• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.22 no.9
• /
• pp.1165-1171
• /
• 2018

CNN requires a large amount of computation and memory in the process of extracting the feature of the object. Also, It is trained from the network that the user has configured, and because the structure of the network is fixed, it can not be modified during training and it is also difficult to use it in a mobile device with low computing power. To solve these problems, we apply a pruning method to the pre-trained weight file to reduce computation and memory requirements. This method consists of three steps. First, all the weights of the pre-trained network file are retrieved for each layer. Second, take an absolute value for the weight of each layer and obtain the average. After setting the average to a threshold, remove the weight below the threshold. Finally, the network file applied the pruning method is re-trained. We experimented with LeNet-5 and AlexNet, achieved 31x on LeNet-5 and 12x on AlexNet.

• KIPS Transactions on Computer and Communication Systems
• /
• v.1 no.1
• /
• pp.11-20
• /
• 2012

Stream data shows a sequence of values changing continuously over time. Due to the nature of stream data, its trend is continuously changing according to various time intervals. Therefore the prediction of stream data must be carried out simultaneously with respect to multiple intervals, i.e. Continuous Multiple Prediction(CMP). In this paper, we propose a Continuous Integrated Hierarchical Temporal Memory (CIHTM) network for CMP based on the Hierarchical Temporal Memory (HTM) model which is a neocortex leraning algorithm. To develop the CIHTM network, we created three kinds of new modules: Shift Vector Senor, Spatio-Temporal Classifier and Multiple Integrator. And also we developed learning and inferencing algorithm of CIHTM network.

• Proceedings of the Korean Information Science Society Conference
• /
• 2006.06b
• /
• pp.232-234
• /
• 2006

본 논문에서는 마르코프 결정 문제 (Markov decision problem)의 풀이 효율을 잴 수 있는 척도를 알아보기 위해 복잡계 네트워크 (complex network) 의 관점에서 MDP를 하나의 그래프로 나타내고, 그 그래프의 위상학적 성질들을 여러 네트워크 척도 (network measurements)들을 이용하여 측정하고 그 MDP의 풀이 효율과의 관계를 분석하였다. 실세계의 여러 문제들이 MDP로 표현될 수 있고, 모델이 알려진 경우에는 평가치 반복(value iteration)이나 모델이 알려지지 않은 경우에도 강화 학습(reinforcement learning) 알고리즘등을 사용하여 풀 수 있으나, 이들 알고리즘들은 시간 복잡도가 높아 크기가 큰 실세계 문제에 적용하기 쉽지 않다. 이 문제를 해결하기 위해 제안된 것이 MDP를 계층적으로 분할하거나, 여러 단계를 묶어서 수행하는 등의 시간적 추상화(temporal abstraction) 방법들이다. 시간적 추상화를 도입할 경우 MDP가 보다 효율적으로 풀리는 꼴로 바뀐다는 사실에 착안하여, MDP의 풀이 효율을 네트워크 척도를 이용하여 측정할 수 있는 여러 위상학적 성질들을 기반으로 분석하였다. 다양한 구조와 파라미터를 가진 MDP들을 사용해 네트워크 척도들과 MDP의 풀이 효율간의 관계를 분석해 본 결과, 네트워크 척도들 중 평균 측지 거리 (mean geodesic distance) 가 그 MDP의 풀이 효율을 결정하는 가장 중요한 기준이라는 사실을 알 수 있었다.

• Journal of KIISE:Information Networking
• /
• v.31 no.4
• /
• pp.375-383
• /
• 2004

As wireless mobile networks have been widely adopted due to their convenience for deployment, the research for improving their performance has been actively conducted. Since their throughput is restrained by the packet corruption rate not by congestion as in wired networks, however, network simulations for performance evaluation need to select the appropriate wireless channel model representing the behavior of propagation errors for the evaluated channel. The selection of the right model should depend on various factors such as the adopted frequency band, the level of signal power, the existence of obstacles against signal propagation, the sensitivity of protocols to bit errors, and etc. This paper analyzes 10-day bit traces collected from real sensor channels exhibiting the high bit error rate to determine a suitable sensor channel model. For selection, it also evaluates the performance of two error recovery algorithms such as a link layer FEC algorithm and three TCPs (Tahoe, Reno, and Vegas) over several channel models. The comparison analysis shows that CM(Chaotic Map) model predicts 3-time less BER variance and 10-time larger PER(Packet Error Rate) than traces while these differences between the other models and traces are larger than 10-time. The simulation experiments, furthermore, prove that CM model evaluates the performance of these algorithms over sensor channels with the precision at least 10-time more accurate than any other models.

• Journal of Intelligence and Information Systems
• /
• v.28 no.3
• /
• pp.185-207
• /
• 2022

Recently, with the development of deep learning technology, researches to apply a deep learning algorithm to analyze unstructured data such as text and images are being actively conducted. Text classification has been studied for a long time in academia and industry, and various attempts are being performed to utilize data characteristics to improve classification performance. In particular, a hierarchical relationship of labels has been utilized for hierarchical classification. However, the top-down approach mainly used for hierarchical classification has a limitation that misclassification at a higher level blocks the opportunity for correct classification at a lower level. Therefore, in this study, we propose a methodology for classifying hierarchical data using the autoencoder-based deeply supervised network that high-level classification does not block the low-level classification while considering the hierarchical relationship of labels. The proposed methodology adds a main classifier that predicts a low-level label to the autoencoder's latent variable and an auxiliary classifier that predicts a high-level label to the hidden layer of the autoencoder. As a result of experiments on 22,512 academic papers to evaluate the performance of the proposed methodology, it was confirmed that the proposed model showed superior classification accuracy and F1-score compared to the traditional supervised autoencoder and DNN model.

