Earticle

신경 회로망을 구현하기 위해 다양한 시도들이 이루어지고 있으며, 하드웨어적인 개선을 위해 전용 칩 개발이 이루 어지고 있다. 이러한 신경 회로망을 웨어러블 디바이스에 적용하기 위해서는 소형화와 저전력 동작이 필수적이다. 이러한 관점에서 적합한 구현 방법은 FPGA (field programmable gate array)를 사용한 디지털 회로 설계이다. 이 시스템을 구현하기 위해서는 성능 향상을 위해 신경 회로망의 많은 부분을 차지하는 학습 알고리즘을 FPGA 내 에 구현하여야 한다. 본 논문에서는 FPGA를 이용하여 다양한 학습 알고리즘 중 역전파 알고리즘을 구현하였으며, 구현 된 신경 회로망은 OR 게이트 연산을 통해 검증되었다. 또한 이러한 신경 회로망을 활용하여 다양한 사용자의 생체 신호 측정 결과를 분석할 수 있음을 확인하였다.

Neural networks can be implemented in variety of ways, and specialized chips is being developed for hardware improvement. In order to apply such neural networks to wearable devices, the compactness and the low power operation are essential. In this point of view, a suitable implementation method is a digital circuit design using field programmable gate array (FPGA). To implement this system, the learning algorithm which takes up a large part in neural networks must be implemented within FPGA for better performance. In this paper, a back propagation algorithm among various learning algorithms is implemented using FPGA, and this neural network is verified by OR gate operation. In addition, it is confirmed that this neural network can be used to analyze various users' bio signal measurement results by learning algorithm.

This study aims to implement the emergency monitoring robot system which predicts the current state of the patients without visiting the medical institutions by measuring the basic health status of the user's blood pressure, heartbeat, and basic health status of body temperature in the disaster emergency situation based on the Smart Grid. By arranging a large number of sensor(blood pressure, heartbeat, body temperature sensor) and measuring the bio signs, so the attached wireless XBee sensor can be stored in DB of robot, and it aims to draw the current state of the patients by analysis of stored bio data. Among 300 data obtained from the sensor, 1st data to 100th data were used for learning, and from 101st data to 300th data were used for assessment. 12 results were different among the total 300 assessment data, so it shows about 96% accuracy.

The back propagation (BP) algorithm is a very popular learning approach in multilayer feedforward networks. However, the most serious problems associated with the BP are local minima problem and slow convergence speeds. Over the years, many improvements and modifications of the BP learning algorithm have been reported. In this research, we propose a new modified BP learning algorithm by introducing adaptive gain together with adaptive momentum and adaptive learning rate into weight update process. By computer simulations, we demonstrate that the proposed algorithm can give a better convergence rate and can find a good solution in early time compare to the conventional BP. We use four common benchmark classification problems to illustrate the improvement in convergence time.

The standard back propagation algorithm for training artificial neural networks utilizes two terms, a learning rate and a momentum factor. The major limitations of this standard algorithm are the existence of temporary, local minima resulting from the saturation behaviour of the activation function, and the slow rates of convergence. Previous research demonstrated that in ‘feed forward’ algorithm, the slope of the activation function is directly influenced by a parameter referred to as ‘gain’. This research proposed an algorithm for improving the performance of the back propagation algorithm by introducing the adaptive gain of the activation function. The efficiency of the proposed algorithm is compared with conventional Gradient Descent Method and verified by means of simulation on four classification problems. In learning the patterns, the simulations result demonstrate that the proposed method converged faster on Wisconsin breast cancer and diabetes classification problem with an improvement ratio of nearly 2.8 and 1.2, 65% better on thyroid data sets and 97% success on IRIS classification problem. The results clearly show that the proposed algorithm significantly improves the learning speed of the conventional back-propagation algorithm.

Imbalanced data sets are difficult to be classified since most classifiers are developed based on the assumption that class distributions are well-balanced. In order to improve the error back-propagation algorithm for the classification of imbalanced data sets, a new error function is proposed. The error function controls weight-updating with regards to the classes in which the training samples are. This has the effect that samples in the minority class have a greater chance to be classified but samples in the majority class have a less chance to be classified. The proposed method is compared with the two-phase, threshold-moving, and target node methods through simulations in a mammography data set and the proposed method attains the best results.

The recent increase in social needs due to the development and aging of devices suitable for healthcare has led to the development of smart healthcare systems and suggests a new medical paradigm. As a result, there is a significant increase in the movement to accommodate changing health care concepts and analyses of measured biometric information using multiple measurement sensors are needed for ongoing healthcare. In order to provide rapid diagnosis and personalized diagnosis through the analyzed biometric information, a smart healthcare system with biometric technology is essential. In this paper, the goal is to store biometric information measured through multiple biometric sensors (blood pressure, heart rate, and body temperature sensors) in the DB using XBee sensors and then analyze stored biometric information through back-propagation algorithm, which is a neural network algorithm to derive the user's current state. As a result of the performance evaluation, the back-propagation algorithm for biosignal analysis showed an error rate of 2.96% on average and showed the lowest error rate when the window size was divided by 3000. In addition, the proposed system showed 87.6% accuracy through 700 performance evaluation data. It is expected that the proposed system will have high reliability and efficiency. The purpose of this study is to help the development of smart healthcare technology through researching the user-based service model and securing high quality data and contents based on smart healthcare using the proposed neural network based smart healthcare system.

