SOFT COMPUTING AND ITS APPLICATIONS. Aniati Murni Arymurthy *

Soft Computing and Its Application (Aniati Murni Arymurthy)

SOFT COMPUTING AND ITS APPLICATIONS Aniati Murni Arymurthy*

ABSTRAK Tema lokakarya ini adalah Komputasi Sains dan Teknologi Nuklir. Topik yang dipilih untuk makalah ini adalah Komputasi Lunak (Soft Computing) dan Aplikasinya. Komputasi sains di bidang ilmu komputer dan teknologi informasi terbagi menjadi analisis numerik, komputasi lunak atau komputasi cerdas, dan metode formal. Ada juga yang memasukkan metode formal dalam komputasi cerdas. Komputasi sains atau ilmiah di bidang sains dan teknologi diterjemahkan sebagai disain dan analisis algoritma (langkah-langkah penyelesaian) untuk penyelesaian masalah matematik di bidang sains dan teknologi. Pentingnya komputasi ilmiah adalah dimungkinkannya simulasi terhadap fenomena alam dimana prototipe disain dan penyelesaiannya dapat dilakukan secara virtual sebelum secara fisik rancangan sistemnya dibangun. Komputasi lunak (soft computing) berbeda dari komputasi keras yang konvensional (hard computing) dalam aspek mampu mengakomodasi ketidak-pastian (uncertainty), ketidak-tepatan (imprecision), kebenaran yang bersifat sebagian (partial truth), dan aproksimasi. Hal tersebut mirip dengan model berpikirnya manusia. Makalah ini membahas logika fuzzy, jaringan syaraf tiruan, support vector machine, dan algoritma genetika. Beberapa aplikasi potensial di bidang nuklir, antara lain mulai dari rancangan reaktor nuklir intinya, pengendalian sistem dinamis operasionalnya, monitoring normal tidaknya fungsi reaktor, sampai ke optimasi resiko dan biaya disain dan pemeliharaan.

ABSTRACT The theme of this seminar is on Scientific Computation and Nuclear Technology area. The selected topic for this paper is on Soft Computing and Its Applications. Scientific computation in the area of computer science and information technology includes numerical analysis, soft computing or computational intelligence, and formal method. Sometimes computational intelligence covers formal method. The term of scientific computing covers the context of design and analysis of algorithms for numerically solving mathematical problems in science and engineering. The importance of scientific computing is to be able to simulate natural phenomena and to virtually prototypes engineering designs before they are actually built. Soft computing differs from conventional (hard) computing in the sense that, unlike hard computing, it is tolerant of imprecision, uncertainty, partial truth, and approximation. In effect, the role model for soft computing is the human mind. This paper discusses fuzzy logic, neural network, support vector machine, and genetic algorithm. Several potential applications in nuclear

*

Laboratory for Pattern Recognition and Image Processing -Faculty of Computer Science, University of Indonesia, UI Campus, Depok 16424, e-mail: [email protected]

23

Risalah Lokakarya Komputasi dalam Sains dan Teknologi Nuklir 2010, Oktober 2010 (23-44 )

technology include nuclear reactor core design, nuclear reactor dynamic system controller, nuclear power monitoring and fault detection; risk and cost optimization in reactor design and maintenance.

I.

BACKGROUNDS

The theme of this seminar is on Scientific Computation and Nuclear Technology area. The selected topic for this paper is on the scientific computation side that is specifically related to computer science and information technology with applications among others related to nuclear technology. The term of scientific computing covers the context of design and analysis of algorithms for numerically solving mathematical problems in science and engineering. Traditionally, it is called numerical analysis. It deals with continuous quantities and considers the effects of approximations. The importance of scientific computing is to be able to simulate natural phenomena and to virtually prototypes engineering designs before they are actually built [1]. In a department of computer science like the Faculty of Computer Science, University of Indonesia, scientific computation has ever become an area (or a field) of interest besides the other fields, such as Information Systems, Computer Architecture and Real-Time Systems, and Software Engineering. The scientific computing field includes numerical analysis, soft computing (computational intelligence) and formal method [2]. Other Department of Computer Science may include formal method in computational intelligence term. Soft computing differs from conventional (hard) computing in the sense that, unlike hard computing, it is tolerant of imprecision, uncertainty, partial truth, and approximation. In effect, the role model for soft computing is the human mind [3]. The soft computing includes Fuzzy Logic, Neural Networks, Support Vector Machines, evolutionary computation like Genetic Algorithm, Machine Learning and Pattern Recognition. This paper discusses the topics under Soft Computing that will include fuzzy logic, neural network, support vector machine, genetic algorithm, in the context of machine learning and pattern recognition. The discussion will only presents the important basic concepts to present the ideas, related applications, and literature reviews on their applications in nuclear technology. As a result, this paper is organized as follows. Fuzzy Logic (FL), Neural Network (NN), Support Vector Machine (SVM), and Genetic Algorithm (GA) are presented in Section II to Section V, consecutively. The GA topic can also be found in other paper in this seminar, among others is written by Hilda Deborah et al. [4]. 24


