• Journal of Aerospace System Engineering
• /
• v.18 no.1
• /
• pp.21-28
• /
• 2024

In this paper, two prediction models based on deep neural network that could predict cross-sectional stiffness of a rotor blade were proposed. Herein, we employed structural and material information of cross-section. In the case of a prediction model that used material properties as the input of the network, it was designed to predict the cross-sectional stiffness by considering elastic modulus of each cross-sectional member. In the case of the prediction model that used structural information as a network input, it was designed to predict the cross-sectional stiffness by considering the location and thickness of cross-sectional members as network input. Both prediction models based on a deep neural network were realized using data obtained by cross-sectional analysis with KSAC2D (Konkuk section analysis code - two-dimensional).

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 1997.10a
• /
• pp.111-114
• /
• 1997

Usually, Evolutionary Algorithms are considered more efficient for optimal system design, However, the performance of the system is determined by fitness function and system environment. In this paper, in order to overcome the limitation of the performance by this factor, we propose a co-evolutionary method that two populations constantly interact and coevolve. In this paper, we apply coevolution to neural network's evolving. So, one population is composed of the structure of neural networks and other population is composed of training patterns. The structure of neural networks evolve to optimal structure and, at the same time, training patterns coevolve to feature patterns. This method prevent the system from the limitation of the performance by random design of neural network structure and inadequate selection of training patterns. In this time neural networks are trained by evolution strategies that are able to apply to the unsupervised learning. And in the coding of neural networks, we propose the method to maintain nonredundancy and character preservingness that are essential factor of genetic coding. We show the validity and the effectiveness of the proposed scheme by applying it to the visual servoing of RV-M2 robot manipulators.

• Proceedings of the Korean Society for Cognitive Science Conference
• /
• 2000.05a
• /
• pp.322-328
• /
• 2000

This study discusses cognition as a computable mapping in cognitive system and relates G del's Incompleteness Theorems to the computability of cognition from a metamathematical perspective. Understanding cognition as a from of computation requires not only Turing machine models but also neural network models. In previous studies of computation by cognitive systems, it is remarkable to note how little serious attention has been given to the issue of computation by neural networks with respect to G del's Incompleteness Theorems. To address this problem, first, we introduce a definition of cognition and cognitive science. Second, we deal with G del's view of computability, incompleteness and speed-up theorems, and then we interpret G del's disjunction on the mind and the machine. Third, we discuss cognition as a Turing computable function and its relation to G del's incompleteness. Finally, we investigate cognition as a neural computable function and its relation to G del's incompleteness. The results show that a second-order representing system can be implemented by a finite recurrent neural network. Hence one cannot prove the consistency of such neural networks in terms of first-order theories. Neural computability, theoretically, is beyond the computational incompleteness of Turing machines. If cognition is a neural computable function, then G del's incompleteness result does not limit the compytational capability of cognition in humans or in artifacts.

• Journal of Information Technology Services
• /
• v.16 no.4
• /
• pp.149-160
• /
• 2017

Sentiment analysis is the process of determining whether a piece of document, text or conversation is positive, negative, neural or other emotion. Sentiment analysis has been applied for several real-world applications, such as chatbot. In the last five years, the practical use of the chatbot has been prevailing in many field of industry. In the chatbot applications, to recognize the user emotion, sentiment analysis must be performed in advance in order to understand the intent of speakers. The specific emotion is more than describing positive or negative sentences. In light of this context, we propose deep learning models for conducting multi-class sentiment analysis for identifying speaker's emotion which is categorized to be joy, fear, guilt, sad, shame, disgust, and anger. Thus, we develop convolutional neural network (CNN), long short term memory (LSTM), and multi-layer neural network models, as deep neural networks models, for detecting emotion in a sentence. In addition, word embedding process was also applied in our research. In our experiments, we have found that long short term memory (LSTM) model performs best compared to convolutional neural networks and multi-layer neural networks. Moreover, we also show the practical applicability of the deep learning models to the sentiment analysis for chatbot.

