CHAPTER 5 RADIAL BASIS FUNCTION (RBF) NEURAL NETWORKS FOR TOOL WEAR MONITORING

Size: px
Start display at page:

Download "CHAPTER 5 RADIAL BASIS FUNCTION (RBF) NEURAL NETWORKS FOR TOOL WEAR MONITORING"

Transcription

1 CHAPTER 5 RADIAL BASIS FUNCTION (RBF) NEURAL NETWORKS FOR TOOL WEAR MONITORING This chapter presents an overview of radial basis function neural networks and their applications to tool wear monitoring. The center of RBF units have been fixed using three different approaches and their learning characteristics are analysed. The performance of RBF neural networks have been compared with MLP for tool wear monitoring. 5. Radial Basis Function (RBF) neural networks There are different forms of designing a supervised neural network. The backpropagation algorithm for the design of a multi-layer perceptron (under supervision) can be viewed as an application of an optimization method known in statistics as stochastic approximation. Another approach can be viewing the design of a neural network as a curve-fitting problem in a high dimensional space. This involves finding a surface in a multi dimensional space that provides a best fit to the training data, with the criterion for "best fit" being measured in some statistical sense. Correspondingly generalization is equivalent to the use of this multi dimensional surface to interpolate the test data. This is indeed the motivation behind the method of radial-basis fiinctions. In the context of a neural network, the hidden units provide a set of 'functions' that constitute an arbitrary "basis" for the input patterns (vectors) when they are expanded into the hidden unit space. Therefore these functions are called 'radial-basis functions'. Broomhead and Lowe (988) were the first to exploit the use of radial-basis functions in the design of neural networks. Other major contributions to the theory, design and application of RBFs include papers by Moody and Darken (989), Renals (989) and Poggio and Girosi (990) [76]. The main advantages claimed for the RBF model are its simplicity and the ease of implementation. The learning and generalization abilities of these networks are excellent. The construction of a radial basis function network in its most basic form involves three entirely different layers. The input layer is made up of source nodes (sensory imits). The 94

2 second layer is a hidden layer of high enough dimension, which serves a different purpose from that in a multi-layer perceptron. The output layer supplies the response of the network to the activation patterns applied to the input layer. The transformation from the input space to the hidden-unit space is non-linear, whereas the transformation from the hidden-unit space to the output space is linear. Fig. 5. shows the typical RBF architecture. Input layer Hidden layer Output layer Cn Fig. 5. RBF architecture The RBF network is a single hidden-layer feed forward neural network. Each node of the hidden layer has two parameters, a center Xj and a width aj. This center is used to compare with the network input vector to produce a radially symmetrical response. The width controls the smoothness properties of the interpolating function. Response of the hidden layer are scaled by the connection weights of the output layer and then combined to produce the network output. RBFs have been shown to have universal approximation capability. In the classical approach to RBF network implementation, the basis functions are usually chosen as Gaussian and the number of hidden units is fixed a priori based on some properties of the input data. The weights connecting the hidden and output units are estimated by linear 95

3 least squares methods, e.g., least mean square (LMS). The disadvantage with the classical approach is that it is not suitable for sequential learning and it also results usually in too many hidden units. The RBF network requires less computation time for the learning and produces a more compact topology than other neural networks [76]. 5.2 Learning Strategies The learning process undertaken by a radial-basis function (RBF) network may be visualized as follows. The linear weights associated with the output unit(s) of the network tend to evolve on a different 'time scale' compared to the non linear activation functions of the hidden units. Thus as the hidden layer's activation functions evolve slowly in accordance with some non linear optimization strategy, the output layer's weights adjust themselves rapidly through a linear optimization strategy. Since different layers of an RBF network perform different tasks, it is reasonable to separate the optimization of the hidden and output layers of the network by using different techniques and perhaps operating on different time scales (Lowe 99). There are different learning strategies available for the design of an RBF network, depending on how centers of the radial basis functions of the network are specified. Three different learning strategies are discussed below Fixed Centers Selected at Random The simplest approach is to assume fixed radial-basis functions defining the activation functions of the hidden units. Specifically, the locations of the centers may be chosen randomly from the training data set. The RBFs use gaussian activation function which is defined as (j)j (x) = exp(- xj-^i /2a j) where Xj is the center andctjis the width (standard deviation), j=l,2, c, where c is the number of centers. The only parameters that would need to be learned in this approach are the linear weights in the output layer of the network. The weights ate learned using a simple LMS algorithm or the gradient descent approach. 96

4 5.2.2 Self-Organized selection of centers- In this approach, the radial-basis functions are permitted to move the locations of their centers in a self-organized fashion, the weights of the output layer are computed using a supervised learning rule. The network undergoes a hybrid learning process (Moody & Darken, 989; Lippmann,989). The self-organized component of the learning process serves to allocate network resources in a meaningful way by placing the centers of the radial-basis functions in only those regions of the input space where significant data are present. The self-organized selection of the centers is done using clustering algorithms like the batch fuzzy c-means. The next step is the determination of the width parameter of the basis functions Cj which must be chosen using some other procedure. An effective scheme to find the widths is the P-nearest neighbor heuristic (Moody & Darken, 989). Consider a given center vector Xj (j=l,...c) and assume Xji, Xj2,... Xjp (l<=jl, j2,...jp<=c) are the P- nearest neighboring centers. The width of the basis function Oj is given by the RMS distance of the given cluster center Xj to the P nearest neighboring centers: aj = sqrt(l/p2:vill'^j-xjpll')[64]. Another heuristic is to choose all the QJ to be equal. This ensures that the basis functions overlap to some degree and hence give a relatively smooth representation of the distribution of training data [66] Supervised selection of centers- In this approach the centers of the radial-basis functions and all other free parameters of the network undergo a supervised learning process. A gradient-descent procedure is used to accomplish the task. The instantaneous value of the cost function is defined as E = '/2E"i=iE^ where n is the number of training examples used to undertake the learning process and Ej is the error signal defined by Ei = Ok - S\=i Wk exp(- xj-^i p/2a^j) where K is the number of nodes in the output layer, c is the number of centers in the hidden layer 97

5 (j=l,2,...c). The requirement is to find the free parameters Wk, Xj and GJ in order to minimize E [76]. 5.3 Comparison of RBF networks and Multi-layer Perceptrons Radial basis fiinction networks and multi-layer perceptrons play very similar roles in that they both provide techniques for approximating arbitrary non-linear functional mappings between multi-dimensional spaces. Both are examples of non-linear layered feed forward networks and they are universal approximators. The comparison between the two networks is given in the table 5.. Table 5.: Comparison ofmlp and RBF Networks SI. No ML? Has more hidden layers Use monotonic Sigmoidal function Hidden and output layer share a common neuron model Compute iimer products Uses Global error for minimization RBF Has a single hidden layer Use non monotonic Gaussian function Hidden and output layer share different neuron models Compute Euclidean norm Uses local error for minimization 5.4 Training & Testing the RBF network In order to train the RBF neural network, three different methods have been selected for center initialization of the RBF units. We have considered En-8 data set for this and the results obtained have been presented. For the other two data sets namely grey cast iron and En-2A the results obtained have been tabulated Centers of the RBF units initialized randomly (a) Training phase - The input patterns have a dimension of 2 and output is of dimension. The complete training set consists of 69 patterns. Sample training and test patterns is given in Table