• The KIPS Transactions:PartD
• /
• v.11D no.4
• /
• pp.975-982
• /
• 2004

In current P2P concept, pure P2P and Hybrid P2P structures are used commonly. Gnutella and Ktella are forms of pure P2P. and forms of Hybrid P2P are innumerable. File searching models exist in these models. These models provide group management for file sharing, searching and indexing. The general file sharing model is good at maintaining connectivity. However, it is defective in group management. Therefore, this study approaches hierarchical structure in file sharing models through routing technique and backup system. This system was designed so that the user was able to maintain group efficiency and connection reliability in large-scale network.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2006.05a
• /
• pp.749-752
• /
• 2006

최근 P2P (Peer-to-Peer) 시스템은 인터넷의 사용량 증가와 네트워크 속도의 증가, 개인 PC 의 성능 향상과 같은 외부적인 요인과 기존의 클라이언트 서버와 비교해 보다 효율적으로 다양한 자원을 공유할 수 있다는 내부적인 장점으로 인해 관심이 증가되고 있다. 초기 P2p 시스템은 냅스터와 같은 중앙 집중형 기반에서 JXTA 와 같은 순수 모델로 변화 되었으며, 최근 두 가지의 장점을 결합한 수퍼 피어 기반의 계층형 시스템이 연구되고 있다. 본 논문에서는 피어를 수퍼 피어와 일반 피어로 분류하는 2 계층 P2P 시스템에 대해 연구하였다. 제안한 시스템은 일반 피어가 동적인 성능과 유사도를 기반으로 최적의 수퍼 피어를 선택하도록 한다. 일반 피어는 가장 적합한 수퍼 피어를 선택함으로서 보다 효율적으로 쿼리를 처리할 수 있으며 일반 피어가 요구하는 서비스와 유사한 서비스를 제공함으로써 일반 피어의 만족도를 향상 시킬 수 있다.

• Electronics and Telecommunications Trends
• /
• v.27 no.4
• /
• pp.40-53
• /
• 2012

통신은 기본적으로 사람과 사람 간의 의사소통에서 출발한다. 그러나 만물지능통신은 통신의 대상을 사람-사물-공간-시스템으로 확장한 초연결(hyperconnectivity) 네트워킹을 전제로 한다. 본고에서는 이러한 만물지능통신 기반의 초연결 산업 구도를 분석하기 위하여, 스마트 혁명 이후의 IT 산업 분석틀로 부상한 CPNT(Content, Platform, Network, Terminal) 계층구조를 적용함과 동시에 초연결 산업의 준거틀로 일본 경제산업성 기술전략맵에 포함된 미래 사회의 삽화 내용을 원용했다. 동 기술전략맵에서는 2025년을 실현연도로 상정하여, 기술이 개발된 미래 생활환경을 삽화를 통하여 기술하고 있다. 동 삽화의 구도와 내용을 분석한 결과 초연결 산업은 사람-사물-공간-시스템 간의 초연결로 특징지을 수 있고, 콘텐츠-플랫폼-단말-네트워크 계층구조로 재구성할 수 있음을 확인할 수 있었다. 이러한 작업을 통하여 초연결 산업 생태계의 기본 구도로서 천지인(天地人) 모델을 제안하였다.

• Journal of KIISE:Information Networking
• /
• v.33 no.2
• /
• pp.175-182
• /
• 2006

The CGA proposed by IETF working group prevents address spoofing and stealing and provides digital signature to users, but key collision problem arises. To solve this critical problem, the CGA defines the SEC field within address format, which is set to high value when high security is required and vice versa, but the CGA faces a dilemma between security and the processing time. As SEC value increases, the processing time to generate the CGA grows dramatically while key collision ratio increases if low SEC value is applied to the CGA. We propose modified CGA (MCGA) that has shorter processing time than the CGA and offers digital signature with small overheads. To solve key collision problem, we employ hierarchical ad hoc network. The MCGA is applicable to IPv6 networks as well public networks. In this paper, we design a mathematical model to analyze the processing time for MCGA and CGA first and evaluate the processing time via simulations, where the processing time for MCGA is reduced down 3.3 times when SEC value is set to 0 and 68,000 times when SEC value is set to 1. Further, we have proved that the CGA is inappropriate for both ad hoc networks and IPv6 networks when the SEC field is set to more than 3.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.13 no.6
• /
• pp.643-648
• /
• 2003

In order to provide efficient retrieving results for user query on the web environment, the various searching algorithms have developed and considered user's preference and convenience. However, the searching algorithms are developed on the horizontal and non hierarchical web environment in general and could not apply to the complex hierarchical and functional web environments such like the enterprise network. In this paper, we purpose the multi-agent based web mining system which can provide the efficient mining results to the user on the special web environment. For doing this, we suggest the network model with the hierarchical web environment and model the multi agent based web mining system which has four corporation agents and fourteen process modules. Then, we explain the detailed functions of each agent considered the hierarchical environment according to the module. Especially, we purpose the new merging agent and improved ranking algorithm by using the graph theory.