Furthermore, in the conventional management information and retrieval system, the information are usually retrieved based on metadata; while in a modern multimedia information retrieval system, the information is retrieved based on content [5]. The system is called Content Based Information (that could be in the form of text, image, audio, or multimedia) Retrieval System (CBIRS). The development of a CBIRS may utilize various soft computing methods. Several relevant applications will also be on CBIRS. Finally, this paper is closed by a summary and several potential research topics in Section VI.

II.

FUZZY LOGIC

II.1. Basic Concepts Standard set theory is difference from fuzzy logic set theory. Standard set theory says that the intersection of a set A and its complement (Ac) is a null set, while in fuzzy logic set theory it is not a null set. It can be explained by Fig. 1 [6].

(a)

(b)

(c) Figure 1. (a) In the standard set theory the air temperature is cool if it is between 50oF-70oF; otherwise it is not cool. (b) The temperatures belong to a fuzzy set only to some extent and belong to the set’s complement to some extent. (c) The 55oF has 50% degree of membership to ‘cool’ and 50% degree of membership to ‘not cool’. (Source: Kosko and Isaka [6])

25


In a standard set theory, a bivalent indicator function IA of a non-fuzzy subset A of X has the following value (Eq. 1). 1 if x is a member of subset A; IA(x)

= 0 if x is not a member of subset A.

(1)

The bivalent indicator function was extended to a multi-valued indicator or membership function as shown in Fig. 1(c).

II.2. Fuzzy Logic and Air Condition System Controller Suppose for an example, membership of the air temperatures can be grouped into three categories that can be illustrated as Fig. 2.

Figure 2. Groups of Air Temperatures Membership Fuzzy Function. Suppose an air conditioner system has a fuzzy controller with the following rules: • If the air temperature is cold (between 0oF-60oF), then the motor speed is slow = (0.6 * temperature + 4) rpm; • If the air temperature is cool (between 50oF-68oF), then the motor speed is medium = (0.5 * temperature + 10) rpm; • If the air temperature is hot ( => 55oF), then the motor speed is fast = 45 rpm. The question is, if the air temperature is 57oF, how fast is the motor speed? First of all, the graphics membership equation for each group of air temperature should be defined. The followings are an example of the graphics equation (Eq. 2) following the Sugeno-style fuzzy inference [7,8,9] for membership fuzzy function in Fig. 2. The equation for ’cold temperature’ group has a value of: 1 if temperature x <= 50; and 26


[(50 – x) / 10] + 1 if temperature 50 < x < 60. The equation for ’cool temperature’ group has a value of: (x – 50) / 10 if temperature 50 <= x <= 60; and [(60 – x) / 8] + 1 if temperature 60 < x <= 68. The equation for ’hot temperature’ group has a value of: (x – 55) / 15 if temperature 55 <= x <= 70; and 1 if temperature x => 70.

(2)

Based on the graphics membership equation for the air temperature, the following is how to determine the membership of the air temperature 57oF. The membership of 57oF to ’cold temperature’ group is [(50 – 57) / 10] + 1 = 0.3. The membership of 57oF to ’cool temperature’ group is (57 – 50) / 10 = 0.7. The membership of 57oF to ’hot temperature’ group is (57 – 55) / 15 = 0.13. By following the fuzzy rules of the air condition system controller, if the air temperature is 57oF, then the air condition motor speed becomes [(0.6 * 57 + 4) * 0.3 + (0.5 * 57 + 10) * 0.7 + 45 * 0.13] / (0.3 + 0.7 + 0.13) that will approach 39.2 rpm.