6 Table 5.2: Sample Input Patterns (En-B) -^h \ I'li No. -> J> 4 5 Tcsl KDC S S Kisc I'iiiR' X Kins vollai;c 0.977X !.0000 Kiicrjjy Kvfiil duralion , , Mean KiM' Tinii; R K, <\ l'at(o- K/ X<) R OTfWfit) " , Cutting. - Or3'J77 V) M).0 OJiJST ', > :. / ' Fcyd The RBF units for training the network are selected arbitrarily and each unit is assigned an input pattern randomly, thus initializing the centers. The algorithm to train the network is given below. Algorithm:. Select the number of RBF units arbitrarily. 2. Initialize their centers from input data randomly. 3. SetEtot=0. 4. Choose the input-output pair {^\, (^\^, where i = l,2,...n is the number of patterns and i =, 2,... p is the number of input features and k = is the output feature. 5. Compute the hidden layer output, Vj = e" V ^ ^ \ where Xj is the center and GJ is the width of the RBF unit. 6. Compute the output using Ok = / (l+e'^'^kj "^j) 7. Compute the square error E =(0k - Ck) * ( Ok - C,v) and Etot = Etot + E. 8. The change in output layer weights are calculated as follows: 5k=(0k-Ck)* Ok*(l-Ok) Awkj= 5k*Vj*Ti*a w"'=«' kj = w""* kj + Awkj 9. If Etot > Emin, then goto Step Save weights, centers and widths and exit. 99

7 The simulation parameters r =0.85 ana a=u.ud are mamtamed constant for all the studies. The training of the network has been done with different number of RBF units. The widths of the RBF units are determined using P-nearest neighbor heuristic (each RBF unit has different width value) and have been kept constant for all the RBF units and studies have been carried out. The upper limit on the learning cycle has been kept at,75,000 epochs to observe the network convergence behavior. For any parameter setting, if the network takes more number of epochs than the set value, it is considered as non convergent and the network parameters are set to new values and training is restarted. Table 5.3 shows the error reached during the training of RBF network with different RBF units. Table 5.3: Variation of Error with Different Number of RBF Units No. of RBF units Error It is evident that the error decreases with increase in the number of RBF units and it is minimum for 66, beyond which the error increases. Similarly optimum number of RBF units has been determined for grey cast iron and En-2A data set. Table 5.4 shows the optimum number of RBF units for three data sets. Table 5.4: Optimum Number of RBF Units (random selection of centers) Data set En-8 steel Grey cast iron En-2A steel No. of Centers Error

8 Fig. 5.2 shows the variation of error with number of epochs for 66 centers E^och Fig. 5.2 Variation of Eiror witli Epochs (2-66-, En-8, Random selection of centers) It is clear that the error reduces gradually with epochs. Fig. 5.3 shows the final network architecture Input layer Hidden layer Hutput layer w Jk ^n Fig. 5.3 RBF: Final networlc architecture (2-66-) Centers selected randomly These results presented are for varying widths, which have been determined using the P-nearest neighbor heuristic. In another study widths have been kept constant for all the RBF units and the network has been trained for different values of widths. Table 5.5 0

9 shows the error reached during training the RBF network with different values of widths for 66 RBF units. Table 5.5: Variation of Error with Different width Values for 66 RBF Units Width value Error It is clear from the table that the error increased with the increase in the value of the width. The optimum value of width has been chosen as 0.9, based on the performance of the network on both training and test data. The network has also been trained with width = 0.9 for different number of RBF units. It is seen that with the increase in the number of RBF units, the error decreases and the error is minimum when the number of centers is equal to the number of data in the training set. (b) Testing phase: In the testing phase, 5 data samples have been selected. The trained RBF network with different number of RBF units has been tested with these data samples. The network output specifies the condition of the tool in terms offlankwear on the tool. For the sake of analysis of the network output, from the tool wear monitoring point of view, only two states have been considered. Based on the flank wear values, the network output can be interpreted as follows: (i) if (network output <=0.4) the tool is in the 'Normal' condition. (ii) if (network output > 0.4) the tool is in the 'abnormal' condition and hence replace the tool. Table 5.6 shows the results of testing seen and unseen data for different number of RBF units. 02

10 Table 5.6 Performance of REF Network with Random Selection of Centers No. of. RBF units Training accuracy Testing accuracy 94% 80% 96% 97% 97% It is clear from the table that as the number of RBF units increase, the performance of the network improves. Table 5.7 shows the sample testing results for the RBF network with 66 hidden units. Table 5.7: Sample Testing Results (2-66-, En-8, random selection of centers) Pattern No. Desired output (mm) T Network output (mm) raining data Test data Classification error X0-^ O.OOUll XIO-^ XIO"^ X0-^ X0-" Table 5.8 shows the performance of the network for different values of widths for 66 RBF units. 03

11 Table 5.8: Performance of RBF Network, for Different aj (En-8) Width value Oj Training accuracy Testing accuracy 00% 87% 99% 97% 90% 80% It is evident from the table that the overall network performance on training and test data improves with a increase in the width value and is maximum for 0.9, beyond which it reduces. Table 5.9 shows the sample testing results for training and test data for width= Table 5.9: Sample testing results (2-66-, width: 0.9, En-8) Pattern No. Desired output (mm) Ti Network output (mm) 'aining data Test data Classification error XIO' X0-* X0-* The results for all the three data sets are presented in table 5.0 below. Table 5.0: Performance of RBF network for three data sets (Random Selection) Data set En-8 steel Grey cast iron En-2A steel No. of RBF units Error Training accuracy 97% 96% 77% Testing accuracy 93 %! 95 % i 7% 04

12 All the data sets except En-2A data set exhibit good generalization. The poor performance of RBF network on En-2A data set could be due to insufficient number of patterns representing all possible tool wear states. In the second approach, Batch fuzzy- c means algorithm has been used to determine the RBF units Center initialization using Batch fuzzy c-means algorithm In this study, batch flizzy-c means algorithm has been used to initialize centers of the RBF network. The number of RBF units has been fixed arbitrarily. This is a clustering technique, which assigns feature vectors Xj into c clusters, which are represented by prototypes Vj. The certainty of the assignment of the feature vector Xj into various clusters is measured by the membership functions Uj(xi) = uy e [0,], I<=j<=c, which satisfy the property ZVi uy =. The M x c matrix U = [uy] e u is a fuzzy partition in the set u defined as u = { U e R"^' Uy e [0,], V i, j; IVi uy =, Vi; 0< Z^i=i uy < M, Vj}, where x e R" are M feature vectors. This has been used to iteratively select the optimum number of centers for the RBF network. The algorithm is presented below [36]. Algorithm:. Select c, no.of centers, and set m=2, flizziness parameter, Emin and iter = Generate an initial set of prototypes {vi,o, V2,o, Vc,o}. 3. iter = iter + Uij, iter = [ IVl ( Xi-Vj,i,er-l '/ Xi-V,,iter-l P)"^"-'^]-', l<=i<=m ; l<=j<=c, Vj,iter= l''i=l(uij.iterrxi/z''i=l(uij,iterr, l<=j<=c, Eiter = ZVl II ^j,'ter " Vj,iter- 4. If iter < N (maximum number of itera;tions) and Euer > Emin, then goto step 3. The centers determined have been used to train the RBF network. In batch algorithms all the prototypes are updated together. 05

