• Proceedings of the Korea Inteligent Information System Society Conference
• /
• 2003.05a
• /
• pp.469-473
• /
• 2003

다층 구조 신경망에서 널리 사용되는 오류 역전파 알고리즘은 초기 가중치와 불충분한 은닉층의 노드수로 인하여 지역 최소화에 빠질 가능성이 있다. 따라서 본 논문에서는 오류 역전파 알고리즘에서 은닉층의 노드 수를 설정하는 문제와 ARTI에서 경계 변수의 설정에 따라 인식률이 저하되는 문제점을 개선하기 위하여 ARTI과 Max-Min 신경망을 결합한 퍼지 다층 퍼셉트론을 제안한다. 제안된 자가 생성을 이용한 퍼지 다층 퍼셉트론은 입력층에서 은닉층으로 노드를 생성시키는 방식은 ARTI을 적용하였고, 가중치 조정은 특정 패턴에 대한 저장 패턴을 수정하도록 하는 winner-take-all 방식을 적용하였다. 제안된 학습 방법의 성능을 평가하기 위하여 학생증 영상을 대상으로 실험한 결과, 기존의 오류 역전파 알고즘보다 연결 가중치들이 지역 최소화에 위치할 가능성이 줄었고 학습 속도 및 정체 현상이 개선되었다.

• Proceedings of the Korean Information Science Society Conference
• /
• 2001.10b
• /
• pp.313-315
• /
• 2001

본 논문에서 소개하는 알고리즘을 은닉층의 뉴런의 수를 학습하는 동안 동적으로 변화시켜 역전파 알고리즘의 단점인 Local maxima를 탈출하고 또한 은닉층의 뉴런의 수를 결정하는 과정을 없애기 위해 연구되었다. 본 알고리즘의 성능을 평가하기 위해 두 가지 실험에 적용하였는데 첫번째는 Exclusive-OR 문제이고 두번째는 7$\times$8 한글 자음과 모음의 폰트 학습에 적용하였다. 이 실험의 결과로 네트웍이 local maxima에 빠져드는 확률이 줄어드는 것을 알 수 있었고 학습속도 또한 일반적인 역전파 알고리즘보다 빠른 것으로 증명되었다.

• Proceedings of the Korean Information Science Society Conference
• /
• 2004.10a
• /
• pp.199-201
• /
• 2004

본 논문에서는 반도체의 Wafer ID 문자인식을 위해 기존의 오류 역전파 학습알고리즘을 개선하여 최적의 학습 조건에 관해 연구하였다. 결과, 오류 역전파 학습알고리즘의 학습 최적 조건은 은닉 층수는 1층, n값은 0.6 이상, 은닉층 노드수는 10개일 때 99%의 높은 인식률을 보였다. 본 논문에서 제안하는 최적조건을 사용함으로써 기존의 오류역전파 학습 알고리즘이 가진 문제점을 해결할 수 있었다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.11a
• /
• pp.469-470
• /
• 2023

이 논문은 이미지 처리와 머신 러닝 분야에서 중요한 주제 중 하나인 알고리즘 성능과 효율성에 대한 연구이다. 역전파 알고리즘과 상대적으로 새로운 Forward-Forward 알고리즘을 비교하고 분석해 보니 Forward-Forward 알고리즘이 역전파 알고리즘보다 비교적 높은 정확도를 확인할 수 있었다.

• The Journal of the Acoustical Society of Korea
• /
• v.33 no.6
• /
• pp.392-399
• /
• 2014

The underwater acoustic communication is characterized by doubly spread channels, which are the delay spread due to multiple paths and the doppler spread due to environmental fluctuations or a moving platform. An equalizer is used to remove the inter-symbol interferences that the delay spread causes, but an equalizer doesn't use an acoustic environment such as a multipath. However, a passive time-reversal mirror is simpler than an equalizer because a matched filter is implemented numerically at the receiver structure along with one-way propagation. In this paper, a passive time-reversal mirror is applied to remove interferences due to a multipath in sea-going experimental data in East Sea in Oct. 2010 and improved communication performance is confirmed. The performance is verified by comparing the signal-to-interference plus noise ratio before/after passive time-reversal mirror. It is also performed independently of the passive time-reversal mirror and adaptive equalizer and the bit error rate is compared to verify the performance of underwater acoustic communication.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.15 no.6
• /
• pp.116-126
• /
• 2016

