• Proceedings of the IEEK Conference
• /
• 2000.07b
• /
• pp.1041-1044
• /
• 2000

The enhancement to the back-propagation algorithm presented in this paper has resulted from the need to extract sparsely connected networks from networks employing product terms. The enhancement works in conjunction with the back-propagation weight update process, so that the actions of weight zeroing and weight stimulation enhance each other. It is shown that the error measure, can also be interpreted as rate of weight change (as opposed to ${\Delta}W_{ij}$), and consequently used to determine when weights have reached a stable state. Weights judged to be stable are then compared to a zero weight threshold. Should they fall below this threshold, then the weight in question is zeroed. Simulation of such a system is shown to return improved learning rates and reduce network connection requirements, with respect to the optimal network solution, trained using the normal back-propagation algorithm for Multi-Layer Perceptron (MLP), Higher Order Neural Network (HONN) and Sigma-Pi networks.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.44 no.1
• /
• pp.40-48
• /
• 2007

This paper presents a novel texture segmentation method using multilayer perceptron (MLP) networks and Markov random fields in multiscale Bayesian framework. Multiscale wavelet coefficients are used as input for the neural networks. The output of the neural network is modeled as a posterior probability. Texture classification at each scale is performed by the posterior probabilities from MLP networks and MAP (maximum a posterior) classification. Then, in order to obtain the more improved segmentation result at the finest scale, our proposed method fuses the multiscale MAP classifications sequentially from coarse to fine scales. This process is done by computing the MAP classification given the classification at one scale and a priori knowledge regarding contextual information which is extracted from the adjacent coarser scale classification. In this fusion process, the MRF (Markov random field) prior distribution and Gibbs sampler are used, where the MRF model serves as the smoothness constraint and the Gibbs sampler acts as the MAP classifier. The proposed segmentation method shows better performance than texture segmentation using the HMT (Hidden Markov trees) model and HMTseg.

• International Journal of Control, Automation, and Systems
• /
• v.2 no.4
• /
• pp.423-434
• /
• 2004

In this paper, we introduce a new topology of Self-Organizing Polynomial Neural Networks (SOPNN) based on genetically optimized Multi-Layer Perceptron (MLP) and discuss its comprehensive design methodology involving mechanisms of genetic optimization. Let us recall that the design of the 'conventional' SOPNN uses the extended Group Method of Data Handling (GMDH) technique to exploit polynomials as well as to consider a fixed number of input nodes at polynomial neurons (or nodes) located in each layer. However, this design process does not guarantee that the conventional SOPNN generated through learning results in optimal network architecture. The design procedure applied in the construction of each layer of the SOPNN deals with its structural optimization involving the selection of preferred nodes (or PNs) with specific local characteristics (such as the number of input variables, the order of the polynomials, and input variables) and addresses specific aspects of parametric optimization. An aggregate performance index with a weighting factor is proposed in order to achieve a sound balance between the approximation and generalization (predictive) abilities of the model. To evaluate the performance of the GA-based SOPNN, the model is experimented using pH neutralization process data as well as sewage treatment process data. A comparative analysis indicates that the proposed SOPNN is the model having higher accuracy as well as more superb predictive capability than other intelligent models presented previously.reviously.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.22 no.6
• /
• pp.173-178
• /
• 2022

Recently, a recommendation system using deep learning in social network services has been actively studied. However, in the case of a recommendation system using deep learning, the cold start problem and the increased learning time due to the complex computation exist as the disadvantage. In this paper, the user-tailored exercise routine recommendation algorithm is proposed using the user's metadata. Metadata (the user's height, weight, sex, etc.) set as the input of the model is applied to the designed model in the proposed algorithms. The exercise recommendation system model proposed in this paper is designed based on the neural collaborative filtering (NCF) algorithm using multi-layer perceptron and matrix factorization algorithm. The learning proceeds with proposed model by receiving user metadata and exercise information. The model where learning is completed provides recommendation score to the user when a specific exercise is set as the input of the model. As a result of the experiment, the proposed exercise recommendation system model showed 10% improvement in recommended performance and 50% reduction in learning time compared to the existing NCF model.