The enhancement to the back-propagation algorithm presented in this paper has resulted from the need to extract sparsely connected networks from networks employing product terms. The enhancement works in conjunction with the back-propagation weight update process, so that the actions of weight zeroing and weight stimulation enhance each other. It is shown that the error measure, can also be interpreted as rate of weight change (as opposed to ${\Delta}W_{ij}$), and consequently used to determine when weights have reached a stable state. Weights judged to be stable are then compared to a zero weight threshold. Should they fall below this threshold, then the weight in question is zeroed. Simulation of such a system is shown to return improved learning rates and reduce network connection requirements, with respect to the optimal network solution, trained using the normal back-propagation algorithm for Multi-Layer Perceptron (MLP), Higher Order Neural Network (HONN) and Sigma-Pi networks.

This paper proposes a modified error function to improve the error back-propagation (EBP) algorithm for multi-Layer perceptrons (MLPs) which suffers from slow learning speed. It can also suppress over-specialization for training patterns that occurs in an algorithm based on a cross-entropy cost function which markedly reduces learning time. In the similar way as the cross-entropy function, our new function accelerates the learning speed of the EBP algorithm by allowing the output node of the MLP to generate a strong error signal when the output node is far from the desired value. Moreover, it prevents the overspecialization of learning for training patterns by letting the output node, whose value is close to the desired value, generate a weak error signal. In a simulation study to classify handwritten digits in the CEDAR [1] database, the proposed method attained 100% correct classification for the training patterns after only 50 sweeps of learning, while the original EBP attained only 98.8% after 500 sweeps. Also, our method shows mean-squared error of 0.627 for the test patterns, which is superior to the error 0.667 in the cross-entropy method. These results demonstrate that our new method excels others in learning speed as well as in generalization.

본 논문에서는 상황인지 시스템의 작업 부하를 줄이고 추론 성능을 향상시키기 위하여 개선된 역전파 알고리즘을 사용하는 진보된 학습 메커니즘을 위한 새로운 프레임워크를 제안한다. 학습 메커니즘은 상황인지 시스템의 전체 성능을 좌우하는 매우 중요한 부분이지만 현재까지 사용자들의 상황 정보를 대상으로 학습 메커니즘의 개선을 통한 상황인지 시스템의 성능을 향상하려는 연구는 많이 진행되지 않았다. 역전파 알고리즘은 상황 인식 시스템의 학습 메커니즘을 위한 가장 적합한 알고리즘 중에 하나로서 제안된 프레임워크는 기존의 역전파 알고리즘을 개선하고, 시스템 캐싱을 이용하여 작업 부하를 효율적으로 관리함으로써 추론 성능을 향상시켜 상황인지 시스템의 전체 성능을 향상시킨다. 성능평가를 통하여 제안된 프레임워크가 상황인지 시스템의 전체 성능을 향상시키는 것을 보인다.

In seeking to improve the workload efficiency and inference capability of context-aware systems, we propose a new framework for an advanced teaming mechanism that uses improved bath propagation (BP) algorithm. Even though a learning mechanism is one of the most important parts in a context-aware system, the existing algorithms focused on facilitating systems by elaborating the learning mechanism with user's context information are rare. BP is the most adaptable algorithm for learning mechanism of context-aware systems. By using the improved BP algorithm, the framework we proposed drastically improves the inference capability so that the overall performance is far better than other systems. Also, using the special system cache, the framework manages the workload efficiently. Experiments show that there is an obvious improvement in overall performanre of the context-awareness systems using the proposed framework.

The results of this study suggest the use of a Back Propagation Network (BPN) algorithm for the detection of esophageal erosions or abnormalities - which are the important signs of esophagitis - in the analysis of the color and textural aspects of clinical images obtained by endoscopy. The authors have investigated the optimization of the learning condition by the number of neurons in the hidden layer within the structure of the neural network. By optimizing learning parameters, we learned and have validated esophageal erosion images and/or ulcers functioning as the critical diagnostic criteria for esophagitis and associated abnormalities. Validation was established by using twenty clinical images. The success rates for detection of esophagitis during calibration and during validation were 97.91% and 96.83%, respectively.

The indoor noise of a ship is usually determined using the A-weighted sound pressure level. However, in order to better understand this phenomenon, evaluation parameters that more accurately reflect the human sense of hearing are required. To find the level of the satisfaction index of the noise inside a naval vessel such as "Loudness" and "Annoyance", psycho-acoustic evaluation of various sound recordings from the naval vessel was performed in a laboratory. The objective of this paper is to develop a single index of "Loudness" and "Annoyance" for noise inside a naval vessel according to a psycho-acoustic evaluation by using psychological responses such as Noise Rating (NR), Noise Criterion (NC), Room Criterion (RC), Preferred Speech Interference Level (PSIL) and loudness level. Additionally, in order to determine a single index of satisfaction for noise such as "Loudness" and "Annoyance", with respect to a human's sense of hearing, a back-propagation neural network is applied.

