Search

We propose a framework for automatic game parameter tuning using a game player model. Two kinds of computational intelligence techniques are used to create the framework: a fuzzy logic system (FS) as the decision maker and evolutionary computa- tion as the model parameter optimizer. Insights from a game developer are integrated into the player model consisting of FS rules. FS membership function parameters are optimized by a differential evolution (DE) algorithm to find optimal model parameters. We conducted experiments in which our player model plays a turn-based strategy video game. DE optimisation was able to evolve our player model such that it could compete well at various levels of game difficulty.

We introduce two strategies into the guided fireworks algorithm (GFWA) to further improve its performance by generating one or more weight-based guiding spark individual(s) for each rework individual. The first strategy assigns different weights to spark individuals under each rework individual according to their _fitness and then calculates one or more guiding vector(s) to guide the firework individual to evolve into potential directions. The second strategy decides the number of weight-based guiding spark individuals dynamically based on the evolution of a firework individual, i.e. if a firework individual does not evolve and survive in the next generation, then the second strategy reduces the number of spark individuals generated around the firework individual and generates the same reduced number of weight based guiding spark individuals additionally. We design a controlled experiment to evaluate the performance of our proposal using CEC2013 benchmark functions with five different dimensions. The experiment results confirm that the proposed strategies can provide effective guidance information to improve the GFWA performance significantly, and its acceleration effect for higher dimensional tasks is more obvious.

We propose a haptic database retrieval system based on user’s perception and media converter that translates it into a physical haptic feedback. We have developed a three-dimensional psychological space that expresses human perception haptic stimuli by using semantic differential pairs. Differential Evolution (DE) algorithm searches and retrieves haptic stimuli within a haptic physical feature space. Moreover, Neural Networks (NN) maps the searched physical feature vectors into the psychological space. The distance between the mapped positions and target position in the psychological space is used for the GA search, and when the distance becomes less than the threshold value, the retrieval search ends. In this study, we have conducted a preliminary experiment to construct psychological space and the NN part of the proposed system. The result of the preliminary experiment and the configuration of the NN will be discussed in detail.

本論文では，最適化問題の探索点生成メカニズムの競争進化に基づく最適化の考え方を示し，その考え方に基づいた数値最適化問題のためのアルゴリズムを実現し，ベンチマーク関数を用いて評価する．提案する最適化法における探索点生成メカニズムは，探索点生成領域とその領域内でのサンプリング法を持つ．従来の進化計算の探索点生成メカニズムは選択，交叉，突然変異などの演算であり生成メカニズムの特性自体は一定であった．このメカニズムを確率モデルに置き換える研究アプローチも確率モデルの構成法は固定であった．本論文での議論は，探索点生成メカニズムの構造や特性パラメータを動的に変化させることによって最適化能力を向上させようとする点に本質がある．, We show a concept of optimization based on competition and evlolution among mechanisms that generate search points, and realize an algorithm for numerical optimization problems based on the concept, and evaluate it with several benchmark functions. The mechanisms that generate search points in the proposal concept have (1) areas where search points are generated and (2) sampling methods within their areas. Conventional evolutionary computation (EC) generates new search points by using EC operators, and the characteristics of its mechanism that generates search points does not change. Conventional probabilistic approaches that replace the EC search mechanism with probabilistic models do not change the characteristics of its mechanism that generates search points, too. The essential point of our discussion in this paper is to dynamically change the characteristics of a mechanism that generates search points and aim to increase its optimization capability.

We apply interactive evolutionary computation technique to processing images in medical and other fields, analyze its performance of image enhancement, and discuss its capability and further improvement. We first show our system scheme and the design of user interface and then show experimental results of image enhancement. We also propose an idea of new interactive genetic programming that has not only arithmetical operators but also image processing operators in generated mathematical equations to increase its performance for image processing

We introduce new types of signal processing for which the characteristics of the signal processing filters are designed automatically by interactive evolutionary computation (IEC) based on human perception, such as hearing or vision. We first describe our existing works that use this approach, such as recovering distorted speech and hearing-aid fitting, as well as other related works in this field. Next, we evalu-ate the capabilities of visual-based image signal processing using IEC and compare it with conventional linear filters for the tasks of edge detection, high pass filtering, and horizon-tal / vertical component filtering. The experimental com-parisons show that the performances of both methods are similar, which means that the new approach, without a pri-ori knowledge on signal processing, is useful when signal processing users are not signal processing experts such as is the case in medical image processing or photo-retouch design.

