• Journal of Advanced Navigation Technology
• /
• 제27권5호
• /
• pp.674-684
• /
• 2023

In this paper, a real-time prediction technique for gravity disturbances is proposed using a multi-layer perceptron (MLP) model. To select a suitable MLP model, 4 models with different network sizes were designed to compare the training accuracy and execution time. The MLP models were trained using the data of vehicle moving along the surface of the sea or land, including their positions and gravity disturbance. The gravity disturbances were calculated using the 2160^th degree and order EGM2008 with SHM. Among the models, MLP4 demonstrated the highest training accuracy. After training, the weights and biases of the 4 models were stored in the embedded computer of the INS to implement the MLP network. MLP4 was found to have the shortest execution time among the 4 models. These research results are expected to contribute to improving the navigation accuracy of INS through gravity disturbance compensation in the future.

• Journal of Korea Multimedia Society
• /
• 제23권9호
• /
• pp.1155-1163
• /
• 2020

Deep learning approach is emerging as a new way to improve the accuracy of tree species identification using bark image. However, the approach has not been studied enough because it is confronted with the problem of acquiring a large volume of bark image dataset. This study solved this problem by utilizing a pretrained off-the-shelf DCNN model. It compares the discrimination power of bark features extracted by each DCNN model. Then it extracts the features by using a selected DCNN model and feeds them to a multi-layer perceptron (MLP). We found out that the ResNet50 model is effective in extracting bark features and the MLP could be trained well with the features reduced by the principal component analysis. The proposed approach gives accuracy of 99.1% and 98.4% for BarkTex and Trunk12 datasets respectively.

• Annual Conference of KIPS
• /
• 한국정보처리학회 2006년도 춘계학술발표대회
• /
• pp.325-328
• /
• 2006

역전파 학습 방법은 속도가 느리고, 지역 최소점으로 빠져 수렴에 실패하는 경우가 많다고 알려져 있다. 이제까지 알려진 역전파의 대체 방법들은 수렴 속도와 인자에 따른 수렴의 안정성에 대한 불균형을 해소 하는데 치중했다. 기존의 전통적인 역전파에서 발생하는 위와 같은 문제를 해결하기 위하여, 본 논문에서는 적은 용량의 저장 공간만을 요구하며 수렴이 빠르고 상대적으로 안정성이 보장되는 알고리즘을 제안한다. 이 방법은 상위연결(upper connections), 은닉층-출력층(hidden to output), 하위 연결(lower connections), 입력층-은닉층(input to hidden)에 대해 개별적으로 훈련을 시키는 교대 학습 방법을 적용한다.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 한국방송∙미디어공학회 2019년도 추계학술대회
• /
• pp.67-68
• /
• 2019

머신러닝을 활용한 이미지 분류는 단순 사물을 넘어서 사람의 감성과 같은 추상적이고 주관적인 개념에도 적용되고 있다. 이 중에서도 합성곱 신경망을 통한 이미지의 감정 분류 연구가 더욱 활성화되고 있다. 그럼에도 다양한 멀티미디어들을 머신러닝 알고리즘으로 분석하고 이를 의미있는 결과로 재생성하기는 매우 복잡하고 까다롭다. 본 연구에서는 기존 연구를 개선시켜 음악 데이터를 다층퍼셉트론 모델을 통해 분류된 이미지와 결합한 동영상을 파이썬의 다양한 라이브러리를 통해 자동으로 생성하였다. 이를 통해 특정 분위기로 분류된 이미지들과 이에 어울리는 음악을 매칭시켜 유의미한 새로운 멀티미디어를 자동으로 생성할 수 있었다.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• 제33B권4호
• /
• pp.209-216
• /
• 1996

In this paper, we propose a learning method of multi-layer perceptrons (MLPs) with 8-bit data precision. The suggested method uses the cross-entropy cost function to remove the slope term of error signal in output layer. To decrease the possibility of overflows, we use 16-bit weighted sum results into the 8-bit data with appropriate range. In the forwared propagation, the range for bit-conversion is determined using the saturation property of sigmoid function. In the backwared propagation, the range for bit-conversion is derived using the probability density function of back-propagated signal. In a simulation study to classify hadwritten digits in the CEDAR database, our method shows similar generalization performance to the error back-propagation learning with 16-bit precision.

