• IEMEK Journal of Embedded Systems and Applications
• /
• v.16 no.1
• /
• pp.9-15
• /
• 2021

Recently, it is increasingly necessary to run high-end neural network applications with huge computation overhead on top of resource-constrained embedded systems, such as wearable devices. While the huge computational overhead can be alleviated by distributed neural networks running on multiple separate devices, existing distributed neural network techniques suffer from a large traffic between the devices; thus are very vulnerable to communication failures. These drawbacks make the distributed neural network techniques inapplicable to wearable devices, which are connected with each other through unstable and low data rate communication medium like human body communication. Therefore, in this paper, we propose a distributed neural network partitioning technique that is resilient to communication failures. Furthermore, we show that the proposed technique also improves the inference accuracy even in case of no communication failure, thanks to the improved network partitioning. We verify through comparative experiments with a real-life neural network application that the proposed technique outperforms the existing state-of-the-art distributed neural network technique in terms of accuracy and resiliency to communication failures.

• Parasites, Hosts and Diseases
• /
• v.54 no.4
• /
• pp.503-507
• /
• 2016

The genus Spirometra belongs to the family Diphyllobothriidae and order Pseudophyllidea, and includes intestinal parasites of cats and dogs. In this study, a plerocercoid labeled as Spirometra mansonoides from the USA was examined for species identification and phylogenetic analysis using 2 complete mitochondrial genes, cytochrome c oxidase I (cox1) and NADH dehydrogenase subunit 3 (nad3). The cox1 sequences (1,566 bp) of the plerocercoid specimen (USA) showed 99.2% similarity to the reference sequences of the plerocercoid of Korean Spirometra decipiens (GenBank no. KJ599679), and 99.1% similarity in regard to nad3 (346 bp). Phylogenetic tree topologies generated using 4 analytical methods were identical and showed high confidence levels with bootstrap values of 1.00, 100%, 100%, and 100% for Bayesian inference (BI), maximum-likelihood (ML), neighbor-joining (NJ), and maximum parsimony (MP) methods, respectively. Representatives of Diphyllobothrium and Spirometra species formed a monophyletic group, and the sister-genera status between these species was well supported. Trapezoic proglottids in the posterior 1/5 region of an adult worm obtained from an experimentally infected cat were morphologically examined. The outer uterine loop of the uterus coiling characteristically consisted of 2 complete turns. The results clearly indicated that the examined Spirometra specimen from the USA matched to S. decipiens very well, and indicated possible presence of the life cycle of this species in this region.

• Journal of the Korean Data and Information Science Society
• /
• v.26 no.5
• /
• pp.1129-1139
• /
• 2015

Inference process generates additional triples from knowledge represented in RDF triples of semantic web technology. Tens of million of triples as an initial big data and the additionally inferred triples become a knowledge base for applications such as QA(question&answer) system. The inference engine requires more computing resources to process the triples generated while inferencing. The additional computing resources supplied by underlying resource pool in cloud computing can shorten the execution time. This paper addresses an algorithm to allocate the number of computing nodes "elastically" at runtime on Hadoop, depending on the size of knowledge data fed. The model proposed in this paper is composed of the layered architecture: the top layer for applications, the middle layer for distributed parallel inference engine to process the triples, and lower layer for elastic Hadoop and server visualization. System algorithms and test data are analyzed and discussed in this paper. The model hast the benefit that rich legacy Hadoop applications can be run faster on this system without any modification.

• Parasites, Hosts and Diseases
• /
• v.57 no.5
• /
• pp.481-487
• /
• 2019

Mitochondrial DNA sequence variability of Spirometra erinaceieuropaei in GenBank was observed by reinvestigation of mitochondrial cox1 and cytb sequences. The DNA sequences were analyzed in this study, comprising complete DNA sequences of cox1 (n=239) and cytb (n=213) genes. The 10 complete mitochondrial DNA sequences of Spirometra species were compared with those of Korea, China and Japan. The sequences were analyzed for nucleotide composition, conserved sites, variable sites, singleton sites and parsimony-informative sites. Phylogenetic analyses was done using neighbor joining, maximum parsimony, Bayesian inference and maximum-likelihood on cox1 and cytb sequences of Spirometra species. These polymorphic sites identified 148 (cox1) and 83 (cytb) haplotypes within 239 and 213 isolates from 3 Asian countries. Phylogenetic tree topologies were presented high-level confidence values for the 2 major branches of 2 Spirometra species containing S. erinaceieuropaei and S. decipiens, and S. decipiens sub-clades including all sequences registered as S. erinaceieuropaei in cox1 and cytb genes. These results indicated that mitochondrial haplotypes of S. erinaceieuropaei and S. decipiens were found in the 3 Asian countries.

• Journal of Intelligence and Information Systems
• /
• v.28 no.3
• /
• pp.45-55
• /
• 2022

Recently, large-scale language models (LPLM) have been shown state-of-the-art performances in various tasks of natural language processing including intent classification. However, fine-tuning LPLM requires much computational cost for training and inference which is not appropriate for dialog system. In this paper, we propose compressed intent classification model for multi-agent in low-resource like CPU. Our method consists of two stages. First, we trained sentence encoder from LPLM then compressed it through knowledge distillation. Second, we trained agent-specific adapter for intent classification. The results of three intent classification datasets show that our method achieved 98% of the accuracy of LPLM with only 21% size of it.