13 (a) Training phase: The training data patterns are described by 2 input features and output feature. Since the cutting conditions have similar values for various patterns, they do not contribute much to the center selection using the batch fuzzy c-means algorithm and have not been included for grey cast iron and En-2A data set. Therefore the number of input features has been reduced to 0 for both these data sets. The simulation parameters T]=0.85 and a=0.05 have been maintained constant for all the experiments. The training of the Batch fuzzy c-means algorithm has been carried out with different number of centers. And the final optimal centers obtained after convergence has been used to train RBF network. For training batch fuzzy c-means algorithm the target error has been fixed at The RBF network has been able to recognize most of the training patterns in epochs. Table 5. shows the results of applying batch fuzzy c-means algorithm on three data sets. Table 5.: Results of applying batch fuzzy c-means algorithm for three data sets Data set En-8 steel (2 input features) Grey cast iron (0 input features) En-2A steel (0 input features) Number of centers No. of Epochs Table 5.2 shows the variation of error with number of RBF units. Table 5.2: Variation of Error with number of RBF units No. of Centers or RBF units Error It is clear that the error decreases with an increase in the number of RBF units and it is minimum for 50 RBF units, beyond which the error increases. Fig. 5.4 shows the variation of error with number of epochs for 50 RBF units. 06

14 V 0.4 ^v,.,,^^^ i 0.3 ^ ^ n U ^ T ' Epoch Fig. 5.4 Variation of Error with Epochs (2-50-, En-8, Batch Fuzzy C Means) The drop in the error with epochs is gradual, when compared to that exhibited by RBF network trained using randomly initialized centers. Fig. 5.5 shows the final RBF architecture. Hidden layer Output layer ^n Fig. 5.5 RBF: Final Network architecture (2-50-) Centers Initialized Using Batch Fuzzy c-means Algorithm (b) Testing Phase: The trained network with different number of RBF Units has been tested with patterns from the training data set as well as patterns, which are not used for training. 07

15 Table 5.3 shows the results of testing both training and test data for different number of centers used in the network. Table 5.3: Performance of RBF network with selection of centers through 5 3 Batch fuzzy-c means algorithm (Input features: 2, En-8) No. RBF units Training accuracy Testing accuracy 20 96% 87% 30 94% 50 97% % It is clear from the table that as the number of RBF units increases, the performance of the network on training and test data improves till 50, beyond which there is no change in the performance. Table 5.4 shows the sample testing results for seen and unseen data for 50 RBF units. Table 5.4: Sample testing results (2-50-, En-8, batch fuzzy c-means) Pattern No. Desired output (mm) Tr Network output (mm) aining data Test data Classiflcation error XIO"'' X0-' XIO"' XIO-^ Table 5.5 shows the results for all the three data sets. 08

16 Table 5.5: Performance of RBF network using batch fuzzy c-means algorithm for center initialization for three data sets Data set En-8 steel (2 input features) Grey cast iron (0 input features) En-2A steel (0 input features) No. of RBF units Error Training accuracy 97% 85% 80% Testing accuracy 90% % The table shows that using 0 features in the input data set, there has not been much degradation in the performance of the network. Also a comparison with table 5.7 reveals that the number of centers required for the desired level of performance by the RBF network is lesser when compared to random selection of centers. In the next learning strategy, gradient descent has been used to adapt all the adjustable parameters of the RBF network Center initialization using Gradient Descent approach In this strategy, a radial basis function network has been trained using the gradient descent approach. All the adjustable parameters of the network are adapted using this approach. Karayiannis [37] proposed a supervised learning algorithm based on gradient descent for training reformulated RBF neural networks. Experiments involved a variety of reformulated RBF networks generated by linear and exponential generator functions. This indicated that gradient descent learning is simple, easily implementable and produces RBF networks that perform considerably better than conventional RBF models trained by existing algorithms. The algorithm for gradient descent learning is presented below. 09

17 Algorithm:. Select the number of centers, leaming constant r, momentum coefficient a and N (maximum no. of epochs). Randomly initialize weights to small values. The centers are selected randomly from the input data. The width of the hidden units is fixed at the begirming (constant for all the units). 2. Set Epoch=0, Etot =0 & present the first training pattern (^i, ^k)- 3. Compute the hidden layer responses Vj = e ""J'^j'"^ j. The network output is given by Ok=l/(I+e-%^j). 4. Compute the error i.e. square error, = (Ck - Ok) * (Ck - Ok) and Etot +=En (n = number of each pattern) 5. Update the adjustable parameters i.e. weights of the output layer using gradient descent method 5k = ( Ok - Ck) * Ok *(- Ok), Awkj = 5k * Vj * TI * a, W "%j = w '\j + Awkj. 6. Present the next training pattern and goto Step After a fixed number of epochs (one epoch is presentation of all the patterns in the training data set), update the center values using gradient descent approach as follows: Xj""^ = Xj"'"* + Ti * a * %, 5j * ^i where 5j = II ^i - xj II / a'j * II ^i - Xj II * e ""^ ^''''''' * Zk=i 5k * Wkj. Update the widths using P- nearest neighbor heuristic. 8. Present the first training pattern and goto Step If Etot< Emin and epoch>n, Stop. 0. Save weights, centers and widths and exit. (a) Training phase - The training of the network has been carried out on En-8 data set. The widths GJ have been kept constant at 0.9 initially for all the hidden units. The centers are adjusted for every 5000 epochs and the corresponding widths have been determined using the P- nearest neighbor heuristic, i.e., the network training started with equal widths for all the hidden units and then adjusted. This methodology has been adopted to achieve the desirable accuracy. Table 5.6 shows the variation of error with number of RBF units. 0

18 Table 5.6: Variation of Error witii Number of RBF Units No. of RBF units Error It is clear that for 60 RBF units, the error converged by the network during training is minimum. Fig. 5.6 shows the final RBF network architecture. Input layer Hidden layer Output layer ;«Fig. 5.8 RBF: Final Network architecture (2-60-) Centers initialized using Gradient Descent approach (b) Testing phase - centers. The final RBF network architecture has been tested using test samples. Table 5.7 shows the testing results for test samples for different number of Table 5.7: Performance of RBF network with initialization of centers using gradient descent No. of RBF units Error Training accuracy Testing accuracy % 73% % 87% % 87% % 80%

19 From the table it is clear that the optimal performance in terms of classification accuracy on test samples and error has been achieved for 60 RBF units. Table 5.8 shows the sample testing results for 60 centers. Table 5.8: Sample Testing Results (2-60-, Gradient Descent) Pattern No. Desired output Network output (mm) (mm) Training data Test data Classiflcation error X0"^ X0"' The network architecture has been able to recognize 9 % of the training and 87 % test patterns respectively. 5.5 Discussion Investigations have been carried out on RBF neural networks using three approaches for center initialization namely random selection from the training data, using batch fiizzy c-means algorithm and gradient descent approach. Random selection of centers from the training data set requires lot of trials for establishing the right number of centers. Where as in case of batch fuzzy c-means algorithm, the algorithm helps in determining the optimum number of centers for desirable performance. The use of batch fiizzy c-means algorithm for establishing the centers is a robust method and will always 2

20 guarantee good performance, because the membership functions used will determine the strength of attraction between the centers and the input vectors. The use of batch fuzzy c- means will definitely prevent the deterioration in the performance of the neural network due to noisy data, which is possible in a practical shop floor environment. The use of gradient descent approach adapts all the network parameters simultaneously. Though during learning, the error does not converge to a global minimum, the developed network architecture has been able to recognize most of the training and test patterns. Table 5.9 shows the comparative evaluation of fixed RBF network with three methods of center initialization and Multi-Layer Perceptron network for En-8 data set. Table 5.9: Comparative Evaluation of RBF Network and MLP (En-8) Network RBF network Network architect ure Training data Accuracy Test data No. of epochs i) Random initialization ii) Batch fuzzy-c means iii) Gradient descent MLP % 97% 9% 00 % 87% 87% MLP takes lesser training time when compared to RBF networks. The performance of both the networks are comparable, as both are robust and accurate in estimating the flank wear values. MLP requires less number of hidden units when compared to RBF networks, for the same level of performance. The generalization capability of RBF networks for three methods of center initialization has been found to be better when compared to MLP. The RBF networks are able to generalize well based on the method adopted for center initialization. The developed network architecture has been able to classify almost all of the seen patterns. The network has been able to classify 93 % of the unseen patterns 3