II.3. An Example of Fuzzy Logic in a Dynamic System Controller In a nuclear reactor core design an optimization process is required. The objective of the optimization process is to minimize the radial power peaking factor in a three-enrichment zone reactor, considering restrictions on the average thermal flux, critical and sub-moderated [10]. Afterwards, in the operational of the nuclear reactor as a dynamic system, a controller is needed. A Fuzzy Auto-Regressive Moving Average (FARMA) was proposed for a dynamic system controller [11]. The FARMA controller does not have any initial rules. It records the history of the plant input and output in a rule based system. After the input output rules are accumulated, the rule based system is updated in a self-organizing manner. A problem formulation can be represented by using a rule based of an expert knowledge as shown in Eq. 3 [11]. R1: if x1 is A1 and x2 is B1, the z is C1; R2: if x1 is A2 and x2 is B2, the z is C2; … … Rn: if x1 is An and x2 is Bn, the z is Cn. (3)

27


where x1 and x2 represent two process variables, and z represents the control variable. Ai, Bi, and Ci are linguistic values or fuzzy sets of the linguistic variables x1, x2, and z, respectively. The output of the system can be described as a mapping function of the plant input and output history. For a non-linear system, the mapping can be represented as follows (Eq. 4) y(k+1) = f (y(k), y(k-1), y(k-2),..., u(k), u(k-1), u(k-2),...)

(4)

where u(k) and y(k) is the input and output variables at the stage k. For a control purpose, the value of input u(k) is required to determine a reference output yref., the equation becomes Eq. 5. u(k) = g((yref, y(k), y(k-1),..., u(k-1), u(k-2),...)

(5)

Furthermore, a new rule with the input and output history can be defined as follows (Eq. 6). ith Rule: IF yref = A1i, y(k) = A2i, y(k-1) = A3i, ... , y(k-n+1) = An+1i, AND u(k-1) = B1i, u(k-2) = B2i, ... , u(k-m) = Bmi, (6) THEN u(k) = Ci. Eq. 6 is known as the FARMA rule [11], where m and n is the number of input and output variables, Aij and Bij is the antecedent values of the ith rule, and Ci is the consequent value of the ith rule.

III.

NEURAL NETWORK

III.1.

Basic Concepts

The fundamental element of a neural network is a neuron, which has multiple inputs and a single output (see Fig. 3). A signal xj at the input of synapse j is connected to the neuron k and multiplied by the synaptic weight wkj. An adder sums the weighted input signals, and an activation function for limiting the amplitude is applied to obtain the output yk of a neuron. The neuron also includes an externally applied bias, denoted by bk, which has the effect of increasing or lowering the network input of the activation function, depending on whether it is positive or negative, respectively [12]. 28


Figure 3. Neuron Model (Source: Bueno et al. [12]).

Figure 4. Self-Organizing Map (SOM) Network.

Figure 5. Back-Propagation (BP) Neural Network.

A neural network is a massively parallel distributed processor made up of simple processing units (called neurons) and there exist various architectures of neural network. Fig. 4 and Fig. 5 show the widely used neural networks called SOM (the unsupervised classifier) and BP (the supervised classifier).

III.2.

The Use of Neural Networks and Other Classifiers in Pap Smear Cell Classification

The steps of Pap smear cell processing can be illustrated in Fig. 6 [13]. The Pap smear test procedure will produce an image of the patient’s substance shown in Fig. 6(a). The first step is to relocate each cell and investigate each cell shown in Fig. 6(b) and Fig. 6(c). The single cell image in Fig. 6(b) is segmented and produces a segmented image in Fig. 6(c) that consists of nucleus, cytoplasm, and background areas. The object shape features are extracted from the image in Fig. 6(c) and the cell is classified into one of the seven categories of normal and abnormal cell in Fig. 6(d).

(a) Pap smear cells image (source: J. Indarti , FKUI).

(b) Single Pap smear cell image and (c) its segmented image supervised by a doctor (source: Jantzen et al. [14].

(d) Seven categories of cell condition (source: Jantzen et al. [14].