• Korean Journal of Computational Design and Engineering
• /
• v.2 no.2
• /
• pp.77-84
• /
• 1997

This paper is concerning the development of multiple neural networks system of problem domains where the complete input space can be decomposed into several different regions, and these are known prior to training neural networks. We will adopt oblique decision tree to represent the divided input space and sel ect an appropriate subnetworks, each of which is trained over a different region of input space. The overall architecture of multiple neural networks system, called the federated architecture, consists of a facilitator, normal subnetworks, and tile networks. The role of a facilitator is to choose the subnetwork that is suitable for the given input data using information obtained from decision tree. However, if input data is close enough to the boundaries of regions, there is a large possibility of selecting the invalid subnetwork due to the incorrect prediction of decision tree. When such a situation is encountered, the facilitator selects a tile network that is trained closely to the boundaries of partitioned input space, instead of a normal subnetwork. In this way, it is possible to reduce the large error of neural networks at zones close to borders of regions. The validation of our approach is examined and verified by applying the federated neural networks system to the configuration design of a midship structure.

• Journal of the Korean Geotechnical Society
• /
• v.18 no.5
• /
• pp.37-42
• /
• 2002

Artificial neural networks are efficient computing techniques that are widely used to solve complex problems in many fields. In this paper a liquefaction potential was estimated by using a back propagation neural network model applicated to cyclic triaxial test data, soil parameters and site investigation data. Training and testing of the network were based on a database of 43 cyclic triaxial test data from 00 sites. The neural networks are trained by modifying the weights of the neurons in response to the errors between the actual output values and the target output value. Training was done iteratively until the average sum squared errors over all the training patterns were minimized. This generally occurred after about 15,000 cycles of training. The accuracy from 72% to 98% was shown for the model equipped with two hidden layers and ten input variables. Important effective input variables have been identified as the NOC,$D_10$ and (N$_1$)$_60$. The study showed that the neural network model predicted a CSR(Cyclic shear stress Ratio) of silty-sand reasonably well. Analyzed results indicate that the neural-network model is more reliable than simplified method using N value of SPT.

• 제어로봇시스템학회:학술대회논문집
• /
• 2001.10a
• /
• pp.169.5-169
• /
• 2001

This study proposes a neural network for solving optimization problems such as the TSP (Travelling Salesman Problem), scheduling, and line balancing. The Hopfield network has been used for solving such problems, but it frequently gives abnormal solutions or non-optimal ones. Moreover, the Hopfield network takes much time especially in solving large size problems. To overcome such disadvantages, this study adopts nodes whose outputs changes with a fixed value at every evolution. The proposed network is applied to solving a TSP, finding the shortest path for visiting all the cities, each of which is visted only once. Here, the travelling path is reflected to the energy function of the network. The proposed network evolves to globally minimize the energy function, and a ...

• ETRI Journal
• /
• v.42 no.3
• /
• pp.366-375
• /
• 2020

The network traffic prediction of a smart substation is key in strengthening its system security protection. To improve the performance of its traffic prediction, in this paper, we propose an improved gravitational search algorithm (IGSA), then introduce the IGSA into a wavelet neural network (WNN), iteratively optimize the initial connection weighting, scalability factor, and shift factor, and establish a smart substation network traffic prediction model based on the IGSA-WNN. A comparative analysis of the experimental results shows that the performance of the IGSA-WNN-based prediction model further improves the convergence velocity and prediction accuracy, and that the proposed model solves the deficiency issues of the original WNN, such as slow convergence velocity and ease of falling into a locally optimal solution; thus, it is a better smart substation network traffic prediction model.

• Proceedings of the KSME Conference
• /
• 2001.11a
• /
• pp.667-672
• /
• 2001

An optimization approach is used to partition the field of view. A cost function is defined to represent the constraints on the solution, which is then mapped onto a two-dimensional Hopfield neural network for minimization. Each neuron in the network represents a possible match between a field of view and one or multiple objects. Partition is achieved by initializing each neuron that represents a possible match and then allowing the network to settle down into a stable state. The network uses the initial inputs and the compatibility measures between a field of view and one or multiple objects to find a stable state.

• Proceedings of the KIEE Conference
• /
• 1990.07a
• /
• pp.539-541
• /
• 1990

A new neural network model, based on the Hopfield's crossbar associative network, is presented and shown to be an effective tool for the NP-Complete problems. This model is applied to a class of layer assignment problems for VLSI routing. The results indicate that this modified Hopfield model improves stability and accuracy.