• The Journal of the Korea institute of electronic communication sciences
• /
• v.15 no.1
• /
• pp.85-92
• /
• 2020

Deep learning shows outstanding performance in image and video analysis, such as object classification, object detection and semantic segmentation. In this paper, it is analyzed that the performances of deep learning models can be affected by characteristics of train dataset. It is proposed as a method for selecting activation function and optimization algorithm of deep learning to classify facial expression. Classification performances are compared and analyzed by applying various algorithms of each component of deep learning model for CK+, MMI, and KDEF datasets. As results of simulation, it is shown that genetic algorithm can be an effective solution for optimizing components of deep learning model.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 2003.09b
• /
• pp.95-98
• /
• 2003

In this paper, an adaptive fuzzy neural control of unknown nonlinear systems based on the rapid learning algorithm is proposed for optimal parameterization. We combine the advantages of fuzzy control and neural network techniques to develop an adaptive fuzzy control system for updating nonlinear parameters of controller. The Fuzzy Neural Network(FNN), which is constructed by an equivalent four-layer connectionist network, is able to learn to control a process by updating the membership functions. The free parameters of the AFN controller are adjusted on-line according to the control law and adaptive law for the purpose of controlling the plant track a given trajectory and it's initial values are off-line preprocessing, In order to improve the convergence of the learning process, we propose a rapid learning algorithm which combines the error back-propagation algorithm with Aitken's $\delta$$\^$2/ algorithm. The heart of this approach ls to reduce the computational burden during the FNN learning process and to improve convergence speed. The simulation results for nonlinear plant demonstrate the control effectiveness of the proposed system for optimal parameterization.

• Journal of Internet Computing and Services
• /
• v.4 no.5
• /
• pp.69-76
• /
• 2003

In this paper we show a new learning algorithm for pattern classification. This algorithm considered a scheme to find a solution to a problem of incremental learning algorithm when the structure becomes too complex by noise patterns included in learning data set. Our approach for this problem uses a pruning method which terminates the learning process with a predefined criterion. In this process, an iterative model with 3 layer feedforward structure is derived from the incremental model by an appropriate manipulations. Notice that this network structure is not full-connected between upper and lower layers. To verify the effectiveness of pruning method, this network is retrained by EBP. From this results, we can find out that the proposed algorithm is effective, as an aspect of a system performence and the node number included in network structure.

• Journal of Korean Association for Spatial Structures
• /
• v.22 no.2
• /
• pp.39-46
• /
• 2022

Recently, machine learning is widely used to solve optimization problems in various engineering fields. In this study, machine learning is applied to development of a control algorithm for a smart control device for reduction of seismic responses. For this purpose, Deep Q-network (DQN) out of reinforcement learning algorithms was employed to develop control algorithm. A single degree of freedom (SDOF) structure with a smart tuned mass damper (TMD) was used as an example structure. A smart TMD system was composed of MR (magnetorheological) damper instead of passive damper. Reward design of reinforcement learning mainly affects the control performance of the smart TMD. Various hyper-parameters were investigated to optimize the control performance of DQN-based control algorithm. Usually, decrease of the time step for numerical simulation is desirable to increase the accuracy of simulation results. However, the numerical simulation results presented that decrease of the time step for reward calculation might decrease the control performance of DQN-based control algorithm. Therefore, a proper time step for reward calculation should be selected in a DQN training process.

• Transactions of the Korean Society of Pressure Vessels and Piping
• /
• v.19 no.2
• /
• pp.93-101
• /
• 2023

In this study, a deep learning algorithm was used to diagnose electric potential signals obtained through CIPS and DCVG, used indirect inspection methods to confirm the soundness of buried pipes. The deep learning algorithm consisted of CNN(Convolutional Neural Network) model for diagnosing the electric potential signal and Grad CAM(Gradient-weighted Class Activation Mapping) for showing the flaw prediction point. The CNN model for diagnosing electric potential signals classifies input data as normal/abnormal according to the presence or absence of flaw in the buried pipe, and for abnormal data, Grad CAM generates a heat map that visualizes the flaw prediction part of the buried pipe. The CIPS/DCVG signal and piping layout obtained from the 3D finite element model were used as input data for learning the CNN. The trained CNN classified the normal/abnormal data with 93% accuracy, and the Grad-CAM predicted flaws point with an average error of 2m. As a result, it confirmed that the electric potential signal of buried pipe can be diagnosed using a CNN-based deep learning algorithm.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.14 no.2
• /
• pp.216-221
• /
• 2004

