• Journal of Advanced Information Technology and Convergence
• /
• v.8 no.2
• /
• pp.119-127
• /
• 2018

Training a very large deep neural network can be painfully slow and prone to overfitting. Many researches have done for overcoming the problem. In this paper, a combination of early stopping and ADAM based deep neural network was presented. This form of deep network is useful for handling the big data because it automatically stop the training before overfitting occurs. Also generalization ability is better than pure deep neural network model.

• Journal of the Korean Institute of Electrical and Electronic Material Engineers
• /
• v.25 no.9
• /
• pp.723-726
• /
• 2012

A current intensity convertible CLD chip was fabricated using small and large FET cell configuration. Pinch-off current of 8.82 mA and 11.56 mA were obtained for small and large cell in the CLD chip, respectively. Constant current was fairly maintained until the breakdown voltage of 60 V. Measured knee voltage, $V_k$ were 3.8 V and 4.5 V for small and large cell, respectively. We configured current amplifying chip with parallel connection of each cells, by connecting 8 individual large cells in parallel network, 92.0 mA of current was obtained. The pinch-off constant current of CLD chip was varied very linearly with respect to the number of parallel connected cell.

• Journal of Communications and Networks
• /
• v.16 no.2
• /
• pp.202-208
• /
• 2014

Wireless network information services allow end systems to discover heterogeneous networks and spectrum available for secondary use at or near their current location, helping them to cope with increasing traffic and finite spectrum resources. We propose a unified architecture that allows end systems to find nearby base stations that are using either licensed, shared or unlicensed spectrum across multiple network operators. Our study evaluates the performance and scalability of spatial databases storing base station coverage area geometries. The measurement results indicate that the current spatial databases perform well even when the number of coverage areas is very large. A single logical spatial database would likely be able to satisfy the query load for a large national cellular network. We also observe that coarse geographic divisions can significantly improve query performance.

• Journal of Korea Multimedia Society
• /
• v.18 no.2
• /
• pp.189-198
• /
• 2015

Despite the fact that traffic engineering techniques have been comprehensively utilized in the past to enhance the performance of communication networks, the distinctive characteristics of Software Defined Networking (SDN) demand new traffic engineering techniques for better traffic control and management. Considering the behavior of traffic, large flows normally carry out transfers of large blocks of data and are naturally packet latency insensitive. However, small flows are often latency-sensitive. Without intelligent traffic engineering, these small flows may be blocked in the same queue behind megabytes of file transfer traffic. So it is very important to identify large flows for different applications. In the scope of this paper, we present an approach to detect large flows in real-time without even a short delay. After the detection of large flows, the next problem is how to control these large flows effectively and prevent network jam. In order to address this issue, we propose an approach in which when the controller is enabled, the large flow is mitigated the moment it hits the predefined threshold value in the control application. This real-time detection, marking, and controlling of large flows will assure an optimize usage of an overall network.

• Journal of the Korea Society of Computer and Information
• /
• v.23 no.9
• /
• pp.57-64
• /
• 2018

In recent years, a new approach to cyber security, called the moving target defense, has emerged as a potential solution to the challenge of static systems. In this paper, we design a protected server network with a large number of decoys to anonymize the protected servers that dynamically mutate their IP address and port numbers according to Hidden Tunnel Networking, which is a network-based moving target defense scheme. In the network, a protected server is one-to-one mapped to a decoy-bed that generates a number of decoys, and the decoys share the same IP address pool with the protected server. First, the protected server network supports mutating the IP address and port numbers of the protected server very frequently regardless of the number of decoys. Second, it provides independence of the decoy-bed configuration. Third, it allows the protected servers to freely change their IP address pool. Lastly, it can reduce the possibility that an attacker will reuse the discovered attributes of a protected server in previous scanning. We believe that applying Hidden Tunnel Networking to protected servers in the proposed network can significantly reduce the probability of the protected servers being identified and compromised by attackers through deploying a large number of decoys.

