pp. 363–366, June 1999. (in Japanese).

[171] T. Sato and M. Hagiwara, “IDSET: Interactive design systemusing evolutionary techniques,” J. of Computer-Aided Design,(accepted).

[172] Y. Sato, “Voice conversation using evolutionary computationof prosodic control,” in 12th Symposium on Human Interface,(Yokohama, Japan), pp. 469–475, Oct. 1996.

[173] Y. Sato, “Voice conversationusing evolutionary computation of

19

prosodic control,” in Intelligent Processing of Manufacturing ofMaterials ’97, pp. 342–348, July 1997.

[174] Y. Sato, “Voice conversation using evolutionary computationof prosodic control,” in Symposium on Artificial Life and itsApplications, pp. 71–78, Dec. 1997. (in Japanese).

[175] R. Sawa and M. Hagiwara, “A system for evolving 3D machinestructure using CG,” in 16th Fuzzy System Symposium, (Akita,Japan), pp. 321–324, Sept. 2000. (in Japanese).

[176] M. Shibuya, H. Kita, and S. Kobayashi, “Generation of pass-motion by hands using interactive genetic algorithms,” in 26thIntelligent System Symposium of SICE, (Yokohama, Japan),Mar. 1999. (in Japanese).

[177] M. Shibuya, H. Kita, and S. Kobayashi, “Integration of multi-objective and interactive genetic algorithms and its applicationto animation design,” in IEEE Int. Conf. on System, Man,and Cybernetics (SMC’99), (Tokyo, Japan), pp. 646–651, Oct.1999. vol.3.

[178] H. Shiraki and H. Saito, “An interactive image retrieval systemusing genetic algorithms,” in Int. Conf. on Virtual Systemsand MultiMedia (VSMM’96), (Gifu, Japan), pp. 257–262, Sept.1996.

[179] K. Sims, “Artificial evolution for computer graphics,” Com-puter Graphics, ACM SIGGRAPH Conf. Proc., vol. 25,pp. 319–328, July/Aug. 1991.

[180] K. Sims, “Interactive evolution of dynamical systems,” in 1stEuropean Conf. on Artificial Life (F. J. Varela and P. Bourgine,eds.), (Paris, France), MIT Press, Dec. 1991.

[181] K. Sims, “Interactive evolution of equations for proceduralmodels,” The Visual Computer, vol. 9, no. 8, pp. 466–476, 1993.

[182] J. R. Smith, “Designing biomorphs with an interactive ge-netic algorithm,” in 4th Int. Conf. on Genetic Algorithms(ICGA’91), (San Diego, CA, USA), pp. 535–538, CA, USA:Morgan Kaufmann Publishers, July 1991.

[183] H. P. S. Snijder and R. Daru, “GECAD: A creativity support-ing sketching tool for genetic layout generation,” in 14th Int.Congress on Cybernetics, (Namur, Belgium), pp. 87–92, Aug.1995.

[184] F. Sugimoto, K. Nishio, and N. Honda, “A method of fitnessassignment in interactive genetic algorithm using fuzzy reason-ing based on psychological measure,” J. of Japanese Societyfor Artificial Intelligence, vol. 13, no. 5, pp. 739–745, 1998. (inJapanese).

[185] F. Sugimoto and N. Honda, “A human interface to searchand draw facial images in mind by using psychometrical spacemodel of faces,” in IEEE Int. Conf. on Fuzzy Systems (FUZZ-IEEE’99), vol. 3, (Seoul, South Korea), pp. 1585–1590, Aug.1999.

[186] F. Sugimoto andM. Yoneyama, “Robustness against instabilityof sensory judgment in a human interface to draw a facial imageusing a psychometrical space model,” in IEEE Int. Conf. onMultimedia and Expo. (ICME2000), vol. 2, (New York, NY,USA), pp. 635–638, July/Aug. 2000.

[187] Y. Sun and Z. Wang, “Interactive algorithm of large scale mul-tiobjective 0-1 linear programming,” in Large Scale Systems:Theory and Applications 1992. 6th IFAC/IFORS/IMACSSymposium, (Beijing, China), pp. 83–86, Oxford, UK: Perg-amon, August 1992.

[188] M. Tabuchi and T. Taura, “Methodology for interactive knowl-edge acquisition between genetic learning engine and human,”J. of Japan Society for Artificial Intelligence, vol. 11, no. 4,pp. 600–607, 1996. (in Japanese).

