인공신경망 학습 과정의 시각적 해석

최설매,김동현,한승기 한국물리학회 2020 새물리 Vol.70 No.10

Recently, artificial neural networks (ANNs) have been utilized for various tasks, including image classification, speech recognition, and machine translation, and for medical diagnosis. For practical applications, we use neural networks composed of a very large number of neurons, but we do not have much information on how to fix the number of hidden layers or the number of neurons in a layer. In this paper, a graphical illustration of neural learning in simple neural networks of a multilayer perceptron (MLP) is presented. In the case of XOR-like problems, the learning process corresponds to finding a line or surface that separates several states into two groups in a higher dimensional space. Here, we illustrate graphically how the bipartition problem depends on the number of neurons in a layer, we also address the meaning of adding a layer in the network. We expect that this intuitive graphical understanding of increasing the number of neurons or layers in simple neural networks will be useful in constructing neural networks for practical problems. 최근에 인공신경망 (ANN) 은 이미지 분류, 음성 인식, 기계어 번역, 그리고 임상 진단 등에 많이활용되고 있다. 실용적 응용에서 사용하고 있는 신경망은 수많은 뉴런으로 구성되어 있지만, 뉴런층을 어떻게 결정하고 또한 하나의 층에서 뉴런의 수는 어떻게 결정하는지 많은 정보가 주어져있지 않다. 본 논문에서는 간단한 다층 퍼셉트론 (MLP) 신경망을 이용하여 신경망 학습 결과에대한 도형적 풀이를 제시한다. XOR과 같은 단순한 문제의 학습 과정은 높은 차원에서 다양한상태를 두 개의 군으로 분리하는 표면을 구하는 문제에 해당한다. 이 논문에서는 간단한 이원 분류문제의 학습이 하나의 층에 있는 뉴런의 수에 따라서 어떻게 달라지는지, 또한 신경망에서 뉴런층을 더하는 것이 도형적으로 어떤 의미를 가지는지를 설명한다. 간단한 신경망에서 뉴런의 수 혹은층을 늘이는 것에 대한 직관적이며 도형적인 이해는 실제 문제에 대한 인공신경망 구축에 도움이 될 것이다.

Control of Nonlinear System with a Disturbance Using Multilayer Neural Networks

Seong, Hong-Seok Institute of Control 2000 Transaction on control, automation and systems eng Vol.2 No.3

The mathematical solutions of the stability convergence are important problems in system control. In this paper such problems are analyzed and resolved for system control using multilayer neural networks. We describe an algorithm to control an unknown nonlinear system with a disturbance, using a multilayer neural network. We include a disturbance among the modeling error, and the weight update rules of multilayer neural network are derived to satisfy Lyapunov stability. The overall control system is based upon the feedback linearization method. The weights of the neural network used to approximate a nonlinear function are updated by rules derived in this paper . The proposed control algorithm is verified through computer simulation. That is as the weights of neural network are updated at every sampling time, we show that the output error become finite within a relatively short time.

Deep neural networks for gas concentration estimation and the effect of hyperparameter optimization on the estimation performance

Hee-Deok Jang,Jae-Hyeon Park,Hyunwoo Nam,Dong Eui Chang 제어로봇시스템학회 2022 제어로봇시스템학회 국제학술대회 논문집 Vol.2022 No.11

Many studies propose gas concentration estimators using machine learning algorithms owing to their high performance. Recently, estimation models using deep neural network have been studied due to their higher performance than conventional machine learning algorithms. The performance of deep neural network can be increased by hyperparameter optimization. In this paper, we propose two deep neural networks for gas concentration estimation and analyze how hyperparameter optimization affects the performance of the proposed deep neural networks. We optimize the hyperparameters of the proposed neural networks and compare the performance with conventional machine learning models. We train the proposed neural networks and evaluate the performance of the models with an open dataset. We confirm that the optimized neural network models show the high performance in gas concentration estimation, and that models using unoptimized parameters may show worse performance than conventional machine learning model.

Hybrid no-propagation learning for multilayer neural networks

Adhikari, Shyam Prasad,Yang, Changju,Slot, Krzysztof,Strzelecki, Michal,Kim, Hyongsuk Elsevier 2018 Neurocomputing Vol.321 No.-

<P><B>Abstract</B></P> <P>A hybrid learning algorithm suitable for hardware implementation of multi-layer neural networks is proposed. Though backpropagation is a powerful learning method for multilayer neural networks, its hardware implementation is difficult due to complexities of the neural synapses and the operations involved in error backpropagation. We propose a learning algorithm with performance comparable to but easier than backpropagation to be implemented in hardware for on-chip learning of multi-layer neural networks. In the proposed learning algorithm, a multilayer neural network is trained with a hybrid of gradient-based delta rule and a stochastic algorithm, called Random Weight Change. The parameters of the output layer are learned using the delta rule, whereas the inner layer parameters are learned using Random Weight Change, thereby the overall multilayer neural network is trained without the need for error backpropagation. Experimental results showing better performance of the proposed hybrid learning rule than either of its constituent learning algorithms, and comparable to that of backpropagation on the benchmark MNIST dataset are presented. Hardware architecture illustrating the ease of implementation of the proposed learning rule in analog hardware vis-a-vis the backpropagation algorithm is also presented.</P>

Neural-based prediction of structural failure of multistoried RC buildings

Sirshendu Hore,Sankhadeep Chatterjee,Sarbartha Sarkar,Nilanjan Dey,Amira S. Ashour,Dana Bălas-Timar,Valentina E. Balas 국제구조공학회 2016 Structural Engineering and Mechanics, An Int'l Jou Vol.58 No.3

