• Proceedings of the KSME Conference
• /
• 2008.11a
• /
• pp.1588-1593
• /
• 2008

This paper describes a real-time isometric pinch force prediction algorithm from surface electromyogram (sEMG) using multilayer perceptron (MLP) for human robot interactive applications. The activities of seven muscles which are observable from surface electrodes and also related to the movements of the thumb and index finger joints were recorded during pinch force experiments. For the successful implementation of the real-time prediction algorithm, an off-line analysis was performed using the recorded activities. Four muscles were selected for the force prediction by using the Fisher linear discriminant analysis among seven muscles, and the four muscle activities provided effective information for mapping sEMG to the pinch force. The MLP structure was designed to make training efficient and to avoid both under- and over-fitting problems. The pinch force prediction algorithm was tested on five volunteers and the results were evaluated using two criteria: normalized root mean squared error (NRMSE) and correlation (CORR). The training time for the subjects was only 2 min 29 sec, but the prediction results were successful with NRMSE = 0.112 ${\pm}$ 0.082 and CORR = 0.932 ${\pm}$ 0.058. These results imply that the proposed algorithm is useful to measure the produced pinch force without force sensors in real-time. The possible applications include controlling bionic finger robot systems to overcome finger paralysis or amputation.

• Journal of Biosystems Engineering
• /
• v.43 no.3
• /
• pp.229-236
• /
• 2018

Purpose: Makgeolli is a traditional alcoholic beverage made from rice with a fermentation starter called "nuruk." The concentration of alcohol in makgeolli depends on the temperature of the fermentation tank. It is important to monitor the alcohol concentration to manage the makgeolli production process. Methods: Data were collected from 84 makgeolli fermentation tanks over a year period. Independent variables included the temperatures of the tanks and the room where the tanks were located, as well as the quantity, acidity, and water concentration of the source. Software for the multilayer perceptron model (MLP) was written in Python using the Scikit-learn library. Results: Many models were created for which the optimization converged within 100 iterations, and their coefficients of determination $R^2$ were considerably high. The coefficient of determination $R^2$ of the best model with the training set and the test set were 0.94 and 0.93, respectively. The fact that the difference between them was very small indicated that the model was not overfitted. The maximum and minimum error was approximately 2% and the total MSE was 0.078%. Conclusions: The MLP model could help predict the alcohol concentration and to control the production process of makgeolli. In future research, the optimization of the production process will be studied based on the model.

• Annual Conference of KIPS
• /
• 2016.04a
• /
• pp.186-188
• /
• 2016

본 논문에서는 모바일 환경에서 동적서명인식에 관해 원본서명과 숙련된 위조서명의 구분을 검증하는 방법을 제안한다. 속도/거리 정보 실험(Data1)과 속도/거리정보와 가속도계를 추가 실험(Data2)을 원본 서명과 위조서명에 대한 테이블을 만들고, 비교하여 원본 서명의 인식률 확인한다. 제시한 방법은 각각 모바일 환경에서 10명이 20 번삑 손가락으로 테스트 하였다. 원본서명에서 딥 러닝중의 하나인 MLP를 실험한 결과 원본 서명에서 Data1은 92%, Data2는 95%의 정확도를 보였으며, 위조서명에서 Data1은 82%, Data2는 85%를 보였다. 그리고 AE에서 실험한 결과 Data1은 원본 서명에서 Data1은 95%, Data2는 97%의 정확도를 보였으며, 위조서명에서 Data1은 91.5%, Data2는 93%의 정확도가 보였다. 실험결과 위조서명에 대해서는 MLP로 위조서명을 분류하는 것보다 OAE에서 분류하는 것이 더 좋은 정확도를 보여준다.

• Journal of Aerospace System Engineering
• /
• v.12 no.2
• /
• pp.15-22
• /
• 2018

A compressible multiphase flow was investigated using a DIM consisting of seven equations, including the fifth-order MLP and a modified HLLC Riemann solver to achieve a precise interface structure of liquid and gas. The numerical methods were verified by comparing the flow structures of the high-pressure water and low-pressure air in the shock tube. A 2D air-helium shock-bubble interaction at the incident shock wave condition (Mach number 1.22) was numerically solved and verified using the experimental results.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 2002.12a
• /
• pp.162-165
• /
• 2002