[189] H. Takagi and K. Ohya, “Improvement of input interface forinteractive genetic algorithms and its evaluation,” in IEICED-156, p. 156, Mar. 1996. (in Japanese).

[190] H. Takagi and K. Ohya, “Discrete fitness values for improv-ing the human interface in an interactive GA,” in IEEE 3rdInt. Conf. on Evolutionary Computation (ICEC’96), (Nagoya,Aichi, Japan), pp. 109–112, May 1996.

[191] H. Takagi, K. Ohya, and M. Ohsaki, “Improvement of inputinterface for interactive genetic algorithms and its evaluation,”in 12th Fuzzy System Symposium, (Tokyo, Japan), pp. 513–516,June 1996. (in Japanese).

[192] H. Takagi, “System optimization without numerical target,” in1996 Biennial Conf. of the North American Fuzzy Informa-tion Processing Society (NAFIPS’96), (Berkeley, CA, USA),pp. 351–354, June 1996.

[193] H. Takagi, K. Ohya, and M. Ohsaki, “Improvement of inputinterface for interactive GA and its evaluation,” in Int. Conf.

on Soft Computing (IIZUKA’96), (Iizuka, Fukuoka, Japan),pp. 490–493, World Scientific, October and September 1996.

[194] H. Takagi, “Interactive GA for system optimization: Problemsand solution,” in 4th European Congress on Intelligent Tech-niques and Soft Computing (EUFIT’96), (Aachen, Germany),pp. 1440–1444, Sept. 1996.

[195] H. Takagi, T. Unemi, and T. Terano, “Perspective on inter-active evolutionary computing,” J. of Japanese Society forArtificial Intelligence, vol. 13, no. 5, pp. 692–703, 1998. (inJapanese).

[196] H. Takagi and K. Aoki, “Interactive evolutionary computation:from creativity support to engineering applications,” in Work-shop on Interactive Evolutionary Computation, (Fukuoka,Japan), pp. 1–6, Mar. 1998. (in Japanese).

[197] H. Takagi, M. Ohsaki, and T. Ingu, “The methods to reducethe burden of human interactive EC operators,” in Workshopon Interactive Evolutionary Computation, (Fukuoka, Japan),pp. 47–52, Mar. 1998. (in Japanese).

[198] H. Takagi, “Interactive evolutionary computation: System op-timization based on human subjective evaluation,” in IEEE Int.Conf. on Intelligent Engineering Systems (INES’98), (Vienna,Austria), pp. 1–6, September 1998.

[199] H. Takagi, “Interactive evolutionary computation — coopera-tion of computational intelligence and human KANSEI —,” inInt. Conf. on Soft Computing (IIZUKA’98), (Iizuka, Fukuoka,Japan), pp. 41–50, World Scientific, October and September1998.

[200] H. Takagi and M. Ohsaki, “Automatic optimization of hearingaids based on how user hears,” in Annual Meeting of AcousticSociety of Japan, 1-2-18, (Tokyo, Japan), pp. 359–360, Mar.1999. (in Japanese).

[201] H. Takagi, S. Kamohara, and T. Takeda, “Introduction of Softcomputing techniques to welfare devices,” in IEEE Midnight -Sun Workshop on Soft Computing Methods in Industrial Appli-cations (SMCia/99), (Kuusamo, Finland), pp. 116–121, June1999.

[202] H. Takagi, S.-B. Cho, and T. Noda, “Evaluation of an IGA-based image retrieval system using wavelet coefficients,” inIEEE Int. Conf. on Fuzzy Systems (FUZZ-IEEE’99), (Seoul,South Korea), pp. 1775–1778, Aug. 1999.

[203] H. Takagi and K. Kishi, “On-line knowledge embedding foran interactive EC-based montage system,” in 3rd Int. Conf.on Knowledge-Based Intelligent Information Engineering Sys-tems (KES’99), (Adelaide, Australia), pp. 280–283, Aug./Sept.1999.

[204] H. Takagi and M. Ohsaki, “IEC-based hearing aids fitting,” inIEEE Int. Conf. on System, Man, and Cybernetics (SMC’99),(Tokyo, Japan), pp. 657–662, Oct. 1999.

[205] H. Takagi, “Auto-optimization for sytems based on kanse:from art to engineering and ausement,” in Fukuoka Academy–Industry Joint Plaza’99, (Fukuoka, Japan), pp. 64–69, Oct.1999. (in Japanese).