Figure 6. Pap Smear Cell Image Processing [13].

29


Figure 7. Diagnosing a Pap Smear Cell Image Based on Content-Based Image Retrieval System [13,14].

(a)

(b)

Figure 8 (a). Classification Accuracy using Several Classifiers [13,15]; (b) The Performance of the Content Based Image Retrieval System [13]. Six classifiers were used in this study, they include: Minimum Distance, BP using Gradient Descent algorithm, Least Square, Nearest Class Gravity Center, BP using Scaled Conjugate Gradient algorithm, and SVM ERBF. The results of the classification accuracy and the content-based image retrieval system (CBIRS) performance are shown in Fig. 8. Fig. 8(a) shows that the best performance is obtained by Support Vector Machine (SVM) Exponential Radial Basis Function 30


(ERBF) follows by the BP classifiers. The performance of the CBIRS in Fig. 7, is shown in Fig. 8(b). The system is good to recognize normal class and abnormal class, but for 7-class application the system average accuracy is only 40%.

III.3. Neural Network for Nuclear Power Monitoring and Fault Detection Bueno et al. [12] has proposed a system based on BP neural network for nuclear power monitoring and fault detection. Several variables were chosen as input data. They include coolant temperature above the reactor core, outlet core temperature, and primary loop flow rate. Different noise levels added to the input variables (0.4% noise was added to the coolant temperature and 1% noise was added to the primary loop flow rate). The output is the nuclear power. A theoretical model was developed to generate the database that will be used in their experiment. The system process equations include the mass and energy inventory balance, and the physical and operational aspects such as length, pipe diameter, flow rate, temperature and pressure drop. Data were generated for different operational conditions, both under normal and faulty conditions. The architecture of the multilayer perceptron network consists of 3 input neurons with linear activation function, 1 to 10 hidden-layer neurons with hyperbolic tangent activation function, and 2 output neurons (normal or fault condition). The database was generated into 21 steps of nuclear power variable (0% to 100% nuclear power in 5% steps) where 20 patterns were recorded for each condition, with a total of 420 patterns. Data in the database is divided into 3 sets: 50% for training (to establish the network by determining the weights and the bias), 25% for validation (to monitor the error during the training process), and 25% for testing (to compare different models). The network was trained using a variable number of neurons in the hidden layer (1 to 10 neurons). The learning process is limited by a maximum number of epoch (which is 1000) and a maximum tolerance error of 0.0001. The limitors are to prevent the energy error function to reach local minimums that will not give adequate solution. The output of the training and testing process provides the information on the number of epoch and recognition error for each arrangement of hidden-layer neurons (1 up to 10). Based on the network performance testing, the architecture of the network is obtained and can be furtherly validated using a real nuclear reactor data.

31


IV.

SUPPORT VECTOR MACHINE

IV.1. Basic Concepts Support vector machine (SVM) is a classification model that determines the best hyperplane to separate two-class input space, it is called a linearly separable binary classifier [16]. The classification model can be illustrated in Fig. 9. A kernel function can be used to map from a non-linearly separable input space to a linearly separable higher dimensional input space shown in Fig. 10. Radial Basis kernel Function (RBF) is a widely used kernel function. If the binary classifier is to be extended to a multiclass classifier, several methods such as One-vs-All, One-vs-One, and Directed Acyclic Graph can be used.

Figure 9. Margin, Hyperplane, and Support Vector [17].

IV.2.

Figure 10. Mapping from non-linearly to linearly separable input space [17].

Support Vector Machine in Multispectral and HyperSpectral Image Classification

Conventionally, multispectral image classification is done based on spectralbased and pixel-based classification approach. In the era of hyperspectral image classification, the use of pixel-based classification approach could not give reasonable good results, the use of spectral and spatial or an object-based classification approach is needed [18]. Fig. 11 shows typical processing diagrams of object-based and pixelbased classification approach [19]. 32


The image classification results using the pixel-based and object-based approach are shown in Fig. 12. The result of the pixel-based classification contains ‘salt and pepper’ noise and the classification accuracy of the result using object-based classification is better than the result using pixel-based classification approach.

(a)

(b) Figure 11. (a) Processing Diagram of Object-Based Classification Approach; and (b) Processing Diagram of Pixel-Based Classification Approach [19].

