• Journal of the Institute of Electronics Engineers of Korea TE
• /
• v.39 no.2
• /
• pp.66-75
• /
• 2002

This paper presents an adaptive back propagation algorithm that its able to enhancement for the learning efficiency with updating the learning parameter and varies the number of hidden layer node by the generated error, adaptively. This algorithm is expected to escaping from the local minimum and make the best environment for the convergence of the back propagation neural network. On the simulation tested this algorithm on three learning pattern. One was exclusive-OR learning and the another was 3-parity problem and 7${\times}$5 dot alphabetic font learning. In result that the probability of becoming trapped in local minimum was reduce. Furthermore, the neural network enhanced to learning efficient about 17.6%~64.7% for the existed back propagation. 

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.30 no.2C
• /
• pp.31-40
• /
• 2005

we propose an enhanced self-generation supervised algorithm that by combining an ART algorithm and the delta-bar-delta method. Form the input layer to the hidden layer, ART-1 and ART-2 are used to produce nodes, respectively. A winner-take-all method is adopted to the connection weight adaption so that a stored pattern for some pattern is updated. we test the recognition of student identification, a certificate of residence, and an identifier from container that require nodes of hidden layers in neural network. In simulation results, the proposed self-generation supervised learning algorithm reduces the possibility of local minima and improves learning speed and paralysis than conventional neural networks.

• Journal of KIISE:Software and Applications
• /
• v.31 no.5
• /
• pp.537-542
• /
• 2004

Multi-Layer Perceptron network has been mainly applied to many practical problems because of its nonlinear mapping ability. However the generalization ability of MLP networks may be affected by the number of hidden nodes, the initial values of weights and the training errors. These factors, if improperly chosen, may result in poor generalization ability of MLP networks. It is important to identify these factors and their interaction in order to control effectively the generalization ability of MLP networks. In this paper, we have empirically identified the factors that affect the generalization ability of MLP networks, and compared their relative effects on the generalization performance for the conventional and visualized weight selecting methods using the controller box.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2018.05a
• /
• pp.335-337
• /
• 2018

음성향상기법은 음성에 포함된 잡음이나 잔향을 제거하는 기술로써 마이크로폰으로 입력된 음성신호는 잡음이나 잔향에 의해 왜곡되어지므로 음성인식, 음성통신 등의 음성신호처리 기술의 핵심 기술이다. 이전에는 음성신호와 잡음신호 사이의 통계적 정보를 이용하는 통계모델 기반의 음성향상기법이 주로 사용되었으나 통계 모델 기반의 음성향상기술은 정상 잡음 환경과는 달리 비정상 잡음 환경에서 성능이 크게 저하되는 문제점을 가지고 있었다. 최근 머신러닝 기법인 심화신경망 (DNN, deep neural network)이 도입되어 음성 향상 기법에서 우수한 성능을 내고 있다. 심화신경망을 이용한 음성 향상 기법은 다수의 은닉 층과 은닉 노드들을 통하여 잡음이 존재하는 음성 신호와 잡음이 존재하지 않는 깨끗한 음성 신호 사이의 비선형적인 관계를 잘 모델링하였다. 이러한 심화신경망 기반의 음성향상기법을 향상 시킬 수 있는 방법 중 하나인 강화학습을 적용하여 기존 심화신경망 대비 성능을 향상시켰다. 강화학습이란 대표적으로 구글의 알파고에 적용된 기술로써 특정 state에서 최고의 reward를 받기 위해 어떠한 policy를 통한 action을 취해서 다음 state로 나아갈지를 매우 많은 경우에 대해 학습을 통해 최적의 action을 선택할 수 있도록 학습하는 방법을 말한다. 본 논문에서는 composite measure를 기반으로 reward를 설계하여 기존 PESQ (Perceptual Evaluation of Speech Quality) 기반의 reward를 설계한 기술 대비 음성인식 성능을 높였다.

• Journal of the Korea Concrete Institute
• /
• v.17 no.3 s.87
• /
• pp.369-374
• /
• 2005

This study proposes a technique for the recognition of crack patterns, which includes horizontal, vertical, diagonal($-45^{\circ}$), diagonal($+45^{\circ}$), and random cracks, based on image processing technique and artificial neural network. A MATLAB code was developed for the proposed image processing algorithm and artificial neural network. Features were determined using total projection technique, and the structure(no. of layers and hidden neurons) and weight of artificial neural network were determined by learning from artificial crack images. In this process, we adopted Bayesian regularization technique as a generalization method to eliminate overfitting Problem. Numerical tests were performed on thirty-eight crack images to examine validity of the algorithm. Within the limited tests in the present study, the proposed algorithm was revealed as accurately recognizing the crack patterns when compared to those classified by a human expert.