[206] H. Takagi, T. Noda, and S.-B. Cho, “Psychological space tohold impression among media in common for media databaseretrieval system,” in IEEE Int. Conf. on System, Man, and Cy-bernetics (SMC’99), (Tokyo, Japan), pp. 263–268, Oct. 1999.

[207] H. Takagi, “Active user intervention in an EC search,” in 5thJoint Conf. on Information Sciences (JCIS2000), (AtlanticCity, New Jersey, USA), pp. 995–998, Feb./Mar. 2000.

[208] H. Takagi, T. Unemi, and T. Terano, “Interactive evolutionarycomputation,” inGenetic Algorithms 4 (H. Kitano, ed.), ch. 11,pp. 325–361, Sangyo Publisher, Aug. 2000. (in Japanese).

[209] H. Takagi andM. Ohsaki, “IEC fitting: New framework of hear-ing aid fitting based on computational intelligence technologyand user’s preference for hearing,” in Poster session PB9, Int.Hearing Aid Research Conf. (IHCON2000), (Lake Tahoe, CA,USA), pp. 49–50, Aug. 2000.

[210] H. Takagi, “Computational intelligence with human capa-bility,” in Int. Symposium on Computational Intelligence(ISCI2000) (P. Sincak and J. Vascak, eds.), Kosice, Slovakia:Springer-Verlag, Aug./Sept. 2000.

[211] H. Takagi, “Fusion technologyof neural networks and fuzzy sys-tems: A chronicledprogression from the laboratory to our dailylives,” Int. J. of Applied Mathematics and Computer Science,vol. 10, no. 4, pp. 647–673, 2001.

[212] T. Tanino, M. Tanaka, and C. Hojo, “An interactive multicri-teria decision making method by using a genetic algorithm,”in 36th Annual Conf. of the Institute of Systems, Control and

20

Information Engineers (ISCIE), (Kyoto, Japan), pp. 335–336,May 1992. (in Japanese).

[213] T. Tannai and M. Hagiwara, “Colorist: A color designingsupport system to reflect Kansei,” in 2nd Annual Conf. ofJSKE2000, p. 200, Sept. 2000. (in Japanese).

[214] T. Terano and Y. Ishino, “Data analysis using simulated breed-ing and inductive learningmethods,” in IJCAI’95 Workshop onData Engineering and Inductive Learning, (Montreal, Canada),pp. 60–96, Aug. 1995.

[215] T. Terano, Y. Ishino, and K. Yoshinaga, “Integrating machinelearning and simulated breeding techniques to analyze the char-acteristics of consumer goods,” in Evolutionary Algorithms inManagement Applications (J. Biethahn and V. Nissen, eds.),pp. 211–224, Springer-Verlag, 1995.

[216] T. Terano, Y. Ishira, and K. Yoshinaga, “Integrating machinelearning and simulated breeding techniques to analyze the char-acteristics of consumer goods,” in Evolutionary Algorithms inManagement Applications (J. B. and. V. Nissen, ed.), pp. 211–224, Springer-Verlag, 1995.

[217] T. Terano and Y. Ishino, “Marketing data analysis using in-ductive learning and genetic algorithms with interactive- andautomated-phases,” in IEEE Int. Conf. on Evolutionary Com-putation (ICEC’95), (Perth, WA, Australia), pp. 771–776,Nov./Dec. 1995. vol.2.

[218] T. Terano and Y. Ishino, “Knowledge acquisition from ques-tionnaire data using simulated breeding and inductive learningmethods,” Expert Systems With Applications, vol. 11, no. 4,pp. 507–518, 1996.

[219] T. Terano and Y. Ishino, “Interactiveknowledge discovery frommarketing questionnaire using simulated breeding and induc-tive learning methods,” in 2nd Int. Conf. on Knowledge Dis-covery and Data Mining. (KDD-96), (Portland, OR, USA),pp. 279–282, Aug. 1996.

[220] T. Terano, “Evolutionary computation with interactive phases– a tutorial –,” in Joint Symposium on System and Informa-tion, (Sendai, Japan), pp. 59–64, Nov. 1997. (in Japanese).

[221] T. Terano and Y. Ishino, “Interactive genetic algorithm basedfeature selection and its application to marketing data anal-ysis,” in Feature Extraction, Construction and Selection: AData Mining Perspective (H. Liu and H. Motoda, eds.),pp. 393–407, Kluwer Academic Publishers, Boston, 1998.