Figure 12. Classified Images of Indian Pine area [19]. (Source of Original Image: http://cobweb.ecn.purdue.edu/~biehl/MultiSpec/)

33


IV.3.

SVM for the Estimation of the Power Peaking Power in Nuclear Reactor

It is important to estimate accurately the local power density at the hottest part of a nuclear reactor core. This information is important for preventing the fuel rod from melting in a nuclear reactor [20]. The power peaking factor can be estimated by SVMs using numerous measured signals of the reactor coolant system. The SVM models were developed by using a training data set and validated by an independent test data set. The predicted values were then applied to a real nuclear power plant to validate the SVM model accuracy. When the validation process has produced a good performance then the SVM can be used for monitoring the low power density which is indirectly protecting the nuclear reactor core.

V.

GENETIC ALGORITHM

V.1. Basic concepts Genetic algorithm (GAs) is a search algorithm that tries to model natural evolution on biological entities. Some biological terminologies are still used in the implementation of GAs in digital computation, such as individual, population, generation, crossover, mutation, etc. [21]. An individual or a chromosome is represented in a string of bits. Different characteristics of individual will be represented in different strings of bits. GAs take a group of individual, i.e. a population as an input and then, this population will be forwarded into some processes that together build up a complete GAs. Those processes are crossover, mutation, and selection. Crossover is a process where a pair of chromosomes is chosen from the current population as parent chromosomes and exchanges the elements between them to create more chromosomes called offspring. In a mutation process, a small amount of individuals is selected from the current population to be mutated. Each of the selected individuals will get some of its element’s value changed to another value. At the end of iteration in GAs, the obtained solutions will be copied to the next iteration as an input. The basic concept of GAs is shown in Fig. 13, and the detail operation of each step can also be found in [4].

34


Figure 13. Basic Concept of Genetic Algorithm. Because GAs is a randomize process, a criteria to measure the goodness of obtained solutions is a key factor. This criterion is what we call as fitness function. Each chromosome will have a fitness value representing its goodness as a solution. Intuitively, the objective is to maximize the fitness value [21], and that is simply means, a chromosome with a higher fitness value will have a higher chance of being selected as an optimal solution to the problem.

V.2.

Application of Genetic Algorithm in Feature Selection

Object recognition or differentiation is done based on object descriptors or features. The question is how many features are to be used and what optimal combination of features will give a good result. This feature selection issues is related to the curse of dimensionality phenomena shown in Fig. 14. If the number of features increases then it can increase the recognition rate, but after an optimum number of features the recognition rate will decrease. In a high dimensional system, more number of training samples is required to build a good recognition system. 35


Figure 14. Curse of Dimensionality or Peaking Phenomenon [22]. False Color Composite Image Correlation Matrix Sequential Forward Floating Selection (SFFS); Channels: (158,156,153); Battacharya Distance = 5.93 Saluran spektrum

153

156

158

153

1

0.3253

0.3508

156

0.3253

1

0.3875

158

0.3508

0.3875

1

Genetic Algorithm (GA); Channels: (109,89,3); Battacharya Distance = 0.804

Figure 15. A Typical Results of Feature Selection Using 3 Spectral Channels [23]. The objective of a feature selection process is to obtain a set of features that have relatively high distance to each other and have low correlation coefficient. Yusnita [23] has compared feature selection results using genetic algorithm and other well known feature selection method called Sequential Forward Floating Selection (SFFS). AVIRIS data with 224 spectrum channels of Lunar Lake area [24] was used

36


for experiment. For a typical feature combination using only 3 spectrum channels, the result of the feature selection process is shown in Fig. 15.

