• 제어로봇시스템학회:학술대회논문집
• /
• 2000.10a
• /
• pp.484-484
• /
• 2000

The problem of determining the proper size of an neural network is recognized to be crucial, especially for its practical implications in such important issues as learning and generalization. Unfortunately, it usually is not obvious what size is best: a system that is too snail will not be able to learn the data while one that is just big enough may learn the slowly and be very sensitive to initial conditions and learning parameters. One popular technique is commonly known as pruning and consists of training a larger than necessary network and then removing unnecessary weights/nodes. In this paper, a new pruning method is developed, based on the penalty-term methods. This method makes the neural network good for the generalization and reduces the retraining time after pruning weights/nodes.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2015.04a
• /
• pp.143-146
• /
• 2015

Software Defined Network (SDN) separates control plane and data plane to achieve benefits such as centralized management, centralized provisioning, lower device cost and more flexibility. In SDN, scalability is an important issue. Centralized controller can be a bottle neck and many research tried to solve this issue on the control plan. However, scalability issue does not only happen in the control plane, but also happen in the data plane. In the data plane, flow table is an important component and its size is limited. In a large network operated by SDN technology, the performance of the network can be highly degraded because of the size limitation of a flow table. In this paper, we propose a ranking-based flow replacement method, Flow Table Management (FTM), to overcome this problem.

• Structural Engineering and Mechanics
• /
• v.73 no.4
• /
• pp.463-479
• /
• 2020

This paper is aimed to address a simultaneous optimization of the size, shape, and topology of steel lattice towers through a combination of the radial basis function (RBF) neural networks and the artificial bee colony (ABC) metaheuristic algorithm to reduce the computational time because mere metaheuristic optimization algorithms require much time for calculations. To verify the results, use has been made of the CIGRE Tower and a 132 kV transmission towers as numerical examples both based on the design requirements of the ASCE10-97, and the size, shape, and topology have been optimized (in both cases) once by the RBF neural network and once by the MSTOWER analyzer. A comparison of the results shows that the neural network-based method has been able to yield acceptable results through much less computational time.

• The Journal of the Korea Contents Association
• /
• v.12 no.9
• /
• pp.348-356
• /
• 2012

This study is identifying early takeoff phenomenon in relational network like SNS and investigates influence of innovator or hub at early takeoff and diffusion size at takeoff point. The major finding of this study are as follows. First, multinominal-logit analysis conducted to explore the impact of innovator and hub at early stage of information diffusion and early takeoff. Hub in early stage of information diffusion impact early takeoff. Second, our study shows that innovator and hub at early stage of information diffusion impact diffusion size at takeoff point. Third, hub is more impact early takeoff and diffusion size at takeoff point then innovator. This study provides implication for professionals by confirming hub's influence in study of successful information diffusion by early takeoff. The limitation and future direction are discussed.

• Journal of IKEEE
• /
• v.25 no.1
• /
• pp.180-192
• /
• 2021

One widely used algorithm for image recognition and pattern detection is the convolution neural network (CNN). To efficiently handle convolution operations, which account for the majority of computations in the CNN, we use hardware accelerators to improve the performance of CNN applications. In using these hardware accelerators, the CNN fetches data from the off-chip DRAM, as the massive computational volume of data makes it difficult to derive performance improvements only from memory inside the hardware accelerator. In other words, data communication between off-chip DRAM and memory inside the accelerator has a significant impact on the performance of CNN applications. In this paper, a simulator for the CNN is developed to analyze the main memory or DRAM access with respect to the size of the on-chip memory or global buffer inside the CNN accelerator. For AlexNet, one of the CNN architectures, when simulated with increasing the size of the global buffer, we found that the global buffer of size larger than 100kB has 0.8x as low a DRAM access count as the global buffer of size smaller than 100kB.

• Asia pacific journal of information systems
• /
• v.18 no.3
• /
• pp.21-43
• /
• 2008