21 for random initialization and batch fuzzy c-means algorithm and 87 % of the unseen patterns for gradient descent approach, indicating that RBF neural networks are powerfiil architecture for pattern classification. 5.6 Conclusion: In this chapter,. RBF neural networks have been applied for tool wear monitoring for determining the tool status in face milling operations. 2. Three center initialization strategies for the RBF units in the hidden layer have been investigated. 3. For each center initialization approach and width selection a number of experiments have been conducted to train the network and the learning characteristics have been thoroughly analyzed. 4. RBF neural networks have been found to exhibit better learning and generalization abilities in estimating the flank wear values, for any of the center initialization strategies, when compared to MLP. 5. RBF neural networks have been effective in monitoring the condition of the tool during face milling operations with 93 % accuracy, when unseen signal patterns are presented to the network. In the next chapter, application of Resource Allocation Network for tool wear monitoring is presented. 4

Neural Networks. CE-725: Statistical Pattern Recognition Sharif University of Technology Spring Soleymani

Neural Networks. CE-725: Statistical Pattern Recognition Sharif University of Technology Spring Soleymani Neural Networks CE-725: Statistical Pattern Recognition Sharif University of Technology Spring 2013 Soleymani Outline Biological and artificial neural networks Feed-forward neural networks Single layer

More information

Function approximation using RBF network. 10 basis functions and 25 data points.

Function approximation using RBF network. 10 basis functions and 25 data points. 1 Function approximation using RBF network F (x j ) = m 1 w i ϕ( x j t i ) i=1 j = 1... N, m 1 = 10, N = 25 10 basis functions and 25 data points. Basis function centers are plotted with circles and data

More information

COMPUTATIONAL INTELLIGENCE

COMPUTATIONAL INTELLIGENCE COMPUTATIONAL INTELLIGENCE Radial Basis Function Networks Adrian Horzyk Preface Radial Basis Function Networks (RBFN) are a kind of artificial neural networks that use radial basis functions (RBF) as activation

More information

Radial Basis Function Neural Network Classifier

Radial Basis Function Neural Network Classifier Recognition of Unconstrained Handwritten Numerals by a Radial Basis Function Neural Network Classifier Hwang, Young-Sup and Bang, Sung-Yang Department of Computer Science & Engineering Pohang University

More information

CHAPTER IX Radial Basis Function Networks

CHAPTER IX Radial Basis Function Networks CHAPTER IX Radial Basis Function Networks Radial basis function (RBF) networks are feed-forward networks trained using a supervised training algorithm. They are typically configured with a single hidden

More information

CSE 5526: Introduction to Neural Networks Radial Basis Function (RBF) Networks

CSE 5526: Introduction to Neural Networks Radial Basis Function (RBF) Networks CSE 5526: Introduction to Neural Networks Radial Basis Function (RBF) Networks Part IV 1 Function approximation MLP is both a pattern classifier and a function approximator As a function approximator,

More information

Pattern Classification Algorithms for Face Recognition

Pattern Classification Algorithms for Face Recognition Chapter 7 Pattern Classification Algorithms for Face Recognition 7.1 Introduction The best pattern recognizers in most instances are human beings. Yet we do not completely understand how the brain recognize

More information

Automatic basis selection for RBF networks using Stein s unbiased risk estimator

Automatic basis selection for RBF networks using Stein s unbiased risk estimator Automatic basis selection for RBF networks using Stein s unbiased risk estimator Ali Ghodsi School of omputer Science University of Waterloo University Avenue West NL G anada Email: aghodsib@cs.uwaterloo.ca

More information

Classification and Regression using Linear Networks, Multilayer Perceptrons and Radial Basis Functions

Classification and Regression using Linear Networks, Multilayer Perceptrons and Radial Basis Functions ENEE 739Q SPRING 2002 COURSE ASSIGNMENT 2 REPORT 1 Classification and Regression using Linear Networks, Multilayer Perceptrons and Radial Basis Functions Vikas Chandrakant Raykar Abstract The aim of the

More information

4.12 Generalization. In back-propagation learning, as many training examples as possible are typically used.

4.12 Generalization. In back-propagation learning, as many training examples as possible are typically used. 1 4.12 Generalization In back-propagation learning, as many training examples as possible are typically used. It is hoped that the network so designed generalizes well. A network generalizes well when

More information

CLASSIFICATION WITH RADIAL BASIS AND PROBABILISTIC NEURAL NETWORKS

CLASSIFICATION WITH RADIAL BASIS AND PROBABILISTIC NEURAL NETWORKS CLASSIFICATION WITH RADIAL BASIS AND PROBABILISTIC NEURAL NETWORKS CHAPTER 4 CLASSIFICATION WITH RADIAL BASIS AND PROBABILISTIC NEURAL NETWORKS 4.1 Introduction Optical character recognition is one of

More information

Neural Network Learning. Today s Lecture. Continuation of Neural Networks. Artificial Neural Networks. Lecture 24: Learning 3. Victor R.

Neural Network Learning. Today s Lecture. Continuation of Neural Networks. Artificial Neural Networks. Lecture 24: Learning 3. Victor R. Lecture 24: Learning 3 Victor R. Lesser CMPSCI 683 Fall 2010 Today s Lecture Continuation of Neural Networks Artificial Neural Networks Compose of nodes/units connected by links Each link has a numeric

More information

742 IEEE TRANSACTIONS ON FUZZY SYSTEMS, VOL. 13, NO. 6, DECEMBER Dong Zhang, Luo-Feng Deng, Kai-Yuan Cai, and Albert So

742 IEEE TRANSACTIONS ON FUZZY SYSTEMS, VOL. 13, NO. 6, DECEMBER Dong Zhang, Luo-Feng Deng, Kai-Yuan Cai, and Albert So 742 IEEE TRANSACTIONS ON FUZZY SYSTEMS, VOL 13, NO 6, DECEMBER 2005 Fuzzy Nonlinear Regression With Fuzzified Radial Basis Function Network Dong Zhang, Luo-Feng Deng, Kai-Yuan Cai, and Albert So Abstract

More information

Radial Basis Function Networks: Algorithms

Radial Basis Function Networks: Algorithms Radial Basis Function Networks: Algorithms Neural Computation : Lecture 14 John A. Bullinaria, 2015 1. The RBF Mapping 2. The RBF Network Architecture 3. Computational Power of RBF Networks 4. Training

More information

Supervised Learning in Neural Networks (Part 2)

Supervised Learning in Neural Networks (Part 2) Supervised Learning in Neural Networks (Part 2) Multilayer neural networks (back-propagation training algorithm) The input signals are propagated in a forward direction on a layer-bylayer basis. Learning

More information

Artificial Neural Networks MLP, RBF & GMDH

Artificial Neural Networks MLP, RBF & GMDH Artificial Neural Networks MLP, RBF & GMDH Jan Drchal drchajan@fel.cvut.cz Computational Intelligence Group Department of Computer Science and Engineering Faculty of Electrical Engineering Czech Technical

More information

CS6220: DATA MINING TECHNIQUES

CS6220: DATA MINING TECHNIQUES CS6220: DATA MINING TECHNIQUES Image Data: Classification via Neural Networks Instructor: Yizhou Sun yzsun@ccs.neu.edu November 19, 2015 Methods to Learn Classification Clustering Frequent Pattern Mining

More information

Support Vector Machines

Support Vector Machines Support Vector Machines RBF-networks Support Vector Machines Good Decision Boundary Optimization Problem Soft margin Hyperplane Non-linear Decision Boundary Kernel-Trick Approximation Accurancy Overtraining

