• The KIPS Transactions:PartB
• /
• v.12B no.3 s.99
• /
• pp.343-348
• /
• 2005

The goal of this paper is to study the joint effect of factors of neural network teaming procedure. There are many factors, which may affect the generalization ability and teaming speed of neural networks, such as the initial values of weights, the learning rates, and the regularization coefficients. We will apply a constructive training algerian for neural network, then patterns are trained incrementally by considering them one by one. First, we will investigate the effect of these factors on generalization performance and learning speed. Based on these factors' effect, we will propose a joint method that simultaneously considers these three factors, and dynamically hue the learning rate and regularization coefficient. Then we will present the results of some experimental comparison among these kinds of methods in several simulated nonlinear data. Finally, we will draw conclusions and make plan for future work.

• Journal of information and communication convergence engineering
• /
• v.9 no.6
• /
• pp.637-640
• /
• 2011

The hybrid neural networks have characteristics such as fast learning times, generality, and simplicity, and are mainly used to classify learning data and to model non-linear systems. The middle layer of a hybrid neural network clusters the learning vectors by grouping homogenous vectors in the same cluster. In the clustering procedure, the homogeneity between learning vectors is represented as the distance between the vectors. Therefore, if the distances between a learning vector and all vectors in a cluster are smaller than a given constant radius, the learning vector is added to the cluster. However, the usage of a constant radius in clustering is the primary source of errors and therefore decreases the recognition success rate. To improve the recognition success rate, we proposed the enhanced hybrid network that organizes the middle layer effectively by using the enhanced ART1 network adjusting the vigilance parameter dynamically according to the similarity between patterns. The results of experiments on a large number of calling card images showed that the proposed algorithm greatly improves the character extraction and recognition compared with conventional recognition algorithms.

• Transactions of the Korean Society of Automotive Engineers
• /
• v.9 no.1
• /
• pp.156-162
• /
• 2001

An effective and practical method is presented for solving the forward kinematics of a 6-DOF Stewart Platform, using neural network algorithm together with Newton-Raphson method. An approximated solution is obtained from trained neural network, then it is used as an initial estimate for Newton-Raphson method. A series of accurate solutions are calculated with reasonable speed for the entire workspace of the platform. The solution procedure can be used for driving a real-time simulation platform.

• Structural Engineering and Mechanics
• /
• v.28 no.2
• /
• pp.153-166
• /
• 2008

This paper suggests the use of wavelet multiresolution analysis (WMRA) and neural network for generation of artificial earthquake accelerograms from target spectrum. This procedure uses the learning capabilities of radial basis function (RBF) neural network to expand the knowledge of the inverse mapping from response spectrum to earthquake accelerogram. In the first step, WMRA is used to decompose earthquake accelerograms to several levels that each level covers a special range of frequencies, and then for every level a RBF neural network is trained to learn to relate the response spectrum to wavelet coefficients. Finally the generated accelerogram using inverse discrete wavelet transform is obtained. An example is presented to demonstrate the effectiveness of the method.

• 제어로봇시스템학회:학술대회논문집
• /
• 2001.10a
• /
• pp.169.2-169
• /
• 2001

In this paper, we propose the face recognition system using the neural network. A difficult procedure in constructing the entire recognition systems is the feature extraction from the face imga. And a key poing is the design of the matching function that relates the set of feature values to the appropriate face candidates. We use the length and angle values as feature values that are extracted from the face image normalized to the range of [0,1]. These features values are applied to the input layer of the neural network. Then, these multi-layered perceptron learns or gives otput result. By using the neural network we need not to design the matching function. This function may have nonlinear attributes considerably and would be ...

• Journal of Advanced Marine Engineering and Technology
• /
• v.23 no.2
• /
• pp.168-175
• /
• 1999

Unlike a general inverted pendulum system which is moved on the cart the proposed inverted pendulum system in this paper has an inverted pendulum which is moved on the two-degree-of-freedom parallelogram link. The dynamic equation of the pendulum system activated by the DD(Direct Drive)motor includes many nonlinear terms and has the high degree of freedoms. The problem is followed hat the exact mathmatical equations can not be analized by a general linear theory However the neural network trained by a simple learning method can control the dynamic system with hard nonlinearities. Learning procedure is the backpropagation algorithm with super-visory signal. The plant inputs obtained by the designed neural network in this paper can stabilize the pendu-lem and get the servo control. Experiment results have proce the effectiveness of the designed neural network controller.