[222] T. Terano and M. Inada, “Knowledge discovery from clini-cal data using interactive evolutionary computation,” in 42ndKnowledge-based Systems of JSAI, vol. SIG-KBS, (Yokohama,Japan), Jan. 1999. (in Japanese).

[223] S. Todd and W. Latham, “Artificial Life or surreal art?,” in1st European Conf. on Artificial Life, pp. 504–513, MIT Press,1992.

[224] S. Todd and W. Latham, Evolutionary Art and Computers.Harcourt, Brace, Jovanovich: Academic Press, 1992.

[225] S. Todd and W. Latham, “The mutation and growth of art bycomputers,” in Evolutionary Design by Computers (P. Bentley,ed.), ch. 9, pp. 221–250, CA, USA: Morgan Kaufmann Publish-ers, 1999.

[226] Y. Todoroki and H. Takagi, “User interface of an interac-tive evolutionary computation for speech processing,” in SOFTKyushu Chapter Meeting, (Saga, Japan), pp. 13–16, Dec. 1999.(in Japanese).

[227] Y. Todoroki and H. Takagi, “User interface of an interactiveevolutionary computation for speech processing,” in 6th Int.Conf. on Soft Computing (IIZUKA2000), (Iizuka, Fukuoka,Japan), pp. 112–118, Oct. 2000.

[228] T. Unemi, “Genetic algorithms and computer graphics arts,”J. of Japan Society for Artificial Intelligence, vol. 9, no. 4,pp. 518–523, 1994. (in Japanese).

[229] T. Unemi, “A user interface for simulated breeding using multi-field,” in Workshop on Interactive Evolutionary Computation,(Fukuoka, Japan), pp. 42–46, Mar. 1998. (in Japanese).

[230] T. Unemi, “A design of multi-field user interface for simu-lated breeding,” in Asia Fuzzy System Symposium (AFSS’98),(Masan, Korea), pp. pp.489–494, June 1998.

[231] T. Unemi, “SBART2.4: Breeding 2D CG images and movies,and creating a type of collage,” in 3rd Int. Conf. on Knowledge-based Intelligent Information Engineering Systems (KES799),(Adelede, Australia), pp. pp.288–291, Aug./Sept. 1999.

[232] T. Unemi, “SBART 2.4: an IEC tool for creating 2D images,movies, and collage,” in Workshop on Genetic Algorithms in

Visual Art and Music, (Las Vegas, NV, USA), pp. pp.21–23,July 2000.

[233] J. Ventrella, “Explorations in the emergence of morphology andlocomotion behavior in animated characters,” in Artificial LifeIV.: 4th Int. Workshop on the Synthesis and Simulation of Liv-ing System, (Cambridge, MA, USA), pp. 436–441, MIT Press,July 1994.

[234] J. Ventrella, “Disney meets Darwin-the evolution of funny ani-mated figures,” in Computer Animation ’95, (Geneva, Switzer-land), pp. 35–43, IEEE Computer Society Press, Apr. 1995.

[235] G. Venturini, M. Slimane, F. Morin, and J. P. A. de Beauville,“On using interactive genetic algorithms for knowledge dis-covery in databases,” in 7th Int. Conf. on Genetic Algorithm(ICGA’97), pp. 696–703, CA, USA: Morgan Kaufmann Pub-lishers, July 1997.

[236] V. Vucic and H. H. Lund, “Self-evolvingarts – organisms versusfetishes,”Muhely – The Hungarian J. of Modern Art, vol. 104,pp. 69–79, 1997.

[237] M. Watanabe, S. Sakamoto, M. Ohsaki, and H. Takagi, “Devel-opment of an IEC fitting system for commercial hearing aids,”in Techmical Report of ASJ, H–2000–6, (Iwate, Japan), pp. 1–8, Jan. 2000. (in Japanese).

[238] T. Watanabe and H. Takagi, “Enhancement of distorted speechusing interactive GA,” in 11th Fuzzy System Symposium, (Ok-inawa, Japan), pp. 183–186, July 1995. (in Japanese).

[239] T. Watanabe and H. Takagi, “Recovering system of the dis-torted speech using interactive genetic algorithms,” in IEEEInt. Conf. on Systems, Man and Cybernetics (SMC’95), vol. 1,(Vancouver, Canada), pp. 684–689, Oct. 1995.

[240] M. Witbrock and S. Neil-Reilly, “Evolving genetic art,” inEvolutionary Design by Computers (P. Bentley, ed.), ch. 10,pp. 251–259, CA, USA: Morgan Kaufmann Publishers, 1999.

21