Q-learning은 강화학습의 한 방법으로서, 여러 분야에 널리 응용되고 있는 기법이다. 최근에는 Linear Quadratic Regulation(이하 LQR) 문제에 성공적으로 적용된 바 있는데, 특히, 시스템모델의 파라미터에 대한 구체적인 정보가 없는 상태에서 적절한 입력과 출력만을 가지고 학습을 통해 문제를 해결할 수 있어서 상황에 따라서 매우 실용적인 대안이 될 수 있다. Neural Q-learning은 이러한 Q-learning의 Q-value를 MLP(multilayer perceptron) 신경망의 출력으로 대치시킴으로써, 비선형 시스템의 최적제어 문제를 다룰 수 있게 한 방법이다. 그러나, Neural Q방식은 신경망의 구조를 먼저 결정한 후 역전파 알고리즘을 이용하여 학습하는 절차를 취하기 때문에, 시행착오를 통하여 신경망 구조를 결정해야 한다는 점, 역전파 알고리즘의 적용으로 인해 신경망의 연결강도 값들이 지역적 최적해로 수렴한다는 점등의 문제점을 상속받는 한계가 있다. 따라서, 본 논문에서는 Neural-0 학습의 도구로, 역전파 알고리즘으로 학습되는 MLP 신경망을 사용하는 대신 최근 들어 여러 분야에서 그 성능을 인정받고 있는 서포트 벡터 학습법을 사용하는 방법을 택하여, $\varepsilon$-SVR(Epsilon Support Vector Regression)을 이용한 Q-value 근사 기법을 제안하고 관련 수식을 유도하였다. 그리고, 모의 실험을 통하여, 제안된 서포트 벡터학습 기반 Neural-Q 방법의 적용 가능성을 알아보았다.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.22 no.5
• /
• pp.1026-1035
• /
• 1997

In this paper, we studied for phoneme recognition using GPFN and PNN as a kind of RBFN. The structure of RBFN is similar to a feedforward networks but different from choosing of activation function, reference vector and learnign algorithm in a hidden layer. Expecially sigmoid function in PNN is replaced by one category included exponential function. And total calculation performance is high, because PNN performs pattern classification with out learning. In phonemerecognition experiment with 5 vowel and 12 consant, recognition rates of GPFN and PNN as a kind of RBFN reflected statistic characteristic of speech are higher than ones of MLP in case of using test data and quantizied data by VQ and LVQ.

• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.61 no.8
• /
• pp.1199-1203
• /
• 2012

H.264 adopted many compression tools to increase image data compression efficiency such as B frame bi-directional predictions, the direct mode coding and so on. Despite its high compression efficiency, H.264 can suffer from its long coding time due to the complicated tools of H.264. To realize a high performance H.264, several fast algorithms were proposed. One of them is adaptive fast direct mode decision algorithm using mode and Lagrangian cost prediction for B frame in H.264/AVC (MLP) algorithm which can determine the direct coding mode for macroblocks without a complex mode decision process. However, in this algorithm, macroblocks not satisfying the conditions of the MLP algorithm are required to process the complex mode decision calculation, yet suffering a long coding time. To overcome the problem, this paper proposes a fast direct mode prediction algorithm. Simulation results show that the proposed algorithm can determine the direct mode coding without a complex mode decision process for 42% more macroblocks and, this algorithm can reduce coding time by up to 23%, compared with Jin's algorithm. This enables to encode B frames fast with a less quality degradation.

• The Transactions of the Korean Institute of Electrical Engineers D
• /
• v.54 no.7
• /
• pp.468-474
• /
• 2005

Brain-machine interface (BMI) based on neuronal spike trains is regarded as one of the most promising means to restore basic body functions of severely paralyzed patients. The spike train decoding algorithm, which extracts underlying information of neuronal signals, is essential for the BMI. Previous studies report that a linear filter is effective for this purpose and there is no noteworthy gain from the use of nonlinear mapping algorithms, in spite of the fact that neuronal encoding process is obviously nonlinear. We designed several decoding algorithms based on the linear filter, and two nonlinear mapping algorithms using multilayer perceptron (MLP) and support vector machine regression (SVR), and show that the nonlinear algorithms are superior in general. The MLP often showed unsatisfactory performance especially when it is carelessly trained. The nonlinear SVR showed the highest performance. This may be due to the superiority of the SVR in training and generalization. The advantage of using nonlinear algorithms were more profound for the cases when there are false-positive/negative errors in spike trains.

• KIISE Transactions on Computing Practices
• /
• v.21 no.8
• /
• pp.567-571
• /
• 2015

Transition-based dependency parsing requires much time and efforts to design and select features from a very large number of possible combinations. Recent studies have successfully applied Multi-Layer Perceptrons (MLP) to find solutions to this problem and to reduce the data sparseness. However, most of these methods have adopted greedy search and can only consider a limited amount of information from the context window. In this study, we use a Recurrent Neural Network to handle long dependencies between sub dependency trees of current state and current transition action. The results indicate that our method provided a higher accuracy (UAS) than an MLP based model.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.14 no.5
• /
• pp.943-950
• /
• 2019

Machine learning can greatly improve the efficiency of work by replacing people. In particular, the importance of machine learning is increasing according to the requests of fourth industrial revolution. This paper predicts monthly power transactions using MLP, RNN, LSTM, and ANFIS of neural network algorithms. Also, this paper used monthly electricity transactions for mount and money, final energy consumption, and diesel fuel prices for vehicle provided by the National Statistical Office, from 2001 to 2017. This paper learns each algorithm, and then shows predicted result by using time series. Moreover, this paper proposed most excellent algorithm among them by using RMSE.