More information

Cursive Handwriting Recognition System Using Feature Extraction and Artificial Neural Network

Cursive Handwriting Recognition System Using Feature Extraction and Artificial Neural Network Cursive Handwriting Recognition System Using Feature Extraction and Artificial Neural Network Utkarsh Dwivedi 1, Pranjal Rajput 2, Manish Kumar Sharma 3 1UG Scholar, Dept. of CSE, GCET, Greater Noida,

More information

In this assignment, we investigated the use of neural networks for supervised classification

In this assignment, we investigated the use of neural networks for supervised classification Paul Couchman Fabien Imbault Ronan Tigreat Gorka Urchegui Tellechea Classification assignment (group 6) Image processing MSc Embedded Systems March 2003 Classification includes a broad range of decision-theoric

More information

IMPLEMENTATION OF RBF TYPE NETWORKS BY SIGMOIDAL FEEDFORWARD NEURAL NETWORKS

IMPLEMENTATION OF RBF TYPE NETWORKS BY SIGMOIDAL FEEDFORWARD NEURAL NETWORKS IMPLEMENTATION OF RBF TYPE NETWORKS BY SIGMOIDAL FEEDFORWARD NEURAL NETWORKS BOGDAN M.WILAMOWSKI University of Wyoming RICHARD C. JAEGER Auburn University ABSTRACT: It is shown that by introducing special

More information

Learning. Learning agents Inductive learning. Neural Networks. Different Learning Scenarios Evaluation

Learning. Learning agents Inductive learning. Neural Networks. Different Learning Scenarios Evaluation Learning Learning agents Inductive learning Different Learning Scenarios Evaluation Slides based on Slides by Russell/Norvig, Ronald Williams, and Torsten Reil Material from Russell & Norvig, chapters

More information

Figure (5) Kohonen Self-Organized Map

Figure (5) Kohonen Self-Organized Map 2- KOHONEN SELF-ORGANIZING MAPS (SOM) - The self-organizing neural networks assume a topological structure among the cluster units. - There are m cluster units, arranged in a one- or two-dimensional array;

More information

Perceptron as a graph

Perceptron as a graph Neural Networks Machine Learning 10701/15781 Carlos Guestrin Carnegie Mellon University October 10 th, 2007 2005-2007 Carlos Guestrin 1 Perceptron as a graph 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0-6 -4-2

More information

Support Vector Machines

Support Vector Machines Support Vector Machines RBF-networks Support Vector Machines Good Decision Boundary Optimization Problem Soft margin Hyperplane Non-linear Decision Boundary Kernel-Trick Approximation Accurancy Overtraining

More information

Machine Learning Classifiers and Boosting

Machine Learning Classifiers and Boosting Machine Learning Classifiers and Boosting Reading Ch 18.6-18.12, 20.1-20.3.2 Outline Different types of learning problems Different types of learning algorithms Supervised learning Decision trees Naïve

More information

Neural Networks (Overview) Prof. Richard Zanibbi

Neural Networks (Overview) Prof. Richard Zanibbi Neural Networks (Overview) Prof. Richard Zanibbi Inspired by Biology Introduction But as used in pattern recognition research, have little relation with real neural systems (studied in neurology and neuroscience)

More information

The exam is closed book, closed notes except your one-page cheat sheet.

The exam is closed book, closed notes except your one-page cheat sheet. CS 189 Fall 2015 Introduction to Machine Learning Final Please do not turn over the page before you are instructed to do so. You have 2 hours and 50 minutes. Please write your initials on the top-right

More information

More on Learning. Neural Nets Support Vectors Machines Unsupervised Learning (Clustering) K-Means Expectation-Maximization

More on Learning. Neural Nets Support Vectors Machines Unsupervised Learning (Clustering) K-Means Expectation-Maximization More on Learning Neural Nets Support Vectors Machines Unsupervised Learning (Clustering) K-Means Expectation-Maximization Neural Net Learning Motivated by studies of the brain. A network of artificial

More information

4. Feedforward neural networks. 4.1 Feedforward neural network structure

4. Feedforward neural networks. 4.1 Feedforward neural network structure 4. Feedforward neural networks 4.1 Feedforward neural network structure Feedforward neural network is one of the most common network architectures. Its structure and some basic preprocessing issues required

More information

CHAPTER 7 MASS LOSS PREDICTION USING ARTIFICIAL NEURAL NETWORK (ANN)

CHAPTER 7 MASS LOSS PREDICTION USING ARTIFICIAL NEURAL NETWORK (ANN) 128 CHAPTER 7 MASS LOSS PREDICTION USING ARTIFICIAL NEURAL NETWORK (ANN) Various mathematical techniques like regression analysis and software tools have helped to develop a model using equation, which

More information

Data Mining. Neural Networks

Data Mining. Neural Networks Data Mining Neural Networks Goals for this Unit Basic understanding of Neural Networks and how they work Ability to use Neural Networks to solve real problems Understand when neural networks may be most

More information

Machine Learning : Clustering, Self-Organizing Maps

Machine Learning : Clustering, Self-Organizing Maps Machine Learning Clustering, Self-Organizing Maps 12/12/2013 Machine Learning : Clustering, Self-Organizing Maps Clustering The task: partition a set of objects into meaningful subsets (clusters). The

More information

Dynamic Analysis of Structures Using Neural Networks

Dynamic Analysis of Structures Using Neural Networks Dynamic Analysis of Structures Using Neural Networks Alireza Lavaei Academic member, Islamic Azad University, Boroujerd Branch, Iran Alireza Lohrasbi Academic member, Islamic Azad University, Boroujerd

More information

1. Introduction. 2. Motivation and Problem Definition. Volume 8 Issue 2, February Susmita Mohapatra

1. Introduction. 2. Motivation and Problem Definition. Volume 8 Issue 2, February Susmita Mohapatra Pattern Recall Analysis of the Hopfield Neural Network with a Genetic Algorithm Susmita Mohapatra Department of Computer Science, Utkal University, India Abstract: This paper is focused on the implementation

More information

Artificial Neural Networks (Feedforward Nets)

Artificial Neural Networks (Feedforward Nets) Artificial Neural Networks (Feedforward Nets) y w 03-1 w 13 y 1 w 23 y 2 w 01 w 21 w 22 w 02-1 w 11 w 12-1 x 1 x 2 6.034 - Spring 1 Single Perceptron Unit y w 0 w 1 w n w 2 w 3 x 0 =1 x 1 x 2 x 3... x

More information

Combined Weak Classifiers

Combined Weak Classifiers Combined Weak Classifiers Chuanyi Ji and Sheng Ma Department of Electrical, Computer and System Engineering Rensselaer Polytechnic Institute, Troy, NY 12180 chuanyi@ecse.rpi.edu, shengm@ecse.rpi.edu Abstract

More information

Visual object classification by sparse convolutional neural networks

Visual object classification by sparse convolutional neural networks Visual object classification by sparse convolutional neural networks Alexander Gepperth 1 1- Ruhr-Universität Bochum - Institute for Neural Dynamics Universitätsstraße 150, 44801 Bochum - Germany Abstract.