• The Journal of Society for e-Business Studies
• /
• v.16 no.4
• /
• pp.75-85
• /
• 2011

Although progress in information technology has made our life prosperous. But it accompanied a number of adverse effects in various aspects. Especially, internet according to the increasing requirements for privacy and security, IP concealment network technologies to ensure the anonymity are constantly being developed. IP concealment network technologies is aiding the user to bypass the blocked sites can be used to access for information gathering, and they could be used for a malicious hacker to hide his attacks. However, due to complex routing path, local communication bandwidth sangyiham, and internode encryption there are also disadvantages that communication speed is significantly less. In this paper, the research for improving the performance of anonymous networks is to proceed by the communication speed measurement that using GeoIP the particular country with high-bandwidth is Specified or path length is limited.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• v.9 no.2
• /
• pp.393-398
• /
• 2005

본 논문에서는 ART2 기반 자가 생성 지도 학습 알고리즘을 이용한 운송 컨테이너 식별자 인식 시스템을 제안한다. 일반적으로 운송 컨테이너의 식별자들은 글자의 색이 검정색 또는 흰색으로 이루어져 있는 특징이 있다. 이러한 특성을 고려하여 원 컨테이너 영상에 대해 검은색과 흰색을 제외한 모든 부분을 잡음으로 처리하기 위해 퍼지를 이용한 잡은 판단 방법을 적용하여 식별자 영역과 잡음을 구별한다. 식별자 영역을 제외한 잡음 영역을 전체 영상의 평균 픽셀값으로 대체시킨다. 그리고 Sobel 마스크를 이용하여 에지를 검출하고, 추출된 에지를 이용하여 수직 블록과 수평 블록을 검출하여 컨테이너의 식별자 영역을 추출하고 이진화한다. 이진화된 식별자 영역에 대해 검정색의 빈도수를 이용하여 흰바탕과 민바탕을 구분하고 8방향 윤곽선 추적 알고리즘을 적용하여 개별 식별자를 추출한다. 개별 식별자 인식을 위해 ART2 기반 자가 생성 지도 학습 알고리즘은 입력층과 은닉층 사이에 ART2를 적용하여 은닉층의 노드를 생성하고, 은닉층과 출력층 사이에 일반화된 델타 학습 방법과 Delta-bar-Delta 알고리즘을 적용하여 학습 성능을 개선한다. 실제 컨테이너 영상을 대상으로 실험한 결과, 기존의 식별자 추출 방법보다 제안된 식별자 추출 방법이 개선되었다. 그리고 기존의 식별자 인식 알고리즘보다 제안된 ART2 기반 자가 생성 지도 학습 알고리즘이 식별자의 학습 및 인식에 있어서 우수한 성능이 있음을 확인하였다.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.23 no.4
• /
• pp.695-707
• /
• 2013

DHT(Distributed Hash Table) networks such as Kademlia are vulnerable to the ID mapping attack caused by the voluntary DHT mapping structure where the location of a node is solely determined by itself on the network topology. This causes security problems such as eclipse, DRDoS and botnet C&C on DHT networks. To prevent ID mapping attacks, we propose a non-voluntary DHT mapping scheme and perform analysis on NAT compatibility, attack resistance, and network dynamicity. Analysis results show that our approach may have an equivalent level of attack resistance comparing with other defense mechanisms and overcome their limitations including NAT compatibility and network dynamicity.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• v.9 no.2
• /
• pp.677-680
• /
• 2005

In numerical analysis on the thermal performance of the heat exchanger with phase change fluids, the numerical values of thermodynamic properties are needed. But the steam table should be modeled properly as the direct use of thermodynamic properties of the steam table is impossible. In this study the function approximation characteristics of neural networks was used in modeling the saturated vapor region of refrigerant R12. The neural network consists of one input layer with one node, two hidden layers with 10 and 20 nodes each and one output layer with 7 nodes. Input can be both saturation temperature and saturation pressure and two cases were examined. The proposed model gives percentage error of ${\pm}$0.005% for enthalpy and entropy, ${\pm}$0.02% for specific volume and ${\pm}$0.02% for saturation pressure and saturation temperature except several points. From this results neural network could be a powerful method in function approximation of saturated vapor region of R12.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.10 no.11
• /
• pp.2008-2013
• /
• 2006

In numerical analysis on evaluating the thermal performance of the thermal equipment, numerical values of thermodynamic properties such as temperature, pressure, specific volume, enthalpy and entropy are required. But the steam table itself cannot be used without modelling. In this study applicability of neural networks in modelling the wide temperature range of wet saturated vapor region was examined. the multi-layer neural network consists of a input layer with 1 node, two hidden layers with 10 and 20 nodes respectively and a output layer with 6 nodes. Quadratic and cubic spline interpoations methods were also applied for comparison. Neural network model revealed similar percentage error to spline interpolation. From these results, it is confirmed that the neural networks could be powerful method in modelling the wide range of the steam table.