V.3. Application of Genetic Algorithm in Nuclear Reactor Core Design There are at least two important applications of genetic algorithm in nuclear reactor core design. The first application is on finding an optimization solution in adjusting several reactor cell parameters, such as dimensions, enrichment and materials to minimize the radial power peaking factor in a three-enrichment zone reactor [25]. The second application is to optimize the reliability of the structures, systems, and components based on risk information [26]. Sacco et al. [25] minimize the fuel and cladding material costs as the objective function, and the radial power peaking factor as an operational constraint. The optimization algorithm is applied and the results of power-peaking and average thermal flux are found. The information is used as a feedback to the algorithm to determine the fitness function. There are several related papers written by this research group to be reviewed including in the use of fuzzy logic approach. This will be a good research and development topic. There is another concept called a risk-informed design that has been widely accepted during the design process of new nuclear plants [26]. The genetic algorithm is applied to optimize the number of structures, systems, and components which belongs to a risk-informed certain group. Another study has also stated that the regulation of nuclear power plant is evolving in a direction to embrace imposition of probabilistic safety criteria [27]. The use of genetic algorithm becomes important to satisfy an optimization process with multiple objective functions both in risk and cost. The multiple objective optimization process for both technical specification and maintenance requirements is expected to be able to obtain its required decision variables and satisfy its probabilistic safety criteria. The basic components that contribute to the risk of unavailability system include random failures, testing, corrective maintenance, and preventive maintenance. On the other side, the basic components that contribute to the cost include testing, preventive maintenance, corrective maintenance, plant shut down (unpredictable repair time), and overhaul maintenance. Based on the component of unavailability and cost parameters, a risk and cost functions modeling is formulated. In a nuclear power plant, two of the most problems are: loss of coolant accident and main steam line break [27]. Based on the selected two problems, decision variables and their upper and lower bound conditions are determined. The information becomes an input to the genetic algorithm.

37


VI.

SUMMARY AND POTENTIAL RESEARCH TOPICS

This paper presents several soft computing methods that include fuzzy logic, neural network, support vector machine, and genetic algorithm. The discussions include the basic concepts, the application both in nuclear applications and other applications. The suggested potential research topics in nuclear applications among others: • Nuclear reactor core design; • Nuclear reactor dynamic system controller; • Nuclear power monitoring and fault detection; and • Estimation of the power peaking power in a nuclear reactor. Their solutions may need to use soft computing methods. In that case, the research could be a collaboration of multi-disciplinary research groups.

REFERENCES 1.

HEATH, M.T., “Scientific Computing: An Introductory Survey”, Department of Computer Science, University of Illinois at Urbana-Champaign, 2002.

2.

Buku Pedoman Kurikulum dan Peraturan Akademik, 2003, Fakultas Ilmu Komputer, Universitas Indonesia.

3.

http://www2.cs.uh.edu/~ceick/6367/Soft-Computing.pdf [accessed October 5, 2010].

4.

H. DEBORAH, A. KOSASIH, and A. MURNI ARYMURTHY, “Genetic Algorithms in Several Applications”, Lokakarya Komputasi Dalam Sains dan Teknologi Nuklir, Serpong, October 14, 2010.

5.

S. DEB., “Multimedia Systems and Content-Based Image Retrieval”, Idea Group Publishing, 2004.

6.

KOSKO, B., and ISAKA, S., “Fuzzy Logic”, Scientific American, pp.76-81, July 1993.

38


7.

SUGENO-style fuzzy inference. http://www.tkk.fi/Units/Control/studies/kurssit/as-74-115/fuzzy2Sugeno.pdf [accessed October 7, 2010].

8.

“Fuzzy Inference Systems”. http://www.google.com/search?hl=en&source=hp&q=Sugeno%27s+fuzzy+infer ence+&btnG=Google+Search [accessed October 7, 2010].

9.

M.R. WIDYANTO, “Fuzzy Logic and Systems”, Bahan Kuliah, Fakultas Ilmu Komputer, Universitas Indonesia, 2009.

10.

SACCO, W.F., FILLHO, H.A. and PEREIRA, C.M.N.A., “Cost-based optimization of a nuclear reactor core design: a preliminary model,” in Proc. of 2007 International Nuclear Atlantic Conference, 2007.

11.

Y-M PARK, U-C MOON, and LEE, K.Y., “A Self-Organizing Fuzzy Logic Controller for Dynamic Systems Using a Fuzzy Auto-Regression Moving Average Model”. IEEE Trans. On Fuzzy System, Vol 3, No 1, pp. 75-82, February 1995.

12.

BUENO, E.I., TING, D.K.S., and GONCALVES, I.M.P., “Development of an Artificial Neural Network for Nuclear Power Monitoring and Fault Detection in the IEA-R1 Research Reactor at IPEN”, International Nuclear Atlantic Conference, INAC 2005, Santos, SP, Brazil, August 28 to September 2, 2005.