More information

Radial Basis Function Networks

Radial Basis Function Networks Radial Basis Function Networks As we have seen, one of the most common types of neural network is the multi-layer perceptron It does, however, have various disadvantages, including the slow speed in learning

More information

COMP 551 Applied Machine Learning Lecture 14: Neural Networks

COMP 551 Applied Machine Learning Lecture 14: Neural Networks COMP 551 Applied Machine Learning Lecture 14: Neural Networks Instructor: (jpineau@cs.mcgill.ca) Class web page: www.cs.mcgill.ca/~jpineau/comp551 Unless otherwise noted, all material posted for this course

More information

Use of Artificial Neural Networks to Investigate the Surface Roughness in CNC Milling Machine

Use of Artificial Neural Networks to Investigate the Surface Roughness in CNC Milling Machine Use of Artificial Neural Networks to Investigate the Surface Roughness in CNC Milling Machine M. Vijay Kumar Reddy 1 1 Department of Mechanical Engineering, Annamacharya Institute of Technology and Sciences,

More information

In the Name of God. Lecture 17: ANFIS Adaptive Network-Based Fuzzy Inference System

In the Name of God. Lecture 17: ANFIS Adaptive Network-Based Fuzzy Inference System In the Name of God Lecture 17: ANFIS Adaptive Network-Based Fuzzy Inference System Outline ANFIS Architecture Hybrid Learning Algorithm Learning Methods that Cross-Fertilize ANFIS and RBFN ANFIS as a universal

More information

CHAPTER 3 TUMOR DETECTION BASED ON NEURO-FUZZY TECHNIQUE

CHAPTER 3 TUMOR DETECTION BASED ON NEURO-FUZZY TECHNIQUE 32 CHAPTER 3 TUMOR DETECTION BASED ON NEURO-FUZZY TECHNIQUE 3.1 INTRODUCTION In this chapter we present the real time implementation of an artificial neural network based on fuzzy segmentation process

More information

CHAPTER 6 COUNTER PROPAGATION NEURAL NETWORK FOR IMAGE RESTORATION

CHAPTER 6 COUNTER PROPAGATION NEURAL NETWORK FOR IMAGE RESTORATION 135 CHAPTER 6 COUNTER PROPAGATION NEURAL NETWORK FOR IMAGE RESTORATION 6.1 INTRODUCTION Neural networks have high fault tolerance and potential for adaptive training. A Full Counter Propagation Neural

More information

RADIAL BASIS FUNCTIONS NETWORK FOR DEFECT SIZING. S. Nair, S. Udpa and L. Udpa Center for Non-Destructive Evaluation Scholl Road Ames, IA 50011

RADIAL BASIS FUNCTIONS NETWORK FOR DEFECT SIZING. S. Nair, S. Udpa and L. Udpa Center for Non-Destructive Evaluation Scholl Road Ames, IA 50011 RADAL BASS FUNCTONS NETWORK FOR DEFECT SZNG S. Nair, S. Udpa and L. Udpa Center for Non-Destructive Evaluation Scholl Road Ames, A 50011 NTRODUCTON An important aspect of non-destructive testing is the

More information

For Monday. Read chapter 18, sections Homework:

For Monday. Read chapter 18, sections Homework: For Monday Read chapter 18, sections 10-12 The material in section 8 and 9 is interesting, but we won t take time to cover it this semester Homework: Chapter 18, exercise 25 a-b Program 4 Model Neuron

More information

An Algorithm For Training Multilayer Perceptron (MLP) For Image Reconstruction Using Neural Network Without Overfitting.

An Algorithm For Training Multilayer Perceptron (MLP) For Image Reconstruction Using Neural Network Without Overfitting. An Algorithm For Training Multilayer Perceptron (MLP) For Image Reconstruction Using Neural Network Without Overfitting. Mohammad Mahmudul Alam Mia, Shovasis Kumar Biswas, Monalisa Chowdhury Urmi, Abubakar

More information

A Class of Instantaneously Trained Neural Networks

A Class of Instantaneously Trained Neural Networks A Class of Instantaneously Trained Neural Networks Subhash Kak Department of Electrical & Computer Engineering, Louisiana State University, Baton Rouge, LA 70803-5901 May 7, 2002 Abstract This paper presents

More information

COMPUTATIONAL INTELLIGENCE SEW (INTRODUCTION TO MACHINE LEARNING) SS18. Lecture 6: k-nn Cross-validation Regularization

COMPUTATIONAL INTELLIGENCE SEW (INTRODUCTION TO MACHINE LEARNING) SS18. Lecture 6: k-nn Cross-validation Regularization COMPUTATIONAL INTELLIGENCE SEW (INTRODUCTION TO MACHINE LEARNING) SS18 Lecture 6: k-nn Cross-validation Regularization LEARNING METHODS Lazy vs eager learning Eager learning generalizes training data before

More information

Improving the way neural networks learn Srikumar Ramalingam School of Computing University of Utah

Improving the way neural networks learn Srikumar Ramalingam School of Computing University of Utah Improving the way neural networks learn Srikumar Ramalingam School of Computing University of Utah Reference Most of the slides are taken from the third chapter of the online book by Michael Nielson: neuralnetworksanddeeplearning.com

More information

Unsupervised Learning : Clustering

Unsupervised Learning : Clustering Unsupervised Learning : Clustering Things to be Addressed Traditional Learning Models. Cluster Analysis K-means Clustering Algorithm Drawbacks of traditional clustering algorithms. Clustering as a complex

More information

CPSC 340: Machine Learning and Data Mining. Principal Component Analysis Fall 2016

CPSC 340: Machine Learning and Data Mining. Principal Component Analysis Fall 2016 CPSC 340: Machine Learning and Data Mining Principal Component Analysis Fall 2016 A2/Midterm: Admin Grades/solutions will be posted after class. Assignment 4: Posted, due November 14. Extra office hours:

More information

3 Nonlinear Regression

3 Nonlinear Regression 3 Linear models are often insufficient to capture the real-world phenomena. That is, the relation between the inputs and the outputs we want to be able to predict are not linear. As a consequence, nonlinear

More information

3 Nonlinear Regression

3 Nonlinear Regression CSC 4 / CSC D / CSC C 3 Sometimes linear models are not sufficient to capture the real-world phenomena, and thus nonlinear models are necessary. In regression, all such models will have the same basic

More information

Univariate and Multivariate Decision Trees

Univariate and Multivariate Decision Trees Univariate and Multivariate Decision Trees Olcay Taner Yıldız and Ethem Alpaydın Department of Computer Engineering Boğaziçi University İstanbul 80815 Turkey Abstract. Univariate decision trees at each

More information

A neural network that classifies glass either as window or non-window depending on the glass chemistry.

A neural network that classifies glass either as window or non-window depending on the glass chemistry. A neural network that classifies glass either as window or non-window depending on the glass chemistry. Djaber Maouche Department of Electrical Electronic Engineering Cukurova University Adana, Turkey

More information

Neural Network Optimization and Tuning / Spring 2018 / Recitation 3

Neural Network Optimization and Tuning / Spring 2018 / Recitation 3 Neural Network Optimization and Tuning 11-785 / Spring 2018 / Recitation 3 1 Logistics You will work through a Jupyter notebook that contains sample and starter code with explanations and comments throughout.

More information

Efficient Object Tracking Using K means and Radial Basis Function

Efficient Object Tracking Using K means and Radial Basis Function Efficient Object Tracing Using K means and Radial Basis Function Mr. Pradeep K. Deshmuh, Ms. Yogini Gholap University of Pune Department of Post Graduate Computer Engineering, JSPM S Rajarshi Shahu College

More information

Unsupervised Learning

Unsupervised Learning Outline Unsupervised Learning Basic concepts K-means algorithm Representation of clusters Hierarchical clustering Distance functions Which clustering algorithm to use? NN Supervised learning vs. unsupervised

More information

5 Learning hypothesis classes (16 points)

5 Learning hypothesis classes (16 points) 5 Learning hypothesis classes (16 points) Consider a classification problem with two real valued inputs. For each of the following algorithms, specify all of the separators below that it could have generated

More information

Fuzzy Signature Neural Networks for Classification: Optimising the Structure

Fuzzy Signature Neural Networks for Classification: Optimising the Structure Fuzzy Signature Neural Networks for Classification: Optimising the Structure Tom Gedeon, Xuanying Zhu, Kun He, and Leana Copeland Research School of Computer Science, College of Engineering and Computer

More information

Hybrid Training Algorithm for RBF Network

Hybrid Training Algorithm for RBF Network Hybrid Training Algorithm for RBF Network By M. Y. MASHOR School of Electrical and Electronic Engineering, University Science of Malaysia, Perak Branch Campus, 3750 Tronoh, Perak, Malaysia. E-mail: yusof@eng.usm.my

More information

Performance Evaluation of a Radial Basis Function Neural Network Learning Algorithm for Function Approximation.