Back-propagation(BP) algorithm needs a lot of time to train the artificial neural network (ANN) to get high accuracy level in classification tasks. So there have been extensive researches to process back-propagation algorithm on parallel processors. This paper prsents a linear systolic array which calculates forward-backward propagation of BP algorithm at the same time using effective space-time transformation and PE structure. First, we analyze data flow of forwared and backward propagations and then, represent the BP algorithm into data dapendency graph (DG) which shows parallelism inherent in the BP algorithm. Next, apply space-time transformation on the DG of ANN is turn with orthogonal direction projection. By doing so, we can get a snakelike systolic array. Also we calculate the interval of input for parallel processing, calculate the indices to make the right datas be used at the right PE when forward and bvackward propagations are processed in the same PE. And then verify the correctness of output when forward and backward propagations are executed at the same time. By doing so, the proposed system maximizes parallelism of BP algorithm, minimizes th enumber of PEs. And it reduces the execution time by 2 times through making idle PEs participate in forward-backward propagation at the same time.

The problems in back-propagation(BP) generally are learning speed and misclassification due to lacal minimum. In this paper, to solve these problems, the classical modified methods of BP are reviewed and an extension of the BP to compensate the sigmoide-prime function around the extremity where the actual output of a unit is close to zero or one is proposed. The proposed method is not onlu faster than the conventional methods in learning speed but has an advantage of setting variables easily because it shows good classification results over the vast and uncharted space about the variations of learning rate, etc.. And it is simple for hardware implementation.

In this paper, we propose a modified error function to improve the EBP(Error Back-Propagation) algorithm of Multi-Layer Perceptrons. Using the modified error function, the output node of MLP generates a strong error signal in the case that the output node is far from the desired value, and generates a weak error signal in the opposite case. This accelerates the learning speed of EBP algorothm in the initial stage and prevents overspecialization for training patterns in the final stage. The effectiveness of our modification is verified through the simulation of handwritten digit recognition.

The error back-propagation(BP) algorithm is widely used for finding optimum weights of multi-layer neural networks. However, the critical drawback of the BP algorithm is its slow convergence of error. The major reason for this slow convergence is the premature saturation which is a phenomenon that the error of a neural network stays almost constant for some period time during learning. An inappropriate selections of initial weights cause each neuron to be trapped in the premature saturation state, which brings in slow convergence speed of the multi-layer neural network. In this paper, to overcome the above problem, Micro-Genetic algorithms(.mu.-GAs) which can allow to find the near-optimal values, are used to select the proper weights and slopes of activation function of neurons. The effectiveness of the proposed algorithms will be demonstrated by some computer simulations of two d.o.f planar robot manipulator.

Neural networks are in the framework of sensorbased tracking control of robot manipulators. They learn by practice movements the relationship between PSD (an analog Position Sensitive Detector) sensor readings for target positions and the joint commands to reach them. Using this configuration, the system can track or follow a moving or stationary object in real time. Furthermore, an efficient neural network architecture has been developed for real time learning. This network uses multiple sets of simple backpropagation networks one of which is selected according to which division (corresponding to a cluster of the self-organizing feature map) in data space the current input data belongs to. This lends itself to a very training and processing implementation required for real time control.

A major problem in automation of turning operation is the difficulty in obtaining a sufficient and reliable chip control. Therefore it becomes desirable to find a method which can detect the chip form. In this paper, a method of the identification of chip form using output of pyrometer and neural network technique is developed. An efficiency of developed method is examined by experiments in turning and the validity of it is confirmed.

Neural networks are works are used in the framework of sensor based tracking control of robot manipulators. They learn by practice movements the relationship between PSD(an analog Position Sensitive Detector) sensor readings for target positions and the joint commands to reach them. Using this configuration, the system can track or follow a moving or stationary object in real time. Furthermore, an efficient neural network architecture has been developed for real time learning. This network uses multiple sets of simple back propagation networks one of which is selected according to which division(corresponding to a cluster of the self-organizing feature map) in data space the current input data belongs to. This lends itself to a very training and processing implementation required for real time control.

The dicing process cuts a semiconductor wafer to lengthwise and crosswise direction by using a rotating circular diamond blade. But inferior goods are made under the influence of several parameters in dicing such as blade, wafer, cutting water and cutting conditions. This paper describes a monitoring algorithm using neural network in order to find out an instant of vibration signal change when bad dicing appears. The algorithm is composed of two steps: feature extraction and decision. In the feature extraction, five features processed from vibration signal which is acquired by accelerometer attached on blade head are proposed. In the decision, back-propagation neural network is adopted to classify the dicing process into normal and abnormal dicing, and normal and damaged blade. Experiments have been performed for GaAs semiconductor wafer in the case of normal/abnormal dicing and normal/damaged blade. Based upon observation of the experimental results, the proposed scheme shown has a good accuracy of classification performance by which the inferior goods decreased from 35.2% to 6.5%.