新しい形の創作を支援する3次元モデラの構成法を提案する.対話型進化計算(IEC)を幾何形状の自動生成と変形に応用することで,初心者でも現実感のある3次元モデルを手軽に作りながらコンピュータグラフィックスの表現力を体験.習得できる.システムは,関数表現された3次元モデルを染色体に符号化し,遺伝的アルゴリズム(GA)による進化シミュレーションを用いて新しい3次元形状を自動生成する.これに対して利用者は,生成されたモデルに主観的な好みを表す適合度を与えて形の善しあしを評価する.この評価とGAによる進化シミュレーションを,利用者が満足する形が出来上がるまで繰り返す。写実性と創造性を問う2種類の造形課題に対して主観評価実験を行い,IECを3次元造形へ応用する効果と問題点について評価・検証した.

We chronicle the research on the fusion technology of neural networks and fuzzy systems (NN+FS), the models that have been proposed from this research, and the commercial products and industrial systems that have adopted these models. First, we review the NN+FS research activity during the early stages in Japan, the US, and Europe. Next, following the classification of NN+FS models, we show the ease of fusing these technologies based on the similarities of the data flow network structures and the non-linearity realization strategies of NNs and FSs. Then, we describe several models and applications of NN+FS. Finally, we introduce some important and recently developed NN+FS patents.

In this paper, we overview Interactive EC (evolutionary computation) research, showing the status quo and its remaining problems. The interactive EC technique optimizes systems from human interaction with computers. Recently, interest in this approach has increased in many application fields that we categorize into the artistic, engineering, and educational fields. We then overview the research within each field. Finally, we show several trials to address the problem of human fatigue.

This paper proposes to display individuals of interactive evolutionary computation(EC) in an evaluation order to reduce the fatigue of human operators. To display in an evaluation order, two prediction methods which do not depend on application tasks-the first using a neural networks(NN), and the second using Euclidean distance measure-are proposed. The prediction method based on an NN learns the relationship between past individuals and their evaluation values that an interactive EC operator gave in the past EC generations. The prediction method based on Euclidean distance measure predicts the human evaluation to a given individual using a weighting average method of the evaluation values to individuals in the past EC generations. These weights are calculated by the Euclidean distances between the given individual and individuals in the past. We evaluate the performances of these two prediction methods through simulation and subjective tests. The simulation test evaluates how the predicted order is closer to the actual one than similarity between the order that GA produces-conventional order-and the actual one. The comparison of cross-correlations between these two similarities has shown that the proposed method is significantly better than the conventional method. Following the simulation, the subjective tests are conducted to test how the proposed method let human operators feel better than the conventional method. The result has shown that there is no significant difference between the proposed method and the conventional method unlikely to the simulation test. We discuss the results of simulation and subjective tests. Furthermore, we discuss four hints for the future improvement obtained through subjective tests and the reports from subjects.

本論文では対話型GA技術を3次元CGのライティングデザイン支援に応用した.5人のアマチュアデザイナーに対し, 与えられたモチーフに最適なライティングパラメータの設定を行わせる実験を, 対話型GAを用いた場合と用いない場合, 更に対話型GAを用いないで設定したパラメータを初期値に, 対話型GAを用いた場合の3通りで行った.この実験タスクには人物像に対する3灯ライティングのデザインを設定した.ライティング 結果に対し, Sheffeの一対比較法および符号検定を評価者20名によって行い, 対話型GAによるデザイン支援の有効性を調べた.その結果, 対話型GAを用いない場合の評価が低いデザイナー, すなわち知識や経験の少ないデザイナー に対しては対話型GAによる支援効果が有意に現れた.一方, 対話型GAを用いない場合でも評価の高いデザイナーに対しては対話型GAによる支援効果は有意とは言えなかった.

