• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.41 no.2
• /
• pp.77-86
• /
• 2004

In general, small-sized neural networks, even though they show good generalization performance, tend to fail to team the training data within a given error bound, whereas large-sized ones learn the training data easily but yield poor generalization. Therefore, a way of achieving good generalization is to find the smallest network that can learn the data, called the optimal-sized neural network. This paper proposes a new scheme for network pruning with ‘impact factor’ which is defined as a multiplication of the variance of a neuron output and the square of its outgoing weight. Simulation results of function approximation problems show that the proposed method is effective in regression.

• Journal of Intelligence and Information Systems
• /
• v.5 no.1
• /
• pp.95-101
• /
• 1999

In this paper, a comparison between different methods to combine predictions from neural networks will be given. These methods are bagging, bumping, and balancing. Those are based on the analysis of the ensemble generalization error into an ambiguity term and a term incorporating generalization performances of individual networks. Neural Networks and AI machine learning models are prone to overfitting. A strategy to prevent a neural network from overfitting, is to stop training in early stage of the learning process. The complete data set is spilt up into a training set and a validation set. Training is stopped when the error on the validation set starts increasing. The stability of the networks is highly dependent on the division in training and validation set, and also on the random initial weights and the chosen minimization procedure. This causes early stopped networks to be rather unstable: a small change in the data or different initial conditions can produce large changes in the prediction. Therefore, it is advisable to apply the same procedure several times starting from different initial weights. This technique is often referred to as training ensembles of neural networks. In this paper, we presented a comparison of three statistical methods to prevent overfitting of neural network.

• Structural Engineering and Mechanics
• /
• v.78 no.1
• /
• pp.53-66
• /
• 2021

The determination of the damage index to reveal the performance level of a structure can constitute the seismic risk generalization approach based on the parametric analysis. This study implemented this concept to one kind of civil engineering structure that is the concrete gravity dam. Different cases of the structure exhibit their individual responses, which constitute different considerations. Therefore, this approach allows the parametric study of concrete as well as soil for evaluating the seismic nature in the generalized case. To ensure that the target algorithm applicable to most of the concrete gravity dams, a very simple procedure has been considered. In order to develop a correlated algorithm (by response surface methodology; RSM) between the ground motion and the structural property, randomized sampling was adopted through a stochastic method called half-fractional central composite design. The responses in the case of fluid-foundation-dam interaction (FFDI) make it more reliable by introducing the foundation as being bounded by infinite elements. To evaluate the seismic generalization of FFDI models, incremental dynamic analysis (IDA) was carried out under the impacts of various earthquake records, which have been selected from the Pacific Earthquake Engineering Research Center data. Here, the displacement-based damage indexed fragility curves have been generated to show the variation in the seismic pattern of the dam. The responses to the sensitivity analysis of the various parameters presented here are the most effective controlling factors for the concrete gravity dam. Finally, to establish the accuracy of the proposed approach, reliable verification was adopted in this study.

• Journal of Information Technology Services
• /
• v.22 no.5
• /
• pp.71-86
• /
• 2023

Recently, various geographic information systems have been used based on spatial information of geographic information systems. Accordingly, it is essential to produce a large-scale map as a small-scale map for various uses of spatial information. However, maps currently being produced have inconsistencies between data due to production timing and limitations in expression, and productivity efficiency is greatly reduced due to errors in products or overlapping processes. In order to improve this, various efforts are being made, such as publishing research and reports for automating domestic mapping, but because there is no specific result, it relies on editors to make maps. This is mainly done by hand, so the time required for mapping is excessive, and quality control for each producer is different. In order to solve these problems, technology that can be automatically produced through computer programs is needed. Research has been conducted to apply the rule base to geometric generalization. The algorithm tolerance setting applied to rule-based modeling is a factor that greatly affects the result, and the level of the result changes accordingly. In this paper, we tried to study the effectiveness of mapping according to tolerance setting. To this end, the utility was verified by comparing it with a manually produced map. In addition, the original data and reduction rate were analyzed by applying generalization algorithms and tolerance values. Although there are some differences by region, it was confirmed that the complexity decreased on average. Through this, it is expected to contribute to the use of spatial information-based services by improving tolerances suitable for small-scale mapping regulations in order to secure spatial information data that guarantees consistency and accuracy.

