• 제어로봇시스템학회:학술대회논문집
• /
• 2000.10a
• /
• pp.20-20
• /
• 2000

Back-propagation neural network (BPNN) is the most prevalently used paradigm in modeling semiconductor manufacturing processes, which as a neuron activation function typically employs a bipolar or unipolar sigmoid function in either hidden and output layers. In this study, applicability of another linear function as a neuron activation function is investigated. The linear function was operated in combination with other sigmoid functions. Comparison revealed that a particular combination, the bipolar sigmoid function in hidden layer and the linear function in output layer, is found to be the best combination that yields the highest prediction accuracy. For BPNN with this combination, predictive performance once again optimized by incrementally adjusting the gradients respective to each function. A total of 121 combinations of gradients were examined and out of them one optimal set was determined. Predictive performance of the corresponding model were compared to non-optimized, revealing that optimized models are more accurate over non-optimized counterparts by an improvement of more than 30%. This demonstrates that the proposed gradient-optimized teaming for BPNN with a linear function in output layer is an effective means to construct plasma models. The plasma modeled is a hemispherical inductively coupled plasma, which was characterized by a 24 full factorial design. To validate models, another eight experiments were conducted. process variables that were varied in the design include source polver, pressure, position of chuck holder and chroline flow rate. Plasma attributes measured using Langmuir probe are electron density, electron temperature, and plasma potential.

• The Transactions of the Korean Institute of Electrical Engineers A
• /
• v.48 no.2
• /
• pp.168-175
• /
• 1999

In this paper, a modified RBF(Radial Basis Function) network structure is suggested for the prediction of a time-series with non-linear, non-stationary characteristics. Coventional RBF network predicting time series by using past outputs sense the trajectory of the time series and react when there exists strong relation between input and hidden activation function's RBF center. But this response is highly sensitive to level and trend of time serieses. In order to overcome such dependencies, hidden activation functions are modified to react to the increments of input variable and multiplied by increment(or dectement) for prediction. When the suggested structure is applied to prediction of Macyey-Glass chaotic time series, Lorenz equation, and Rossler equation, improved performances are obtained.

• Journal of The Korean Society of Agricultural Engineers
• /
• v.63 no.1
• /
• pp.103-116
• /
• 2021

Analysis of runoff is substantial for effective water management in the watershed. Runoff occurs by reaction of a watershed to the rainfall and has non-linearity and uncertainty due to the complex relation of weather and watershed factors. ANN (Artificial Neural Network), which learns from the data, is one of the machine learning technique known as a proper model to interpret non-linear data. The performance of ANN is affected by the ANN's structure, the number of hidden layer nodes, learning rate, and activation function. Especially, the activation function has a role to deliver the information entered and decides the way of making output. Therefore, It is important to apply appropriate activation functions according to the problem to solve. In this paper, ANN models were constructed to estimate runoff with different activation functions and each model was compared and evaluated. Sigmoid, Hyperbolic tangent, ReLU (Rectified Linear Unit), ELU (Exponential Linear Unit) functions were applied to the hidden layer, and Identity, ReLU, Softplus functions applied to the output layer. The statistical parameters including coefficient of determination, NSE (Nash and Sutcliffe Efficiency), NSEln (modified NSE), and PBIAS (Percent BIAS) were utilized to evaluate the ANN models. From the result, applications of Hyperbolic tangent function and ELU function to the hidden layer and Identity function to the output layer show competent performance rather than other functions which demonstrated the function selection in the ANN structure can affect the performance of ANN.

• Preventive Nutrition and Food Science
• /
• v.1 no.1
• /
• pp.41-45
• /
• 1996

This study was conducted to develop the fermentation kinetic model for the prediction of acidity and pH changes in Kimchi as a function of fermentation temperatures. The fitness of the model was evaluated using traditional two-step method and an alternative non-linear regression method. The changes in acidity and pH during fermentation followed the pattern of the first order reaction of a two-step method. As the fermentation temperature increased from 4$^{\circ}C$ to 28, the reaction rates of acidity and pH were increased 8.4 and 7.6 times, respectively. The activation energies of acidity and pH were 16.125 and 16.003kcal/mole. The average activation energies of acidity and pH using a non-linear method were 16.006 by the first order and 15.813 kcal/mole by the zero order, respectively. The non-linear procedure had better fitting 개 experimental data of the acidity and pH than two-step method. The shelf-lives based on the time to reach the 1.0% of acidity were 33.1day at 4$^{\circ}C$ and 2.8 day 28$^{\circ}C$.

• The Journal of the Korea Contents Association
• /
• v.21 no.3
• /
• pp.616-625
• /
• 2021

Deep neural networks are an approximation method that approximates an arbitrary function to a linear model and then repeats additional approximation using a nonlinear active function. In this process, the method of evaluating the performance of approximation uses the loss function. Existing in-depth learning methods implement approximation that takes into account loss functions in the linear approximation process, but non-linear approximation phases that use active functions use non-linear transformation that is not related to reduction of loss functions of loss. This study proposes parametric activation functions that introduce scale parameters that can change the scale of activation functions and location parameters that can change the location of activation functions. By introducing parametric activation functions based on scale and location parameters, the performance of nonlinear approximation using activation functions can be improved. The scale and location parameters in each hidden layer can improve the performance of the deep neural network by determining parameters that minimize the loss function value through the learning process using the primary differential coefficient of the loss function for the parameters in the backpropagation. Through MNIST classification problems and XOR problems, parametric activation functions have been found to have superior performance over existing activation functions.