This paper proposes a new input method for human operators of an interactive GA to reduce their psychological burden. This method utilizes discrete fitness values to reduce the psychological stress involved in the input procedure. The advantage and disadvantage of the proposed method are evaluated and discussed. We conducted simulations to investigate the influence of quantization noise stemming from the use of discrete fitness values on convergence. Results show that the quantization noise does not significantly worsen convergence in practical use. The proposed method and a variation of the method which combines both discrete and continuous fitness values are evaluated using two subjective tests involving the task of drawing faces. The results of the subjective tests indicate that our proposed method and its variation can significantly reduce the psychological stress level of human interactive GA operators. The experimental results of the proposed method are discussed on its universality.

コンピュータグラフィックス(CG)上の人物が自然に視線を移動させるための頭部と眼球の移動時間遷移モデルを提案する. はじめに, 実際に人間が視線を移動させるときの頭部の3次元移動特性を計測し, 過去の研究での計測データと照らし合わせて, 角速度をシグモイド状に変化しながら頭部を移動させることが共通的特徴であるとの知見を得る. この知見に基づいた視線移動に伴う頭部と眼球移動の数式モデルを提案する. このモデルのパラメータを変化させることで, 上目使いの動き, 不気味な動き, 驚いた動き, 複数人との対話のような動きのように, いろいろな状況に応じた動きが生成できる. 更に, 実用的観点からこのモデルの簡易的実現モデルも参考に示す. 最後にこのモデルがCG像の視線移動を自然に表現できるかどうかを主観評価実験を通じて評価し, 考察を加えて今後の研究の一方向についても述べる.

We introduce two techniques to approximate and analyze fitness landscape for evolutionary search enhancement. One involves dimensionality reduction method used for fitness landscape approximation to reduce the computational complexity of the fitting. The other uses Fourier transform to obtain the frequency information of fitness landscape for search acceleration and multi-modal optimization. We briefly describe the inspirations, principles and results of the two techniques.

This study describes a tinnitus test using interactive evolutionary computation (IEC) based on acoustic characteristics of onomatopoeia of subjects with tinnitus. Using acoustic characteristic of onomatopoeias helped to select test tones for subjects with tinnitus. Two methods of tinnitus pitch-match testing, i.e., a tinnitus test using IEC based on acoustic characteristics of onomatopoeia and a conventional clinical pitch-match testョwereused for three subjects with tinnitus for two days. Their subjective tinnitus sound did not change for two days. For each subject, the difference between two days for tinnitus testing using IEC was smaller than that for conventional testing. These results of tinnitus testing using IEC were consistent with the subjective reports of unchanged tinnitus pitch. This study suggests that tinnitus testing using IEC based on acoustic characteristics of onomatopoeia of subjects with tinnitus made it possible to measure the fine pitch of tinnitus

A method of tuning a fuzzy logic controller (FLC) by a genetic algorithm (GA) is proposed for lane following maneuvers in an automated highway system. The GA simultaneously detennines the shape of membership functions, number of rules, and consequent parameters of the FLC. The GA approach operates on binary representations of FLCs and uses an expression for a fitness score to be maximized, which takes into account the tracking error, yaw rate error, lateral acceleration error, rate of lateral acceleration, front wheel steering angle, and rate of front wheel steering angle, to find an optimal controller. Apriori knowledge about both the physical application and FLCs is incorporated into the design method to increase the performance of the design method and the resulting controller. The controllers designed by this method are compared in simulation to a conventional PID controller, a frequency shaped linear quadratic controller (FSLQ), and previously designed FLCs tuned manually

This paper overviews four combinations of fuzzy logic, neural networks and genetic algorithms: (J) neural networks to auto-design fuzzy systems, (2) employing fuzzy rule structure to construct structured neural networks, (3) genetic algorithms to auto-design fuzzy systems, and (4) a fuzzy knowledge-based system to control genetic parameter dynamically

This paper proposes using fuzzy logic techniques to dynamically control parameter settings of genetic algorithms (GAs). We describe the Dynamic Parametric GA: a GA that uses a fuzzy knowledge-based system to control GA parameters. We then introduce a technique for automatically designing and tuning the fuzzy knowledgebase system using GAs. Results from initial experiments show a performance improvement over a simple static GA. One Dynamic Parametric GA system designed by our automatic method demonstrated improvement on an application not included in the design phase, which may indicate the general applicability of the Dynamic Parametric GA to a wide range of applications.