13.

R. AMALIA, P. ANGELINA, and A. MURNI ARYMURTHY, “Two Approaches for Diagnosing a Pap Smear Cell Image: Based on Image Segmentation and Classification Methods and Based on Content-Based Image Retrieval System”, Asia Pacific Conference on Art Science Engineering and Technology (ASPAC ASET), pp.S-180 – S-186, Solo, May 2008.

14.

JANTZEN, J., NORUP, J., DOUNIAS, G., and BJERREGAARD, G., “Papsmear Benchmark Data For Pattern Classification”, Technical University of Denmark, Denmark, 2005.

15.

D. ADDIATI, T. FARIDA, and A. MURNI, “Pap Smear Cell Image Diagnostic System Based on Fuzzy C-Means Clustering and BackPropagation Neural

39


Network”, Proceedings of Asialink International Conference on Biomedical Engineering and Technology, AICBET-2007, p.81-84, Jakarta, Indonesia, 2007. 16.

SOMAN, K.P., LOGANATHAN, R., VIJAYA, M.S., AJAY, V., and SHIVSUBRAMANI, K., “Fast Single-Shot Multiclass Proximal Support Vector Machine and Perceptrons”, Proceedings of the International Conference on Computing: Theory and Applications, p.294-298, March 05-07, 2007.

17.

SVM – Support Vector Machines. http://www.dtreg.com/svm.htm [accessed October 5, 2010].

18.

L. BRUZZONE, L., Carlin, L., and MELGANI, F., “A Multilevel Hierarchical Approach to Classification of High Spatial Resolution Images with Support Vector Machines,” Proceedings of Geoscience and Remote Sensing Symposium, 2004.

19.

W. PRIBADI and A. MURNI ARYMURTHY,” Perbandingan Analisis Citra Berbasis Piksel Dan Analisis Citra Berbasis Objek Pada Citra Remote Sensing Hiperspektral”, Seminar Ilmiah Ilmu Komputer Nasional (SILICON), pp.D17D23, Jakarta, 13-14 November 2008.

20.

BAE, I.H., M. G. NA, LEE, Y.J., and PARK, G.C., “Estimation of the Power Peaking Factor in a Nuclear Reactor Using Support Vector Machines and Uncertainty Analysis”, Nuclear Engineering And Technology, Vol.41 No.9, November 2009.

21.

GOLDBERG, D.E., “Genetic Algorithms in Search, Optimization & Machine Learning”, Reading, MA: Addison-Wesley, 1989.

22.

LANDGREBE, D., “On Information Extraction Principles for Hyperspectral Data, School of Electrical and Computer Engineering”, Purdue University, 1997.

23.

L. YUSNITA, “Metode Seleksi Ciri Sequential Forward Floating Selection dengan Algoritma Genetika dan Kombinasi Keduanya Untuk Citra Hiperspektra”, S1 Final Project, Faculty of Computer Science, University of Indonesia, 2004.

40


24.

California Institute of Technology, AVIRIS Free Standard Data Products, http://popo.jpl.nasa.gov/html/aviris.freedata.html, California Institute of Technology. [accessed August 13, 2003].

25.

SACCO, W.F., FILHO, H.A., and PEREIRA C.M.N.A., “Cost-based optimization of a nuclear reactor core design: a preliminary model,” in Proc. of 2007 International Nuclear Atlantic Conference, 2007.

26.

YANG, J-E., and LEE, W-J., “Risk-informed SSCS Classification at a Design Stage via a Reliability Allocation Approach”, Transactions, SMiRT, Toronto, August, 2007.

27.

MISHRA, A., and PANDEY, M.D., “Regulation of Nuclear Power Plants: A Multi Objective Approach”, Transactions, SMiRT 19, Toronto, August 2007.

DISKUSI

ANIK Apakah beda soft computing (Neural Network, unsupervised classifier, dan supervised classifier) dengan data mining.