The task network which is formed of different individuals can be recognized as a social network. Therefore, the way to communicate with people inside or outside the network has considerable influence on their outcome. Moreover, the position on which a member stands in a network shows the different effects of the information systems supporting communication with others. In this paper, it is to be studied how personal CMC (computer-mediated communication) tools affect the mission that those who work for a network perform through diverse task networks. Especially, we focused on synchronicity of CMC. On this score, the perspective of Media Synchronicity Theory was taken that had been suggested by criticizing Media Richness Theory. It is the objective, from this perspective, to find which characteristics of networks make the value of IT supporting synchronicity high. In the research trends of social networks, there have been two traditional perspectives to explain the effect of network: embeddedness and diversity ones. These differ from the aspect which type of social network can provide much more economic benefits. As similar studies have been reported by various researchers, these are also divided into the bonding and bridging views which are based on internal and external tie, respectively, Size, density, and centrality were measured as the characteristics of personal task networks. Size means the level of relationship between members. It is the total number of other colleagues who work with a specific member for a certain project. It means, the larger the size of task network, the more the number of coworkers who interact each other through the job. Density is the ratio of the number of relationships arranged actually to the total number of available ones. In an ego-centered network, it is defined as the ratio of the number of relationship made really to the total number of possible ones between members who are actually involved each other. The higher the level of density, the larger the number of projects on which the members collaborate. Centrality means that his/her position is on the exact center of whole network. There are several methods to measure it. In this research, betweenness centrality was adopted among them. It is measured by the position on which one member stands between others in a network. The determinant to raise its level is the shortest geodesic that represents the shortest distance between members. Centrality also indicates the level of role as a broker among others. To verify the hypotheses, we interviewed and surveyed a group of employees of a nationwide financial organization in which a groupware system is used. They were questioned about two CMC applications: MSN with a higher level of synchronicity and email with a lower one. As a result, the larger the size of his/her own task network, the smaller its density and the higher the level of his/her centrality, the higher the level of the effect using the task network with CMC tools. Above all, this positive effect is verified to be much more produced while using CMC applications with higher-level synchronicity. Among the a variety of situations under which the use of CMC gives more benefits, this research is considered as one of rare cases regarding the characteristics of task network as moderators by focusing ITs for the operation of his/her own task network. It is another contribution of this research to prove empirically that the values of information system depend on the social, or comparative, characteristic of time. Though the same amount of time is shared, the social characteristics of users change its value. In addition, it is significant to examine empirically that the ITs with higher-level synchronicity have the positive effect on productivity. Many businesses are worried about the negative effect of synchronous ITs, for their employees are likely to use them for personal social activities. However. this research can help to dismiss the concern against CMC tools.

• Journal of Information Processing Systems
• /
• v.2 no.1
• /
• pp.34-38
• /
• 2006

In the wireless network, TCP performs poorly because it was originally designed for wired networks and does not take into consideration wireless characteristics such as mobility, high-loss probability, and hidden-terminal problems. In particular, in the wireless multi-hop networks, a large congestion window increases the probability of contention and packet losses, and TCP performance is degraded severely as a result. So, it is necessary to limit the TCP congestion window size in order keep the probability of contention loss in the system to a minimum. In this paper, we propose a new scheme for determining the maximum congestion window size based on the measured bandwidth and Round-Trip-Time (RTT). Using ns-2 simulation, we show that the proposed scheme reduces the probability of packet contention and improves TCP performance.

• Journal of Power Electronics
• /
• v.15 no.6
• /
• pp.1619-1627
• /
• 2015

This paper presents a power system harmonic elimination using the mixed adaptive linear neural network and variable step-size leaky least mean square (ADALINE-VSSLLMS) control algorithm based active power filter (APF). The weight vector of ADALINE along with the variable step-size parameter and leakage coefficient of the VSSLLMS algorithm are automatically adjusted to eliminate harmonics from the distorted load current. For all iteration, the VSSLLMS algorithm selects a new rate of convergence for searching and runs the computations. The adopted shunt-hybrid APF (SHAPF) consists of an APF and a series of 7^th tuned passive filter connected to each phase. The performance of the proposed ADALINE-VSSLLMS control algorithm employed for SHAPF is analyzed through a simulation in a MATLAB/Simulink environment. Experimental results of a real-time prototype validate the efficacy of the proposed control algorithm.

• Environmental Engineering Research
• /
• v.14 no.2
• /
• pp.102-110
• /
• 2009

Application of artificial intelligence (AI) approaches in eco-environmental modeling has gradually increased for the last decade. Comprehensive understanding and evaluation on the applicability of this approach to eco-environmental modeling are needed. In this study, we reviewed the previous studies that used AI-techniques in eco-environmental modeling. Decision Tree (DT) and Artificial Neural Network (ANN) were found to be major AI algorithms preferred by researchers in ecological and environmental modeling areas. When the effect of the size of training data on model prediction accuracy was explored using the data from the previous studies, the prediction accuracy and the size of training data showed nonlinear correlation, which was best-described by hyperbolic saturation function among the tested nonlinear functions including power and logarithmic functions. The hyperbolic saturation equations were proposed to be used as a guideline for optimizing the size of training data set, which is critically important in designing the field experiments required for training AI-based eco-environmental modeling.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.9 no.8
• /
• pp.3169-3181
• /
• 2015

Nowadays, online word-of-mouth has become a powerful influencer to marketing and sales in business. Opinion mining and sentiment analysis is frequently adopted at market research and business analytics field for analyzing word-of-mouth content. However, there still remain several challengeable areas for 1) sentiment analysis aiming for Korean word-of-mouth content in film market, 2) availability of machine learning models only using linguistic features, 3) effect of the size of the feature set. This study took a sample of 10,000 movie reviews which had posted extremely negative/positive rating in a movie portal site, and conducted sentiment analysis with four machine learning algorithms: naïve Bayesian, decision tree, neural network, and support vector machines. We found neural network and support vector machine produced better accuracy than naïve Bayesian and decision tree on every size of the feature set. Besides, the performance of them was boosting with increasing of the feature set size.