Performance Evaluation of a Radial Basis Function Neural Network Learning Algorithm for Function Approximation. Performance Evaluation of a Radial Basis Function Neural Network Learning Algorithm for Function Approximation. A.A. Khurshid PCEA, Nagpur, India A.P.Gokhale VNIT, Nagpur, India Abstract: This paper presents

More information

Image Compression: An Artificial Neural Network Approach

Image Compression: An Artificial Neural Network Approach Image Compression: An Artificial Neural Network Approach Anjana B 1, Mrs Shreeja R 2 1 Department of Computer Science and Engineering, Calicut University, Kuttippuram 2 Department of Computer Science and

More information

A Topography-Preserving Latent Variable Model with Learning Metrics

A Topography-Preserving Latent Variable Model with Learning Metrics A Topography-Preserving Latent Variable Model with Learning Metrics Samuel Kaski and Janne Sinkkonen Helsinki University of Technology Neural Networks Research Centre P.O. Box 5400, FIN-02015 HUT, Finland

More information

Assignment 1: CS Machine Learning

Assignment 1: CS Machine Learning Assignment 1: CS7641 - Machine Learning Saad Khan September 18, 2015 1 Introduction I intend to apply supervised learning algorithms to classify the quality of wine samples as being of high or low quality

More information

IMPROVEMENTS TO THE BACKPROPAGATION ALGORITHM

IMPROVEMENTS TO THE BACKPROPAGATION ALGORITHM Annals of the University of Petroşani, Economics, 12(4), 2012, 185-192 185 IMPROVEMENTS TO THE BACKPROPAGATION ALGORITHM MIRCEA PETRINI * ABSTACT: This paper presents some simple techniques to improve

More information

Efficient Training of RBF Neural Networks for Pattern Recognition

Efficient Training of RBF Neural Networks for Pattern Recognition IEEE TRANSACTIONS ON NEURAL NETWORKS, VOL. 12, NO. 5, SEPTEMBER 2001 1235 Efficient Training of RBF Neural Networks for Pattern Recognition Francesco Lampariello and Marco Sciandrone Abstract The problem

More information

The rest of the paper is organized as follows: we rst shortly describe the \growing neural gas" method which we have proposed earlier [3]. Then the co

The rest of the paper is organized as follows: we rst shortly describe the \growing neural gas method which we have proposed earlier [3]. Then the co In: F. Fogelman and P. Gallinari, editors, ICANN'95: International Conference on Artificial Neural Networks, pages 217-222, Paris, France, 1995. EC2 & Cie. Incremental Learning of Local Linear Mappings

More information

Classification Lecture Notes cse352. Neural Networks. Professor Anita Wasilewska

Classification Lecture Notes cse352. Neural Networks. Professor Anita Wasilewska Classification Lecture Notes cse352 Neural Networks Professor Anita Wasilewska Neural Networks Classification Introduction INPUT: classification data, i.e. it contains an classification (class) attribute

More information

LECTURE NOTES Professor Anita Wasilewska NEURAL NETWORKS

LECTURE NOTES Professor Anita Wasilewska NEURAL NETWORKS LECTURE NOTES Professor Anita Wasilewska NEURAL NETWORKS Neural Networks Classifier Introduction INPUT: classification data, i.e. it contains an classification (class) attribute. WE also say that the class

More information

Introduction to Machine Learning CMU-10701

Introduction to Machine Learning CMU-10701 Introduction to Machine Learning CMU-10701 Clustering and EM Barnabás Póczos & Aarti Singh Contents Clustering K-means Mixture of Gaussians Expectation Maximization Variational Methods 2 Clustering 3 K-

More information

Vulnerability of machine learning models to adversarial examples

Vulnerability of machine learning models to adversarial examples ITAT 216 Proceedings, CEUR Workshop Proceedings Vol. 1649, pp. 187 194 http://ceur-ws.org/vol-1649, Series ISSN 1613-73, c 216 P. Vidnerová, R. Neruda Vulnerability of machine learning models to adversarial

More information

Neural Networks Laboratory EE 329 A

Neural Networks Laboratory EE 329 A Neural Networks Laboratory EE 329 A Introduction: Artificial Neural Networks (ANN) are widely used to approximate complex systems that are difficult to model using conventional modeling techniques such

More information

The exam is closed book, closed notes except your one-page (two-sided) cheat sheet.

The exam is closed book, closed notes except your one-page (two-sided) cheat sheet. CS 189 Spring 2015 Introduction to Machine Learning Final You have 2 hours 50 minutes for the exam. The exam is closed book, closed notes except your one-page (two-sided) cheat sheet. No calculators or

More information

DESIGN AND EVALUATION OF MACHINE LEARNING MODELS WITH STATISTICAL FEATURES

DESIGN AND EVALUATION OF MACHINE LEARNING MODELS WITH STATISTICAL FEATURES EXPERIMENTAL WORK PART I CHAPTER 6 DESIGN AND EVALUATION OF MACHINE LEARNING MODELS WITH STATISTICAL FEATURES The evaluation of models built using statistical in conjunction with various feature subset

More information

Dr. Qadri Hamarsheh Supervised Learning in Neural Networks (Part 1) learning algorithm Δwkj wkj Theoretically practically

Dr. Qadri Hamarsheh Supervised Learning in Neural Networks (Part 1) learning algorithm Δwkj wkj Theoretically practically Supervised Learning in Neural Networks (Part 1) A prescribed set of well-defined rules for the solution of a learning problem is called a learning algorithm. Variety of learning algorithms are existing,

More information

TWRBF Transductive RBF Neural Network with Weighted Data Normalization

TWRBF Transductive RBF Neural Network with Weighted Data Normalization TWRBF Transductive RBF eural etwork with Weighted Data ormalization Qun Song and ikola Kasabov Knowledge Engineering & Discovery Research Institute Auckland University of Technology Private Bag 9006, Auckland

More information

Cluster Analysis. Mu-Chun Su. Department of Computer Science and Information Engineering National Central University 2003/3/11 1

Cluster Analysis. Mu-Chun Su. Department of Computer Science and Information Engineering National Central University 2003/3/11 1 Cluster Analysis Mu-Chun Su Department of Computer Science and Information Engineering National Central University 2003/3/11 1 Introduction Cluster analysis is the formal study of algorithms and methods

More information

Neural Network Neurons

Neural Network Neurons Neural Networks Neural Network Neurons 1 Receives n inputs (plus a bias term) Multiplies each input by its weight Applies activation function to the sum of results Outputs result Activation Functions Given

More information

CHAPTER 6 MODIFIED FUZZY TECHNIQUES BASED IMAGE SEGMENTATION

CHAPTER 6 MODIFIED FUZZY TECHNIQUES BASED IMAGE SEGMENTATION CHAPTER 6 MODIFIED FUZZY TECHNIQUES BASED IMAGE SEGMENTATION 6.1 INTRODUCTION Fuzzy logic based computational techniques are becoming increasingly important in the medical image analysis arena. The significant