• The Journal of Information Systems
• /
• v.6 no.1
• /
• pp.223-256
• /
• 1997

Since the business scales of retailing companies become to be very large and the number of items dealt increases explosively, automation of inventory management becomes one of the most important issues to solve in retailing industry. In order to accomplish this automation of inventory management, there must be a great need to a method which can perform real-time decision making on inventory control in an automatic fashion, while communicating with inventory information systems like POS system and automatic warehousing system. But even in this circumstance, there are also many obstructions to such automation like varying demands, limited capacity of warehouse and exhibition room, need for strategic consideration on inventory control, etc., in a real sense. Due to these reasons, it seems very difficult that most large-scaled retailing companies get fully automated inventory management system. To overcome those difficulties and reflect them into inventory control, we propose a automated inventory control methodology for retailing industry based on neural network and policy model. Especially, policy model is devised to deal with dynamic varying demands and using this model, strategic goals on inventory can be considered into inventory control mechanism. Our proposed approach is implemented in workstation and its performance is also empirically verified also against to real case of one of the major retailing firm in Korea.

• Journal of Korea Society of Digital Industry and Information Management
• /
• v.9 no.4
• /
• pp.251-265
• /
• 2013

The emergence of social network services, is changing the foundation of human relationship formation and method of communication of individuals through sharing of free information. Social network service is a service to support or facilitate an on-line extension of off-line network among people by helping them to share personal profile. History of social network services very short. But users of the various layers is increasing rapidly and ripple effect social as a result is very large. The focus of existing research was mainly devoted to motivation of use and acceptance of social network services. Currently the use of SNS was maturing. Thus, in-depth research on the use pattern of SNS users is needed. The purpose of this study is that, for Facebook in social network services, to analyze the changes in the initial stage of use, medium-term, usage patterns at the current time. Results of the study by analyzing the characteristics of the change in the pattern of usage of user of Facebook, it can be used as basic materials for SNS researchers and service provider.

• 제어로봇시스템학회:학술대회논문집
• /
• 2005.06a
• /
• pp.1568-1570
• /
• 2005

The internet is already a part of life. It is very convenient and people can do almost everything with internet that should be done in real life. Along with the increase of the number of internet user, various network attacks through the internet have been increased as well. Also, Large-scale network attacks are a cause great concern for the computer security communication. These network attack becomes biggest threat could be down utility of network availability. Most of the techniques to detect and analyze abnormal traffic are statistic technique using mathematical modeling. It is difficult accurately to analyze abnormal traffic attack using mathematical modeling, but network simulation technique is possible to analyze and simulate under various network simulation environment with attack scenarios. This paper performs modeling and simulation under virtual network environment including $NGSS^{1}$ system to analyze abnormal traffic-flooding attack.

• International Journal of Computer Science & Network Security
• /
• v.24 no.1
• /
• pp.45-51
• /
• 2024

The subject of navigation has drawn a large interest in the last few years. The navigation within a city is to find the path between two points, source location and destination location. In many cities, solving the routing problem is very essential as to find the route between different locations (starting location (source) and an ending location (destination)) in a fast and efficient way. This paper considers streets with diagonal streets. Such streets pose a problem in determining the directions of the route to be followed. The paper presents a solution for the path planning using the reconfigurable mesh (R-Mesh). R-Mesh is a parallel platform that has very fast solutions to many problems and can be deployed in moving vehicles and moving robots. This paper presents a solution that is very fast in computing the routes.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 1996.10a
• /
• pp.253-256
• /
• 1996

Rainfall is one of the major and complicated elements of hydrologic system. Accurate prediction of rainfall is very important to mitigate storm damage. The neural network is a good model to be applied for the classification problem, large combinatorial optimization and nonlinear mapping. In this dissertation, rainfall predictions by the neural network theory were presented. A multi-layer neural network was constructed. The network learned continuous-valued input and output data. The network was used to predict rainfall. The online, multivariate, short term rainfall prediction is possible by means of the developed model. A multidimensional rainfall generation model is applied to Seoul metropolitan area in order to generate the 10-minute rainfall. Application of neural network to the generated rainfall shows good prediction. Also application of neural network to 1-hour real data in Seoul metropolitan area shows slightly good predictions.