This paper introduces (1) how neural networks and genetic algorithms have been used for autodesigning fuzzy systems, (2) how neural networks are combined with fuzzy systems in commercial applications, and (3) how fuzzy systems are used to improve the performance of neural networks and genetic algorithms

This paper proposes that the best reasoning (i.e. rule evaluation) method which should be used in a fuzzy system significantly depends on the reasoning environment. It is shown that allowing for dynamic switching of reasoning methods leads to better performance, even when only two different reasoning methods are considered. This paper discusses DSFS (Dynamic Switching Fuzzy System) which dynamically switches and finds the best reasoning method (from among 80 different possible reasoning methods) to use depending on the reasoning situation. To overcome the reasoning speed and memory problem of DSFS due to its computational requirements, the DSFS Switching Reasoning Table method is proposed and its higher performance as compared to a conventional fuzzy system is shown. Finally, efforts to obtain general relationships between the characteristics of different reasoning methods and the actual control surface/environment is discussed.

This paper (1) introduces a.ri automatic fuzzy system design method that uses a Genetic Algorithm to automatically determine the shape of membership functions, consequent parts, and the number of rules, (2) proposes techniques for incorporating apriori knowledge to improve performance of both the design method and the resulting systems, and (3) evaluates how effective embedding apriori knowledge accomplishes these goals. The method has been applied to and evaluated on the cart-pole control task and shows that embedded apriori knowledge does indeed lead to improvements in both the design method performance and final system performance

In this paper, we describe some applications of fuzzy logic to the design of image processing equipment, and how fuzzy technology is used in this area. Most of this equipment is available on the market. The equipment described here includes mechanisms for autofocus, auto-exposure, auto-zooming, and automatic white balancing for cameras and video camcorders; image stabilization for camcorders; electrophotographic controls in copying machines; television image quality control; image codec systems and stepper alignment for semiconductor device processes

International audience; We propose Interactive Differential Evolution (IDE) based on paired comparisons for reducing user fatigue and evaluate its convergence speed in comparison with Interactive Genetic Algorithms (IGA) and tournament IGA. User interface and convergence performance are two big keys for reducing Interactive Evolutionary Computation (IEC) user fatigue. Unlike IGA and conventional IDE, users of the proposed IDE and tournament IGA do not need to compare whole individuals each other but compare pairs of individuals, which largely decreases user fatigue. In this paper, we design a pseudo-IEC user and evaluate another factor, IEC convergence performance, using IEC simulators and show that our proposed IDE converges significantly faster than IGA and tournament IGA, i.e. our proposed one is superior to others from both user interface and convergence performance points of view.

We developed a system to investigate acoustic characteristics of tinnitus tones using genetic algorithms of interactive evolutionary computation (IEC). The utility of tinnitus measurement using the IEC system was evaluated for normal hearing subjects with a pseudo-tinnitus tone. A pure tone and a band noise were employed as pseudo-tinnitus tones. For the pseudo-tinnitus pure tone, the highest scored tone in the fifth IEC generation was very close to the frequency of the pseudo-tinnitus pure tone. However, the highest scored tone showed a poor convergence for sensation level. For the pseudo-tinnitus band noise, the highest scored tones in the seventh IEC generation showed almost similar acoustic characteristics in center frequency, bandwidth and sound pressure level to the pseudo-tinnitus band noise.