• Proceedings of the Korea Concrete Institute Conference
• /
• 2002.10a
• /
• pp.81-84
• /
• 2002

Activation energy value is different according to cement, admixture and water-cement ratio also the relation of age-temperature is as non-linear as activation energy value is large. So to make accurate explanation for the effect of temperature on concrete strength development property, it is necessary to investigation for activation energy value. This study compares activation energy value recommended by Freiesleben and ASTM with activation energy value obtained by consequence of mortar examination according to ASTM C 1074-93. As the result of this study, activation energy value obtained by the study is 37.19KJ/mol, and in case of activation energy value obtained by the study explain temperature's influence about concrete strength development more accurate than activation energy value recommend by Freiesleben and ASTM.

• ETRI Journal
• /
• v.15 no.2
• /
• pp.35-51
• /
• 1993

This paper presents function approximation based on nonparametric estimation. As an estimation model of function approximation, a three layered network composed of input, hidden and output layers is considered. The input and output layers have linear activation units while the hidden layer has nonlinear activation units or kernel functions which have the characteristics of bounds and locality. Using this type of network, a many-to-one function is synthesized over the domain of the input space by a number of kernel functions. In this network, we have to estimate the necessary number of kernel functions as well as the parameters associated with kernel functions. For this purpose, a new method of parameter estimation in which linear learning rule is applied between hidden and output layers while nonlinear (piecewise-linear) learning rule is applied between input and hidden layers, is considered. The linear learning rule updates the output weights between hidden and output layers based on the Linear Minimization of Mean Square Error (LMMSE) sense in the space of kernel functions while the nonlinear learning rule updates the parameters of kernel functions based on the gradient of the actual output of network with respect to the parameters (especially, the shape) of kernel functions. This approach of parameter adaptation provides near optimal values of the parameters associated with kernel functions in the sense of minimizing mean square error. As a result, the suggested nonparametric estimation provides an efficient way of function approximation from the view point of the number of kernel functions as well as learning speed.

• Food Science and Biotechnology
• /
• v.16 no.6
• /
• pp.954-957
• /
• 2007

Gelation mechanism of xanthan-carob mixture (X/C) was investigated based on thermorheological behavior. Three X/C ratios (1:3, 1:1, and 3:1) were studied. Small amplitude oscillatory shear tests were performed to measure linear viscoelastic behavior during gelation. Temperature sweep ($-1^{\circ}C/min$) experiments were conducted. Using a non-isothermal kinetic model, activation energy (Ea) during gelation was calculated. At 1% total concentration, the Ea for xanthan fraction (${\phi}_x$)=0.25, 0.5, and 0.75 were 178, 159, and 123 kJ/mol, respectively. However, a discontinuity was observed in the activation energy plots. Based on this, two gelation mechanisms were presumed-association of xanthan and carob molecules and aggregation of polymer strands. The association process is the primary mechanism to form 3-D networks in the initial stage of gelation and the aggregation of polymer strands played a major role in the later stage.

• Journal of Korean Institute of Industrial Engineers
• /
• v.28 no.2
• /
• pp.147-161
• /
• 2002

In novelty detection, one attempts to discriminate abnormal patterns from normal ones. Novelty detection is quite difficult since, unlike usual two class classification problems, only normal patterns are available for training. Auto-Associative Multi-Layer Perceptron (AAMLP) has been shown to provide a good performance based upon the property that novel patterns usually have larger auto-associative errors. In this paper, we give a mathematical analysis of 2-layer AAMLP's output characteristics and empirical results of 2-layer and 4-layer AAMLPs. Various activation functions such as linear, saturated linear and sigmoid are compared. The 2-layer AAMLPs cannot identify non-linear boundaries while the 4-layer ones can. When the data distribution is multi-modal, then an ensemble of AAMLPs, each of which is trained with pre-clustered data is required. This paper contributes to understanding of AAMLP networks and leads to practical recommendations regarding its use.

• Structural Engineering and Mechanics
• /
• v.12 no.6
• /
• pp.685-698
• /
• 2001

In this paper, a general method for the automatic search for Strut-and-Tie (S&T) models representative of possible resistant mechanisms in reinforced concrete elements is proposed. The representativeness criterion here adopted is inspired to the principle of minimum strain energy and requires the consistency of the model with a reference stress field. In particular, a highly indeterminate pin-jointed framework of a given layout is generated within the assigned geometry of the concrete element and an optimum truss is found by the minimisation of a suitable objective function. Such a function allows us to search the optimum truss according to a reference stress field deduced through a F.E.A. and assumed as representative of the given continuum. The theoretical principles and the mathematical formulation of the method are firstly explained; the search for a S&T model suitable for the design of a deep beam shows the method capability in handling the reference stress path. Finally, since the analysis may consider the structure as linear-elastic or cracked and non-linear in both the component materials, it is shown how the proposed procedure allows us to verify the possibilities of activation of the design model, oriented to the serviceability condition and deduced in the linear elastic field, by following the evolution of the resistant mechanisms in the cracked non-linear field up to the structural failure.