This paper describes an optimal synthesis method of binary neural network for pattern recognition. Our objective is to minimize the number of connections and the number of neurons in hidden layer by using a Newly Expanded and Truncated Learning Algorithm (NETLA) for the multilayered neural networks. The synthesis method in NETLA uses the Expanded Sum of Product (ESP) of the boolean expressions and is based on the multilayer perceptron. It has an ability to optimize a given binary neural network in the binary space without any iterative learning as the conventional Error Back Propagation (EBP) algorithm. Furthermore, NETLA can reduce the number of the required neurons in hidden layer and the number of connections. Therefore, this learning algorithm can speed up training for the pattern recognition problems. The superiority of NETLA to other learning algorithms is demonstrated by an practical application to the approximation problem of a circular region.

• Proceedings of the IEEK Conference
• /
• 1999.06a
• /
• pp.1017-1020
• /
• 1999

Recently, a new learning algorithm for multilayer neural networks has been proposed 〔1〕. In the new learning algorithm, each output neuron is considered as a function of weights and the weights are adjusted so that the output neurons produce desired outputs. And the adjustment is accomplished by taking gradients. However, the gradient computation was performed numerically, resulting in a long computation time. In this paper, we derive the all necessary equations so that the gradient computation is performed analytically, resulting in a much faster learning time comparable to the backpropagation. Since the weight adjustments are accomplished by summing the gradients of the output neurons, we will call the new learning algorithm “multi-gradient.” Experiments show that the multi-gradient consistently outperforms the backpropagation.

• Journal of information and communication convergence engineering
• /
• v.1 no.3
• /
• pp.157-162
• /
• 2003

We propose a learning method for improving the performance of the backpropagation algorithm. The proposed method is using a fuzzy logic system for automatic tuning of the learning rate of each weight. Instead of choosing a fixed learning rate, the fuzzy logic system is used to dynamically adjust the learning rate. The inputs of fuzzy logic system are delta and delta bar, and the output of fuzzy logic system is the learning rate. In order to verify the effectiveness of the proposed method, we performed simulations on the XOR problem, character classification, and function approximation. The results show that the proposed method considerably improves the performance compared to the general backpropagation, the backpropagation with momentum, and the Jacobs'delta-bar-delta algorithm.

• 제어로봇시스템학회:학술대회논문집
• /
• 1989.10a
• /
• pp.1069-1074
• /
• 1989

A discrete-time learning control for robotic manipulators is studied using its pulse transfer function. Firstly, discrete-time learning stability condition which is applicable to single-input two-outputs systems is derived. Secondly, stability of learning algorithm with position signal is studied. In this case, when sampling period is small, the algorithm is not stable because of an unstable zero of the system. Thirdly, stability of algorithm with position and velocity signals is studied. In this case, we can stabilize the learning control system which is unstable in learning with only position signal. Finally, simulation results on the trajectory control of robotic manipulators using the discrete-time learning control are shown. This simulation results agree well with the analytical ones.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 1997.11a
• /
• pp.120-123
• /
• 1997

In this paper, we propose the Fuzzy Inference-based Reinforcement Learning Algorithm. We offer more similar learning scheme to the psychological learning of the higher animal's including human, by using Fuzzy Inference in Reinforcement Learning. The proposed method follows the way linguistic and conceptional expression have an effect on human's behavior by reasoning reinforcement based on fuzzy rule. The intervals of fuzzy membership functions are found optimally by genetic algorithms. And using Recurrent state is considered to make an action in dynamical environment. We show the validity of the proposed learning algorithm by applying to the inverted pendulum control problem.