• 제어로봇시스템학회:학술대회논문집
• /
• 2001.10a
• /
• pp.102.4-102
• /
• 2001

An evolutionary structure optimization method for the Gaussian radial basis function (RBF) network is presented, for modelling and predicting nonlinear time series. Generalization performance is significantly improved with a much smaller network, compared with that of the usual clustering and least square learning method.

• Honam Mathematical Journal
• /
• v.32 no.3
• /
• pp.481-491
• /
• 2010

In this paper, we investigate a generalization of the Adams-Bashforth method by using the Taylor's series. In case of m-step method, the local truncation error can be expressed in terms of m - 1 coefficients. With an appropriate choice of coefficients, the proposed method has produced much smaller error than the original Adams-Bashforth method. As an application of the generalized Adams-Bashforth method, the accuracy performance is demonstrated in the satellite orbit prediction problem. This implies that the generalized Adams-Bashforth method is applied to the orbit prediction of a low-altitude satellite. This numerical example shows that the prediction of the satellite trajectories is improved one order of magnitude.

• 제어로봇시스템학회:학술대회논문집
• /
• 1997.10a
• /
• pp.1557-1560
• /
• 1997

In this paper, we have built an experimental magnetic levitation system for a possible use of control education. We have give a mathermatical model of the nonlinear system and have shown the stability region of the linearized system when it is controlled by a PD controller. We also proposed a neural network control system which uses a neural network as a feedforward controller thgether with a conventional feedback PF controller. We have generated a desired output trajectory, which was designed for the benefit of the generalization of the neural network controller, and trained the desired output trajectory, which was desigend for the benefit of the generalization of the neural netowrk controller, and trained a neural network controller with the data of the actual input and the output of the system obtained by applying the desired output trajectroy. A good tracking performance was observed for both the desired trajectiories used and not used for the neural network training.

• Smart Structures and Systems
• /
• v.15 no.4
• /
• pp.1103-1120
• /
• 2015

A self-sensing magnetorheological (MR) damper with embedded piezoelectric force sensor has recently been devised to facilitate real-time close-looped control of structural vibration in a simple and reliable manner. The development and characterization of the self-sensing MR damper are presented based on experimental work, which demonstrates its reliable force sensing and controllable damping capabilities. With the use of experimental data acquired under harmonic loading, a nonparametric dynamic model is formulated to portray the nonlinear behaviors of the self-sensing MR damper based on NARX modeling and neural network techniques. The Bayesian regularization is adopted in the network training procedure to eschew overfitting problem and enhance generalization. Verification results indicate that the developed NARX network model accurately describes the forward dynamics of the self-sensing MR damper and has superior prediction performance and generalization capability over a Bouc-Wen parametric model.

• Proceedings of the KSPS conference
• /
• 2007.05a
• /
• pp.270-273
• /
• 2007

The purpose of this study was to evaluate the effects of multisensory(AVK: Auditory, Visual and Kinethetic) treatment on reading pronunciation with phonological prcessing - tensification, palatalization, and lateralization for the middle school students with delayed language development caused by mental retarded. Participants were three children with reading pronunciation difficulties in phonological processing. The following conclusions were arrived. First, three children are improved on tensifiication, palatalization, and lateralization by multisensory treatment program. Second, multisensory treatment was effective in facilitating generalization. Three children presented prominent generalization effcects in lateralization. Third, they were found to maintain partially their performance rates of the later phase of the reading with phonological processing intervention three weeks after the termination of the intervention.

• 제어로봇시스템학회:학술대회논문집
• /
• 1999.10a
• /
• pp.116-120
• /
• 1999

The step response of the Manabe standard form[1］has little overshoot and shows almost same waveforms regardless of the order of the characteristic polynomials. In some situations it is difficult to control the rise time and settling time simultaneously of the step response of the Manabe standard form. To control its rise time and settling time efficiently, we develop the generalization of the Manabe standard form: we try to find out the SRFS(Slow Rise time & Fast Settling time) form which has the slower rise time and faster settling time than those of the Manabe standard form. we also consider the other three forms: FRSS(Fast Rise time & Slow Settling time), SRFS(Slow Rise time ＆ Fast Settling time) and SRSS(Slow Rise time & Slow Settling time) forms. In this paper, by using the genetic algorithm, we obtain all the coefficient of the four forms we mention above. Finally, we design a controller for a given plant so that the overall system has the performance that the rise time is faster, the settling time is slower than those of the Manabe standard form.