• Journal of the Society of Naval Architects of Korea
• /
• v.52 no.1
• /
• pp.88-95
• /
• 2015

For the frequency-domain spectral fatigue analysis, the probability density function of stress range needs to be estimated based on the stress spectrum only, which is a frequency domain representation of the response. The probability distribution of the stress range of the narrow-band spectrum is known to follow the Rayleigh distribution, however the PDF of wide-band spectrum is difficult to define with clarity due to the complicated fluctuation pattern of spectrum. In this paper, efforts have been made to figure out the links between the probability density function of stress range to the structural response of wide-band Gaussian random process. An artificial neural network scheme, known as one of the most powerful system identification methods, was used to identify the multivariate functional relationship between the idealized wide-band spectrums and resulting probability density functions. To achieve this, the spectrums were idealized as a superposition of two triangles with arbitrary location, height and width, targeting to comprise wide-band spectrum, and the probability density functions were represented by the linear combination of equally spaced Gaussian basis functions. To train the network under supervision, varieties of different wide-band spectrums were assumed and the converged probability density function of the stress range was derived using the rainflow counting method and all these data sets were fed into the three layer perceptron model. This nonlinear least square problem was solved using Levenberg-Marquardt algorithm with regularization term included. It was proven that the network trained using the given data set could reproduce the probability density function of arbitrary wide-band spectrum of two triangles with great success.

• Proceedings of the Korea Database Society Conference
• /
• 1999.06a
• /
• pp.261-270
• /
• 1999

Sintering transforms powdered ore into lumped ore so that the latter can be used in a blast furnace. The powdered ore combined with coke and other materials is loaded into a container and moved along by a pallet while the ignited coke bums. The speed by which the pallet moves determines how much sintering takes place. Since the process is complicated and lacks an accurate mathematical model, human operators manually control the speed by monitoring various factors in the plant. In this paper, we propose a neural network-based pallet speed controller which copies human operator knowledge. Actual process data were collected from a sintering plant fer eight months and preprocessed to remove noisy and inconsistent data. A multilayer perceptron was trained using a back-propagation learning algorithm. In on-line testing at the sinter plant, the proposed model reliably controlled pallet speed during normal operation without the help of human operators. Moreover, the duality and productivity was as good as with human operators.

• Proceedings of the Korea Inteligent Information System Society Conference
• /
• 1999.03a
• /
• pp.261-270
• /
• 1999

Sintering transforms powdered ore into lumped ore so that the latter can be used in a blast furnace. The powdered or combined with coke and other materials is loaded into a container and moved along by a pallet while the ignited coke burns. The speed by which the pallet moves determines how much sintering takes place. Since the process is complicated and lacks an accurate mathematical model, human operators manually control the speed by monitoring various factors in the plant. In this paper, we propose a neural network-based pallet speed controller which copies human operator knowledge. Actual process data were collected from a sintering plant for eight months and preprocessed to remove noisy and inconsistent data. A multilayer perceptron was trained using a back-propagation learning algorithm. In on-line testing at the sinter plant, the proposed model reliably controlled pallet speed during normal operation without the help of human operators. Moreover, the quality and productivity was as good as with human operators.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.16 no.1
• /
• pp.13-22
• /
• 1991

In this paper we designed 2 bit ECC(Error Checking and Correction) circuit using Single Layer Perceptron type neural networks. We used (11, 6) block codes having 6 data bits and 8 check bits with appling cyclic hamming codes. All of the circuits are layouted by CMOs 2um double metal design rules. In the result of circuit simulation, 2 bit ECC circuit operates at 67MHz of input frequency.

• 제어로봇시스템학회:학술대회논문집
• /
• 2001.10a
• /
• pp.172.6-172
• /
• 2001

In recent years, many attempts have been made to predict the behavior of bonds, currencies, stock, or other economic markets. Most previous experiments used multilayer perceptrons(MLP) for stock market forecasting, The Kospi 200 Index is modeled using different neural networks and fuzzy system predictions. In this paper, a multilayer perceptron architecture, a dynamic polynomial neural network(DPNN) and a fuzzy system are used to predict the Kospi 200 index. The results of prediction is compared with the root mean squared error(RMSE) and the scatter plot. The results show that the fuzzy system is performing slightly better than DPNN and MLP. We can develop the desired fuzzy system by learning methods ...

• Journal of Electrical Engineering and Technology
• /
• v.5 no.4
• /
• pp.666-671
• /
• 2010

Multi-layer perceptron neural networks have been constructed to classify four types of defects in steam generator tubes. Three features are extracted from the signals of the eddy current testing method. These include maximum impedance, phase angle at the point of maximum impedance, and an angle between the point of maximum impedance and the point of half the maximum impedance. Two hundred sets of these features are used for training and assessing the networks. Two approaches are involved to train the networks and to classify the defect type. One is the conjugate gradient method and the other is the Broydon-Fletcher-Goldfarb-Shanno method which is recognized as the most popular algorithm of quasi-Newton methods. It is found from the computation results that the training time of the Broydon-Fletcher-Goldfarb-Shanno method is much faster than that of the conjugate gradient method in most cases. On the other hand, no significant difference of the classification performance between the two methods is observed.