• Annual Conference on Human and Language Technology
• /
• 한국정보과학회언어공학연구회 1995년도 제7회 한글 및 한국어 정보처리 학술대회
• /
• pp.137-142
• /
• 1995

코호넨의 SOFM(Self-Organizing Feature Map)온 빠른 검증 학습이 가능하여 다층 퍼셉트론의 단점을 보완할 수 있는 패턴분류기로 부각되고 있다. 그러나 기본적으로 고정된 크기와 구조의 네트워크를 사용하기 때문에 실재 문제에 적용하기가 쉽지 않다는 문제가 있다. 본 논문에서는 패턴에 대한 사전 정보없이 복잡한 패턴공간을 적응적으로 분할하기 위해 구조적응되는 자기조직화 신경망을 소개하고 이를 인쇄체 한글 문자의 인식에 적용한 결과를 보여준다. 여기에서 제안하는 신경망은 SOFM의 각 셀이 좀더 자세한 SOFM으로 확장될 수 있도록하며, 확률분포가 0인 셀을 제거함으로써 패턴 공간에 보다 근사한 분류를 가능하게 한다. 실제로 이러한 방식이 한글과 같은 복잡한 분류 문제에서 어떻게 작동하는지 설명하고, 한글 완성형 2350자에 대해 실험한 결과를 보여준다.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 한국퍼지및지능시스템학회 2002년도 춘계학술대회 및 임시총회
• /
• pp.223-226
• /
• 2002

This Paper presents an interval type-2 fuzzy perceptron algorithm that is an extension of the type-1 fuzzy perceptron algorithm proposed in [1]. In our proposed method, the membership values for each Pattern vector are extended as interval type-2 fuzzy memberships by assigning uncertainty to the type-1 memberships. By doing so, the decision boundary obtained by interval type-2 fuzzy memberships can converge to a more desirable location than the boundary obtained by crisp and type-1 fuzzy perceptron methods. Experimental results are given to show the effectiveness of our method

• The Transactions of the Korea Information Processing Society
• /
• 제4권1호
• /
• pp.177-183
• /
• 1997

The cumulative function of the noncentral t distribution calculate power in testing equality of means of two normal populations and confidence intervals for the ratio of population mean to standard deviation. In this paper, the evaluation of the cumulative function of noncentral t distribution is applied to the neural network consists of the multi-layer perception structure and learning process has the algorithm of the backpropagation. Numerical comparisons are made between the Fisher's values and the results obtained by neural network theory.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• 제16권2호
• /
• pp.128-136
• /
• 1991

A neural network of multi layer perception type, learned by error back propagation learning rule, is generally used for the verification or clustering of similar type of patterns. When learning is completed, the network has a constant value of output depending on a pattern. This paper shows that the intensity of neuron's out put can be controlled by a function which intensifies the excitatory interconnection coefficients or the inhibitory one between neurons in output layer and those in hidden layer. In this paper the value of factor in the function to control the output is derived from the know values of the neural network after learning is completed And also this paper show that the amount of an increased neuron's output in output layer by arbitary value of the factor is derived. For the applications increased recognition performance of a pattern than has distortion is introduced and the output of partially damaged neurons are first managed and this paper shows that the reduced recognition performance can be recovered.

• The Journal of the Acoustical Society of Korea
• /
• 제18권1호
• /
• pp.61-66
• /
• 1999

In this paper, we proposed various speaker adaptation methods and studied the performance of these methods. Methods which were studied in this paper are MAPE(Maximum A Posteriori Probability Estimation), Linear Spectral Estimating, Multi-Layer Perceptron and ARTMAP. In order to evaluate the performance of these methods, we used Korean isolated digits as the experimental data, the hybrid speaker adaptation method, which unified MAPE, linear spectral estimating and output probability of SCHMM, showed the better recognition result than those which performed other methods. And the method using ARTMAP showed the similar result to above hybrid method.