ANIATI Neural network dapat berfungsi sebagai classifier. Unsupervised classifier bekerja tanpa bantuan pakar (contohnya Self-Organizing Map Neural Network). Supervised classifier bekerja dengan bantuan pakar yang dimanifestasikan dalam bentuk training sample set yang terdiri dari sampel contoh karakteristik setiap kategori objek yang digunakan pada proses pengenalan (contohnya Back-Propagation Neyral Network). Data mining merupakan suatu metode yang dapat digunakan untuk menggali karakteristik perilaku data atau karakteristik pengelompokan kategori data yang sulit dipastikan. Metode yang digunakan pada proses data mining adalah berbagai metode classifier unsupervised maupun supervised. Salah satu software data mining yang banyak digunakan adalah Weka. Weka memberikan pilihan penggunaan fasilitas,

41


K-Means clustering, Fuzzy K-Means clustering, Bayesian, Decision tree, Neural network, dll. TAUFIK 1. 2.

Apakah Proportional Integration Derivation (PID) controller juga termasuk smart controller? Apakah masing-masing metode soft computing (Fuzzy Logic, Neural Network) punya kelebihan yang khusus atau dari metode tersebut yang terbaik yang mana?

ANIATI 1.

2.

PID merupakan controller yang banyak dipakai di industri. Bila bentuknya sudah hardware biasanya akan lebih cepat eksekusinya daripada bentuk software. Setiap controller mempunyai model tertentu yang cocok untuk aplikasi tertentu. Tidak ada istilah the best controller karena masalah control yang diperlukan bisa mempunyai karakteristik berbeda. Kalau masalah mana yang lebih mempunyai kelebihan relatif sifatnya tergantung dari masalah yang akan diselesaikan dan metode yang digunakan, dan harus dilakukan studi perbandingan dengan kasus tertentu dan metode yang ingin dibandingkan.

MASKUR Uji biodistribusi obat radiofor apakah bisa disimulasikan menggunakan soft computing, kalau memang bisa kira-kira menggunakan software apa?

ANIATI Tergantung pada tahap mana. Bila masih dicobakan pada binatang dapat dicatat misalnya data dosis dan efeknya terhadap binatang. Bila pada tahap uji klinis dengan manusia perlu juga dicatat keadaaan sampel apakah mempunyai penyakit-penyakit tertentu. Jadi data terdiri dari dosis, keadaan pasien, dan impact obat terhadap pasien. 42


Data tersebut merupakan fitur objek yang akan digunakan untuk melakukan pengelompokan. Simulasi selalu dapat dilakukan dengan membangun basis data hipotetik menurut model karakteristik sistem. Selanjutnya dengan data hipotetik, simulasi dapat dilakukan menggunakan metode apapun, juga soft computing dimana predictive values dari model klasifikasi dapat ditentukan. Software yang dapat digunakan bisa MATLAB, Weka, dan open-source software yang dapat dicari di internet.

SULASNO 1. 2.

Mohon dijelaskan aplikasi atau tools untuk soft computing yang up to date. Apakah kaitan antara soft computing dan knowledge management system.

ANIATI 1. 2.

Tadi telah disebutkan sebagian dapat dilakukan dengan MATLAB, Weka, atau open-source software yang ada di internet. Content-based multimedia information retrieval system dapat bekerja sebagai knowledge management system. Tadi sudah diberi contoh bahwa data-data yang sudah didiagnosa dokter senior, medical recod yang menjelaskan treatment yang dilakukan terhadap pasien termasuk akibatnya dapat disimpan di suatu basis data. Seorang dokter junior yang sudah dapat melakukan diagnose selalu dapat memanfaatkan knowledge dari seniornya dengan melakukan proses query berdasarkan contoh query data kasusnya. Jadi knowledge dokter seniornya walaupun mereka sudah tiada, tetap dapat diacu dan dipelajari oleh generasi berikutnya.

DAFTAR RIWAYAT HIDUP

Nama

: Prof. Dr. Ir. Aniati Murni Arymurthy

Tempat & Tanggal Lahir

: Magelang, 29 Mei 1948

Pendidikan

: Doktor di Bidang Komputer 43

Daftar Isi


Riwayat Pekerjaan

: 1974 s.d. sekarang, Pengajar dan Peneliti Fakultas Ilmu Kompter - Universitas Indonesia

Keanggotaan

: IEEE

Makalah

: Soft Computing and Its Applications.

44

SOFT COMPUTING AND ITS APPLICATIONS. Aniati Murni Arymurthy *

Recommend Documents