Various vague and unstructured problems encountered the civil engineering/ designers that persuaded by their experiences. One of these problems is the structural failure of the reinforced concrete (RC) building determination. Typically, using the traditional Limit state method is time consuming and complex in designing structures that are optimized in terms of one/many parameters. Recent research has revealed the Artificial Neural Networks potentiality in solving various real life problems. Thus, the current work employed the Multilayer Perceptron Feed-Forward Network (MLP-FFN) classifier to tackle the problem of predicting structural failure of multistoried reinforced concrete buildings via detecting the failure possibility of the multistoried RC building structure in the future. In order to evaluate the proposed method performance, a database of 257 multistoried buildings RC structures has been constructed by professional engineers, from which 150 RC structures were used. From the structural design, fifteen features have been extracted, where nine features of them have been selected to perform the classification process. Various performance measures have been calculated to evaluate the proposed model. The experimental results established satisfactory performance of the proposed model.

다중 생체신호를 이용한 신경망 기반 전산화 감정해석

이지은 ( Jee-eun Lee ),김병남 ( Byeong-nam Kim ),유선국 ( Sun-kook Yoo ) 한국감성과학회 2017 감성과학 Vol.20 No.2

Emotion affects many parts of human life such as learning ability, behavior and judgment. It is important to understand human nature. Emotion can only be inferred from facial expressions or gestures, what it actually is. In particular, emotion is difficult to classify not only because individuals feel differently about emotion but also because visually induced emotion does not sustain during whole testing period. To solve the problem, we acquired bio-signals and extracted features from those signals, which offer objective information about emotion stimulus. The emotion pattern classifier was composed of unsupervised learning algorithm with hidden nodes and feature vectors. Restricted Boltzmann machine (RBM) based on probability estimation was used in the unsupervised learning and maps emotion features to transformed dimensions. The emotion was characterized by non-linear classifiers with hidden nodes of a multi layer neural network, named deep belief network (DBN). The accuracy of DBN (about 94 %) was better than that of back-propagation neural network (about 40 %). The DBN showed good performance as the emotion pattern classifier.

방사형기저함수망을 이용한 열간 사상압연의 압연하중 예측에 관한 연구

손준식,이덕만,김일수,최승갑 한국공작기계학회 2004 한국생산제조학회지 Vol.13 No.6

A major concern at present is the simultaneous control of transverse thickness profile and flatness in the finishing stages of hot rolling process. The mathematical modeling of hot rolling process has long been recognized to be a desirable approach to investigate rolling operating practice and the design of mill equipment to improve productivity and quality. However, many factors make the mathematical analysis of the rolling process very complex and time-consuming. In order to overcome these problems and to obtain an accurate rolling force, the predicted model of rolling force using neural networks has widely been employed. In this paper, Radial Basis Function Network(RBFN) is applied to improve the accuracy of rolling force prediction in hot rolling mill. In order to verify and analyze the performance of applied neural network, the comparison with the measured rolling force and the predicted results using two different neural networks-RBFN, MLP, has respectively been carried out. The results obtained using RBFN neural network are much more accurate those obtained the MLP.

신경회로망을 이용한 밀링 작업에서의 공구 감시

최기상 서울市立大學校 1992 論文集 Vol.26 No.-

An on-line tool wear detection system for face milling operations is developed, and experimentally evaluated. The system employs multiple sensors(AE sensor and dynamometer), and the signals from these sensors are processed using multichannel autoregressive (AR) series model. Decision on the state of cutting tool was made by neural network (multilayered perceptron) using the AR coefficients as input pattern. To learn the necessary input/output mapping for tool wear detection, the parameters of the network are adjusted according to the back propagation (BP) method during off-line training. The results of experimental evaluation show that the system works well over a wide range of cutting conditions (feed rate, depth of cut), and the ability of the system to detect tool wear is improved due to the generalization and self-organizing properties of the neural network.

다중퍼셉트론을 이용한 기초자치단체의 군집화

최기철 정보과학연구소부산외국어대학교 1999 情報科學論集 Vol.1 No.-

There are 232 basic self-governing communities in KOREA. Each communities can be classified as large city type, medium/small city type, and rural area type based on the location and number of population. These communities may have different characters on the type. In this paper, we have classified the 232 communities into two groups and tried discriminant analysis by using the housing and demographic variables to find whether the groups can be explained by the variables. We have also tried the multilayer perceptron method to find the classification. By comparing the classification results of multilayer perceptron and discriminant analysis, we have tried to understand the system of multilayer perceptron that is a major classification method of neural network.

제2형 당뇨병의 위험인자 분석을 위한 다층 퍼셉트론과 로지스틱 회귀 모델의 비교

서혜숙,최진욱,이홍규 대한의용생체공학회 2001 의공학회지 Vol.22 No.4

The statistical regression model is one of the most frequently used clinical analysis methods. It has basic assumption of linearity, additivity and normal distribution of data. However, most of biological data in medical field are nonlinear and unevenly distributed. To overcome the discrepancy between the basic assumption of statistical model and actual biological data, we propose a new analytical method based on artificial neural network. The newly developed multilayer perceptron(MLP) is trained with 120 data set (60 normal, 60 patient). On applying test data, it shows the discrimination power of 0.76. The diabetic risk factors were also identified from the MLP neural network model and the logistic regression model. The signigicant risk factors identified by MLP model were post prandial glucose level(PP2), sex(male), fasting blood sugar(FBS) level, age, SBP, AC and WHR. Those from the regression model are sex(male), PP2, age and FBS. The combined risk factors can be identified using the MLP model. Those are total cholesterol and body weight, which is consistent with the result of other clinical studies. From this experiment we have learned that MLP can be applied to the combined risk factor analysis of biological data which can not be provided by the conventional statistical method.