• Proceedings of the Korean Operations and Management Science Society Conference
• /
• 2006.05a
• /
• pp.354-358
• /
• 2006

This study is concerned with the part-machine grouping(PMG) based on the non-binary part-machine incidence matrix in which real manufacturing Factors such as the operation sequences with multiple visits to the same machine and production volumes of parts are incorporated and each entry represents actual moves due to different operation sequences. The proposed approach adopts Fuzzy ART neural network to quickly create the initial part families and their associated machine cells. To enhance the poor solution due to category proliferation inherent to most artificial neural networks, a supplementary procedure reassigning parts and machines is added. To show effectiveness of the proposed approach to large-size PMG problems, a psuedo-replicated clustering procedure is designed and implemented.

• The Transactions of the Korean Institute of Electrical Engineers D
• /
• v.53 no.1
• /
• pp.40-49
• /
• 2004

In this paper, we propose a new architecture of Genetic Algorithms(GAs)-based Self-Organizing Polynomial Neural Networks(SOPNN), discuss a comprehensive design methodology and carry out a series of numeric experiments. The conventional SOPNN is based on the extended Group Method of Data Handling(GMDH) method and utilized the polynomial order (viz. linear, quadratic, and modified quadratic) as well as the number of node inputs fixed (selected in advance by designer) at Polynomial Neurons (or nodes) located in each layer through a growth process of the network. Moreover it does not guarantee that the SOPNN generated through learning has the optimal network architecture. But the proposed GA-based SOPNN enable the architecture to be a structurally more optimized network, and to be much more flexible and preferable neural network than the conventional SOPNN. In order to generate the structurally optimized SOPNN, GA-based design procedure at each stage (layer) of SOPNN leads to the selection of preferred nodes (or PNs) with optimal parameters- such as the number of input variables, input variables, and the order of the polynomial-available within SOPNN. An aggregate performance index with a weighting factor is proposed in order to achieve a sound balance between approximation and generalization (predictive) abilities of the model. A detailed design procedure is discussed in detail. To evaluate the performance of the GA-based SOPNN, the model is experimented with using two time series data (gas furnace and NOx emission process data of gas turbine power plant). A comparative analysis shows that the proposed GA-based SOPNN is model with higher accuracy as well as more superb predictive capability than other intelligent models presented previously.

• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.58 no.9
• /
• pp.1821-1826
• /
• 2009

An adaptive neural controller for perturbed strict-feedback nonlinear system is proposed. All the previous adaptive neural (or fuzzy) controllers are based on the backstepping scheme where the universal approximators are employed in every design steps. These schemes involve virtual controls and their time derivatives that make the stability analysis and implementation of the controller very complex. This fact is called 'explosion of complexty ' since the complexity grows exponentially as the system dynamic order increases. The proposed adaptive neural control scheme adopt the backstepping design procedure only for determining ideal control law and employ only one neural network to approximate the finally selected ideal controller, which makes the controller design procedure and stability analysis considerably simple compared to the previously proposed controllers. It is shown that all the time-varing signals containing tracking error are stable in the Lyapunov viewpoint.

• Proceedings of the Korea Inteligent Information System Society Conference
• /
• 1999.10a
• /
• pp.365-373
• /
• 1999

Recently, numerous studies have demonstrated that artificial intelligence such as neural networks can be an alternative methodology for classification problems to which traditional statistical methods have long been applied. In building neural network model, the selection of independent and dependent variables should be approached with great care and should be treated as a model construction process. Irrespective of the efficiency of a learning procedure in terms of convergence, generalization and stability, the ultimate performance of the estimator will depend on the relevance of the selected input variables and the quality of the data used. Approaches developed in statistical methods such as correlation analysis and stepwise selection method are often very useful. These methods, however, may not be the optimal ones for the development of neural network models. In this paper, we propose a genetic algorithms approach to find an optimal or near optimal input variables for neural network modeling. The proposed approach is demonstrated by applications to bankruptcy prediction modeling. Our experimental results show that this approach increases overall classification accuracy rate significantly.