In this study, lane-wise traffic flow characteristics were analysed on uninterrupted flow using a new notion of "lane-wise travel speed reversal (LTSR)" which is defined as a phenomena that travel speed in the median lane is lower than other lanes. Mathematical formulation was also proposed to calculate the strength of LTSR. The experiment road site is Seoul Outer Ring Expressway (Jayuro-IC~Jangsoo-IC), and travel trajectories for each four lane were collected for weekdays (Mon. through Fri.) during morning peak. Comparing lane-wise travel speeds for entire test road section, no LTSR was observed, meaning that the travel speed in the median lane is the fastest, followed by 2nd, 3rd, and 4th lane as in order. Howerver, the result of microscopic analysis using 100-meter discrete road section based data shows that LTSR occurs many times. Especially the strength of LTSR is higher in congestion area and freeway merge and diverge segment. It is expected that these results could be used as a fundamental data when establishing lane-by-lane traffic operation strategy and developing lane-wise traffic information collection and dissemination technology.

• Journal of KIISE:Software and Applications
• /
• v.33 no.12
• /
• pp.999-1007
• /
• 2006

The Backpropagation learning algorithm is known to have slow and false convergence aroused from plateau and local minima. Many substitutes for backpropagation announced so far appear to pay some trade-off for convergence speed and stability of convergence according to parameters. Here, a new algorithm is proposed, which avoids some of those problems associated with the conventional backpropagation problems, especially with local minima, and gives relatively stable and fast convergence with low storage requirement. This is the separate learning algorithm in which the upper connections, hidden-to-output, and the lower connections, input-to-hidden, separately trained. This algorithm requires less computational work than the conventional backpropagation and other improved algorithms. It is shown in various classification problems to be relatively reliable on the overall performance.

• The Transactions of the Korea Information Processing Society
• /
• v.4 no.7
• /
• pp.1759-1769
• /
• 1997

The Error BackPropagation (EBP) algorithm for multi-layered neural networks is widely used in various areas such as associative memory, speech recognition, pattern recognition and robotics, etc. Nevertheless, many researchers have continuously published papers about improvements over the original EBP algorithm. The main reason for this research activity is that EBP is exceeding slow when the number of neurons and the size of training set is large. In this study, we developed new learning speed acceleration methods using variable learning rate, variable momentum rate and variable slope for the sigmoid function. During the learning process, these parameters should be adjusted continuously according to the total error of network, and it has been shown that these methods significantly reduced learning time over the original EBP. In order to show the efficiency of the proposed methods, first we have used binary data which are made by random number generator and showed the vast improvements in terms of epoch. Also, we have applied our methods to the binary-valued Monk's data, 4, 5, 6, 7-bit parity checker and real-valued Iris data which are famous benchmark training sets for machine learning.

• Journal of Korea Society of Industrial Information Systems
• /
• v.5 no.3
• /
• pp.117-124
• /
• 2000

A systolic array is a parallel processing system which consists of processing elements of basic computation capabilities, connected with regular and local communication lines. It has been known that a systolic array is on of effective systems to solve complicated communication problems occurred between densely connected neurons on ANN(Artificial Neural Network). In this paper, a systolic array simulator for the back-propagation ANN, which automatically constructs the proper systolic array for a given number of neurons of the ANN, is designed and constructed. With animation techniques of the simulators, it is easy for users to be able to examine the execution of the back-propagation algorithm on the designed systolic array step by step. Moreover the simulator can perform forward and backward operations of the back-propagation algorithm either in sequence or in parallel on the designed systolic array. Parallel execution can be performed by feeding continuous input patterns and by executing bidirectional propagations on all of processing elements of a systolic array at the same time.

• Proceedings of the Korean Information Science Society Conference
• /
• 2001.04b
• /
• pp.295-297
• /
• 2001

최근 생물 유전자 정보를 효과적으로 분석하기 위한 적절한 도구의 필요성이 대두되고 있다. 본 논문에서는 백혈병 환자의 골수로부터 얻어낸 DNA Microarray 유전 정보를 분류하여 환자가 가지고 있는 암의 종류를 예측하기 위한 최적의 특징추출방법과 분류 방법을 찾고자 한다. 이를 위해 피어슨 상관관계, 유클리디안 거리, 코사인 계수, 스피어맨 상관관계, 정보 이득, 상호 정보, 신호 대잡음비의 7가지 특징 추출 방법을 사용하였으며, 역전과 신경망, 의사결정 트리, 구조 적응형 자기구성 지도, $textsc{k}$-최근접 이웃 등 가지의 기계학습 분류기를 이용하여 분류 실험을 하였다. 실험결과, 피어슨 상관관계와 역전파 신경망을 이용한 분류 방법이 97.1%의 인식률을 보임을 알 수 있었다.