More information

Statistical Learning Part 2 Nonparametric Learning: The Main Ideas. R. Moeller Hamburg University of Technology

Statistical Learning Part 2 Nonparametric Learning: The Main Ideas. R. Moeller Hamburg University of Technology Statistical Learning Part 2 Nonparametric Learning: The Main Ideas R. Moeller Hamburg University of Technology Instance-Based Learning So far we saw statistical learning as parameter learning, i.e., given

More information

Supervised vs. Unsupervised Learning

Supervised vs. Unsupervised Learning Clustering Supervised vs. Unsupervised Learning So far we have assumed that the training samples used to design the classifier were labeled by their class membership (supervised learning) We assume now

More information

Accurate modeling of SiGe HBT using artificial neural networks: Performance Comparison of the MLP and RBF Networks

Accurate modeling of SiGe HBT using artificial neural networks: Performance Comparison of the MLP and RBF Networks Accurate modeling of SiGe HBT using artificial neural networks: Performance Comparison of the MLP and RBF etworks Malek Amiri Abdeboochali Department of Electrical Engineering Razi University Kermanshah,

More information

arxiv: v1 [stat.ml] 21 Feb 2018

arxiv: v1 [stat.ml] 21 Feb 2018 Detecting Learning vs Memorization in Deep Neural Networks using Shared Structure Validation Sets arxiv:2.0774v [stat.ml] 2 Feb 8 Elias Chaibub Neto e-mail: elias.chaibub.neto@sagebase.org, Sage Bionetworks

More information

Today. Gradient descent for minimization of functions of real variables. Multi-dimensional scaling. Self-organizing maps

Today. Gradient descent for minimization of functions of real variables. Multi-dimensional scaling. Self-organizing maps Today Gradient descent for minimization of functions of real variables. Multi-dimensional scaling Self-organizing maps Gradient Descent Derivatives Consider function f(x) : R R. The derivative w.r.t. x

More information

Artificial neural networks are the paradigm of connectionist systems (connectionism vs. symbolism)

Artificial neural networks are the paradigm of connectionist systems (connectionism vs. symbolism) Artificial Neural Networks Analogy to biological neural systems, the most robust learning systems we know. Attempt to: Understand natural biological systems through computational modeling. Model intelligent

More information

CHAPTER 6 IMPLEMENTATION OF RADIAL BASIS FUNCTION NEURAL NETWORK FOR STEGANALYSIS

CHAPTER 6 IMPLEMENTATION OF RADIAL BASIS FUNCTION NEURAL NETWORK FOR STEGANALYSIS 95 CHAPTER 6 IMPLEMENTATION OF RADIAL BASIS FUNCTION NEURAL NETWORK FOR STEGANALYSIS 6.1 INTRODUCTION The concept of distance measure is used to associate the input and output pattern values. RBFs use

More information

11/14/2010 Intelligent Systems and Soft Computing 1

11/14/2010 Intelligent Systems and Soft Computing 1 Lecture 7 Artificial neural networks: Supervised learning Introduction, or how the brain works The neuron as a simple computing element The perceptron Multilayer neural networks Accelerated learning in

More information

Neural Network Weight Selection Using Genetic Algorithms

Neural Network Weight Selection Using Genetic Algorithms Neural Network Weight Selection Using Genetic Algorithms David Montana presented by: Carl Fink, Hongyi Chen, Jack Cheng, Xinglong Li, Bruce Lin, Chongjie Zhang April 12, 2005 1 Neural Networks Neural networks

More information

MLPQNA-LEMON Multi Layer Perceptron neural network trained by Quasi Newton or Levenberg-Marquardt optimization algorithms

MLPQNA-LEMON Multi Layer Perceptron neural network trained by Quasi Newton or Levenberg-Marquardt optimization algorithms MLPQNA-LEMON Multi Layer Perceptron neural network trained by Quasi Newton or Levenberg-Marquardt optimization algorithms 1 Introduction In supervised Machine Learning (ML) we have a set of data points

More information

Pattern Recognition. Kjell Elenius. Speech, Music and Hearing KTH. March 29, 2007 Speech recognition

Pattern Recognition. Kjell Elenius. Speech, Music and Hearing KTH. March 29, 2007 Speech recognition Pattern Recognition Kjell Elenius Speech, Music and Hearing KTH March 29, 2007 Speech recognition 2007 1 Ch 4. Pattern Recognition 1(3) Bayes Decision Theory Minimum-Error-Rate Decision Rules Discriminant

More information

Deep Neural Networks Optimization

Deep Neural Networks Optimization Deep Neural Networks Optimization Creative Commons (cc) by Akritasa http://arxiv.org/pdf/1406.2572.pdf Slides from Geoffrey Hinton CSC411/2515: Machine Learning and Data Mining, Winter 2018 Michael Guerzhoy

More information

Neural Networks. Theory And Practice. Marco Del Vecchio 19/07/2017. Warwick Manufacturing Group University of Warwick

Neural Networks. Theory And Practice. Marco Del Vecchio 19/07/2017. Warwick Manufacturing Group University of Warwick Neural Networks Theory And Practice Marco Del Vecchio marco@delvecchiomarco.com Warwick Manufacturing Group University of Warwick 19/07/2017 Outline I 1 Introduction 2 Linear Regression Models 3 Linear

More information

Clustering in Ratemaking: Applications in Territories Clustering

Clustering in Ratemaking: Applications in Territories Clustering Clustering in Ratemaking: Applications in Territories Clustering Ji Yao, PhD FIA ASTIN 13th-16th July 2008 INTRODUCTION Structure of talk Quickly introduce clustering and its application in insurance ratemaking

More information

STEREO-DISPARITY ESTIMATION USING A SUPERVISED NEURAL NETWORK

STEREO-DISPARITY ESTIMATION USING A SUPERVISED NEURAL NETWORK 2004 IEEE Workshop on Machine Learning for Signal Processing STEREO-DISPARITY ESTIMATION USING A SUPERVISED NEURAL NETWORK Y. V. Venkatesh, B. S. Venhtesh and A. Jaya Kumar Department of Electrical Engineering

More information

Fuzzy Segmentation. Chapter Introduction. 4.2 Unsupervised Clustering.

Fuzzy Segmentation. Chapter Introduction. 4.2 Unsupervised Clustering. Chapter 4 Fuzzy Segmentation 4. Introduction. The segmentation of objects whose color-composition is not common represents a difficult task, due to the illumination and the appropriate threshold selection

More information

Unsupervised Learning

Unsupervised Learning Unsupervised Learning Unsupervised learning Until now, we have assumed our training samples are labeled by their category membership. Methods that use labeled samples are said to be supervised. However,

More information

THREE PHASE FAULT DIAGNOSIS BASED ON RBF NEURAL NETWORK OPTIMIZED BY PSO ALGORITHM

THREE PHASE FAULT DIAGNOSIS BASED ON RBF NEURAL NETWORK OPTIMIZED BY PSO ALGORITHM THREE PHASE FAULT DIAGNOSIS BASED ON RBF NEURAL NETWORK OPTIMIZED BY PSO ALGORITHM M. Sivakumar 1 and R. M. S. Parvathi 2 1 Anna University, Tamilnadu, India 2 Sengunthar College of Engineering, Tamilnadu,

More information

Machine Learning 13. week

Machine Learning 13. week Machine Learning 13. week Deep Learning Convolutional Neural Network Recurrent Neural Network 1 Why Deep Learning is so Popular? 1. Increase in the amount of data Thanks to the Internet, huge amount of

More information