ニューラルネット(NN)学習に非線形最適化手法を組み込み，更に四つの改良を加えてその高速化への寄与を評価した．はじめにbackpropagation学習法の課題を指摘し，この対策として非線形最適化手法の有効性を述べる．次に各非線形最適化手法をNN学習の観点から考察し，Davies, Swann, Campeyの直線探索法とFletcher-Reeves法による共役こう配法を選択し，これら二つの基本手法を組み込んだNN学習アルゴリズムを定式化する．更にこの二つの非線形最適化手法をもとに，NN学習に適した四つの改良手法を提案する．提案する改良手法は，(1)直線探索時のforward計葬をメモリの増加と引換えに短縮する方法，(2)直線探索初期の局所最小点へのトラップを避ける方法，(3)並列処理に向いた直線探索方法，(4)こう配方向の切換えによる共役傾斜方法，の4手法である．排他的論理和問題による評価実験を通じて以上の手法の評価を行う．基本的非線形最適化手法はもちろんのこと，本論文で提案する四つの改良手法がいずれも高速化に有用であることを示す．

We first describe the background and technical aspects of an interactive EC (evolutionary computation). Then, we overview the research on its application in three major fields| artistic, engineering, and educational| and on human interface. We also list approximately 90 references on interactive EC for further study.

We discuss new type of signal processing approach, auditory and visual-based signal processing, and show some examples of filter design based on this approach. This approach allows us to design filters based on our auditory or visual evaluation of the filter outputs, while the conventional signal processing is based on mathematical expression. Interactive evolutionary computation is a main technology of this approach, and the evolutionary computation optimizes the target filter based on our auditory or visual inspection to the processed signal. Recovering speech from distorted speech, compensation of hearing impairment, and enhancing images are discussed as examples of this new approach of signal processing.

We introduce computational intelligence technologies, especially, neural networks, fuzzy systems, and evolutionary computation, for researchers in geo-science to help them to apply these technologies to their fields. Following the general introduction of each technology, we explain how these technologies handle nonlinearity, how to deign these nonlinear systems, how to acquire knowledge from observed data, what kind of feature they have, and how to embed expert's knowledge in the systems. We demonstrate three examples of computational intelligence techniques applied to geological inversion, estimation of rock location and density from their gravity responses, and designing image enhancement filters.

We propose a new approach to easily creating 3D geometric models. A technique called interactive evolutionary computation (IEC) is introduced to accelerate user’s invention and inspiration of new shapes. The proposed IEC-based design system generates aesthetically pleasing shapes through the simulation of natural evolutionary processes. The user is responsible for specifying his/her subjective preference to each shape generated by the IEC. The system allows even beginners with little knowledge and experiences of the 3D modeling to acquire innovative shapes. It also provides skilled experts with an advanced geometric modeling interface based on the implicit surface method. The experts can directly modify the internal parameters of the 3D models to make them more elaborate ones. The result of a preliminary experiment is presented to show the potencial of the proposed modeling method.

PSO vs. IGA does not show this tendency. We analyzed them and found that it was due to the too much sensitivity of interactive PSO to quantization noise in fitness. We propose its solutions and show their better performance for less complex tasks than conventional IGA as expected from PSO vs. GA. Tournament IEC requests an IEC user to compare pair individuals, which reduces IEC user fatigue, while conventional IEC requests him/her to compare all individuals and evaluate them to make offspring. Simulated breeding requests his/her to compare all individuals and choose better parents. Although choosing and evaluating look similar, one click for the former and thinking evaluation level relatively and inputting the evaluation level for the latter are quite different from the user fatigue point of view. Drawback of both tournament IEC and simulated breeding are slow conversion due to lack of evaluation information. Paired comparison-based interactive DE has both features of pair comparisons like a tournament IEC and full evaluation like a conventional IEC. Thanks to them, it causes less user fatigue and converges faster. At this moment, this method looks the best IEC framework.

We introduce our recent topics of interactive evolutionary computation (IEC) research. First, we introduce new types of IEC research unlike normal IEC-based system optimization: (a) estimation of IEC user's psychological scale or mental situation by analyzing the outputs of the system optimized by the user, (b) Extended IEC that optimizes a target system using IEC user's physiological data, and (c) IEC combined with evolutionary multi-objective optimization. Secondly, we describe how to realize an IEC simulator for repeated experiments under the same experimental conditions. Thirdly, we introduce research on reducing IEC user fatigue using a model of IEC user's evaluation characteristics.