• ETRI Journal
• /
• v.16 no.1
• /
• pp.35-43
• /
• 1994

In this paper, the robustness of the artificial neural networks to noise is demonstrated with a multilayer perceptron, and the reason of robustness is due to the statistical orthogonality among hidden nodes and its hierarchical information extraction capability. Also, the misclassification probability of a well-trained multilayer perceptron is derived without any linear approximations when the inputs are contaminated with random noises. The misclassification probability for a noisy pattern is shown to be a function of the input pattern, noise variances, the weight matrices, and the nonlinear transformations. The result is verified with a handwritten digit recognition problem, which shows better result than that using linear approximations.

• Proceedings of the Acoustical Society of Korea Conference
• /
• spring
• /
• pp.27-30
• /
• 2000

본 논문은 HMM(Hidden Markov Model)을 이용하여 인식을 수행할 경우의 오류를 최소화 할 수 있는 후 처리 과정으로 신경망을 결합시켜 HMM 단독으로 사용하였을 때 보다 높은 인식률을 얻을 수 있는 HMM과 신경망의 하이브리드시스템을 제안한다. HMM을 이용하여 학습한 후 학습에 참여하지 않은 데이터를 인식하였을 때 오인식 데이터를 정인식으로 인식하도록 HMM의 출력으로 얻은 각 출력확률을 후 처리에 사용될 MLP(Multilayer Perceptrons)의 학습용으로 사용하여 MLP를 학습하여 HMM과 MLP을 결합한 하이브리드 모델을 만든다. 이와 같은 HMM과 신경망을 결합한 하이브리드 모델을 사용하여 단독 숫자음과 4연 숫자음 데이터에서 실험한 결과 HMM 단독으로 사용하였을 때 보다 각각 약 $4.5\%$, $1.3\%$의 인식률 향상이 있었다. 기존의 하이브리드 시스템이 갖는 많은 학습시간이 소요되는 문제점과 실시간 음성인식시스템을 구현할 때의 학습데이터의 부족으로 인한 인식률 저하를 해결할 수 있는 방법임을 확인할 수 있었다.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 2003.09a
• /
• pp.216-219
• /
• 2003

Among the techniques to protect private information by adopting biometrics, speaker verification is expected to be widely used due to advantages in convenient usage and inexpensive implementation cost Speaker verification should achieve a high degree of the reliability in the verification nout the flexibility in speech text usage, and the efficiency in verification system complexity. Continuants have excellent speaker-discriminant power and the modest number of phonemes in the category, and multilayer perceptrons (MLPs) have superior recognition ability and fast operation speed. In consequence, the two provide viable ways for speaker verification system to obtain the above properties. This paper implements a system to which continuants and MLPs are applied, and evaluates the system using a Korean speech database. The results of the experiment prove that continuants and MLPs enable the system to acquire the three properties.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 2003.09a
• /
• pp.300-303
• /
• 2003

An Algorithm based on direct implementation of variable structure systems theory approach is proposed for on-line training of multilayer perceptrons. Network structures which have multiple inputs, single output and one hidden layer are considered and the weights are assumed to have capabilities for continuous time adaptation. The zero level set of the network learning error is regarded as a sliding surface in the learning parameters space. A sliding mode trajectory can be brought on and reached in finite time on such a sliding manifold. Results from simulated on-line identification task for a two-link planar manipulator dynamics are also presented.

• Proceedings of the Korea Contents Association Conference
• /
• 2007.11a
• /
• pp.217-220
• /
• 2007

LLE(Locally Linear Embedding) has been proposed to compute low dimensional, neighborhood preserving embeddings of high dimensional data. Here, we should perform whole processes of LLE when untrained patterns are presented. In this paper, we propose a training of MLPs(Multilayer Perceptrons) to perform the mapping of LLE from high dimensional data to low dimensional ones.

• The Journal of the Korea Contents Association
• /
• v.3 no.3
• /
• pp.98-102
• /
• 2003

For elucidating the key role of hidden neurons in information processing of Multilayer perceptrons(MLPs), we prove that the correlation coefficient between weighted sums to hidden neurons decreases under element-wise nonlinear transformations. This is verified through training of MLPs for an isolated word recognition problem. From this result, we can say that the element-wise nonlinear functions reduces redundancy in the information contents of hidden neurons.

• ETRI Journal
• /
• v.22 no.3
• /
• pp.10-18
• /
• 2000

Although an n-th order cross-entropy (nCE) error function resolves the incorrect saturation problem of conventional error backpropagation (EBP) algorithm, performance of multilayer perceptrons (MLPs) trained using the nCE function depends heavily on the order of nCE. In this paper, we propose an adaptive learning rate to markedly reduce the sensitivity of MLP performance to the order of nCE. Additionally, we propose to limit error signal values at out-put nodes for stable learning with the adaptive learning rate. Through simulations of handwritten digit recognition and isolated-word recognition tasks, it was verified that the proposed method successfully reduced the performance dependency of MLPs on the nCE order while maintaining advantages of the nCE function.

• The Transactions of the Korean Institute of Electrical Engineers
• /
• v.39 no.9
• /
• pp.985-996
• /
• 1990

An inverse kinematic solution of a robot manipulator using multilayer perceptrons is proposed. Neural networks allow the solution of some complex nonlinear equations such as the inverse kinematics of a robot manipulator without the need for its model. However, the back-propagation (BP) learning rule for multilayer perceptrons has the major limitation of being too slow in learning to be practical. In this paper, a new algorithm named Dynamically Reconfiguring BP is proposed to improve its learning speed. It uses a modified version of Kohonen's Self-Organizing Feature Map (SOFM) to partition the input space and for each input point, select a subset of the hidden processing elements or neurons. A subset of the original network results from these selected neuron which learns the desired mapping for this small input region. It is this selective property that accelerates convergence as well as enhances resolution. This network was used to learn the parity function and further, to solve the inverse kinematic problem of a robot manipulator. The results demonstrate faster learning than the BP network.

• Journal of Intelligence and Information Systems
• /
• v.14 no.3
• /
• pp.25-39
• /
• 2008

The objective of this study is to forecast the operational continuous ability using Artificial Neural Networks in battalion defensive operation for the commander decision making support. The forecasting of the combat result is one of the most complex issue in military science. However, it is difficult to formulate a mathematical model to evaluate the combat power of a battalion in defensive operation since there are so many parameters and high temporal and spatial variability among variables. So in this study, we used company combat power level data in Battalion Command in Battle Training as input data and used Feed-Forward Multilayer Perceptrons(MLP) and General Regression Neural Network (GRNN) to evaluate operational continuous ability. The results show 82.62%, 85.48% of forecasting ability in spite of non-linear interactions among variables. We think that GRNN is a suitable technique for real-time commander's decision making and evaluation of the commitment priority of troops in reserve.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.10 no.4
• /
• pp.374-378
• /
• 2000

in information retrieval, lack of positive examples is a main cause of poor performance. In this case most learning algorithms may not characteristics in the data to low recall. To solve the problem of unbalanced data, we propose a boosting method that uses linear perceptrons as weak learnrs. The perceptrons are trained on local data sets. The proposed algorithm is applied to text filtering problem for which only a small portion of positive examples is available. In the experiment on category crude of the Reuters-21578 document set, the boosting method achieved the recall of 80.8%, which is 37.2% improvement over multilayer with comparable precision.