• Proceedings of the Korean Information Science Society Conference
• /
• 2005.11a
• /
• pp.829-831
• /
• 2005

최근 컴포넌트 기반 미들웨어의 최적화에 대한 연구가 활발히 이루어지고 있다. CPU점유율이 높은 어플리케이션의 동시 실행은 시스템에 부하를 주기 때문에, 시스템 성능을 약화시키고 실행중인 어플리케이션에 영향을 준다. 컴포넌트 기반 미들웨어는 여러 개의 재사용 가능한 컴포넌트를 조합하여 어플리케이션을 구성하기 때문에 동적으로 재구성이 가능하다. 본 논문은 컴포넌트 기반 미들웨어가 시스템 상황에 대한 정보를 받아들여 시스템의 상황을 스스로 판단하고 자가치료 또는 시스템의 성능을 최적화시키는 컴포넌트를 선택하는 방법을 제안한다. 상황판단을 위해 유연한 추론이 가능하고, 데이터로부터 자동학습이 가능한 베이지안 네트워크를 사용하였다. 두 시간 가량의 데이터를 리눅스 사용자로부터 획득하여 실험한 결과, 테스트 데이터에 대해 $76.5\%$의 성능을 보였다.

• Proceedings of the Korean Information Science Society Conference
• /
• 2003.10c
• /
• pp.559-561
• /
• 2003

최근 차세대 Web 기술로서 Semantic Web이 주목 받고 있다. Semantic Web에서는 Web상에 존재하는 문서에 Web Resource들에 대한 Ontology를 기반으로 Semantic Annotation을 하고 Ontology 추론 Agent를 통하여 의미 기반으로 Web을 검색할 수 있도록 해준다. 이와 같은 Semantic Web 기술의 핵심 요소는 Web Ontology이며 W3C에서는 이를 표현 할 수 있는 표준 언어로서 RDF기반의 OWL(Web Ontology Language) 명세를 제정하고 있다. 따라서 표준 Web Ontology 언어인 OWL을 위한 추론 시스템은 Semantic Web 검색 Agent의 구현을 위한 필수적인 기반 기술이라 할 수 있으나 아직 그 개발이 미비한 상태이다. OWL 추론 시스템을 구현하기 위해서는 OWL의 이론적인 기반을 제공하는 DL(Description Logic)을 추론할 수 있는 엔진을 사용하는 것이 한가지 방법이 될 수 있으나 OWL이 Rule과 같은 DL의 범주를 벗어나는 Vocabulary를 지원하는 언어로 확장되는 경우에 이를 처리하기가 어렵다. 또 다른 방법으로서 Logic Programming을 통하여 OWL 언어의 Semantic을 기술하고 정리 증명(Theorem Proving)을 통하여 Ontology를 추론하는 공리화(Axiomatisation) 기법이 있는데 이러한 방법의 장점은 기반이 되는 Logic의 범주 내에서 새로운 언어를 위한 Vocabulary의 확장이 용이하다는 점이다. 본 논문에서는 Axiomatisation 방법을 이용하여 OWL로 기술된 Ontology를 추론할 수 있는 시스템의 설계 및 구현에 대해 설명하기로 한다.

• Journal of Korea Society of Digital Industry and Information Management
• /
• v.9 no.2
• /
• pp.73-79
• /
• 2013

Cloud computing is a form which provides IT resources through network and pays the cost as much as you used. And it has advantages that it doesn't need to construct infrastructure and can be offered a variety of environments. The main core of these computing is virtualization technology. Security mechanism about attacks using vulnerabilities of virtualization technology isn't provided right and existing security tools can't be applied as it is. In this paper, we proposed honeyVM structure that can cope actively by collecting the information about attacks using virtualization vulnerability. Mamdani fuzzy inference is used to adjust dynamically the number of formed honeyVM depending on the load of system. Security structure to protect actual virtual machine from attacks and threats is proposed. The performance of the proposed structure in this paper measured occurred attack detection rate and resource utilization rate.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 2006.11a
• /
• pp.314-317
• /
• 2006

This study proposes an ontology design framework to support the cooperation among devices by using PROLOG, Conceptual Graph (CG), and Resource Description Framework (RDF). Quite a large number of representation languages for representing ontology on the Web have been established over the last decade. Most of these researches are focused on design of independent resources description. In Semantic Web, however, cooperation ontology will be needed. In this study, the CG could make an entire conceptual view of knowledge and RDF can represent that knowledge. Then the PROLOG could support the natural inference based on that knowledge. Therefore, our proposed ontology will be used in the designing of Semantic Web-based cooperation systems.

• Asia Marketing Journal
• /
• v.20 no.3
• /
• pp.63-81
• /
• 2018

This research examines how the announcement of a celebrity endorsement contract with a brand can influence consumers' judgment of its future growth in a market. Using both behavioral experiments and an event study analysis, we find that when the brand is perceived to be technology-focused (versus marketing-focused), the information about the celebrity contract can lead consumers to view the brand's future as less promising. Furthermore, we identify consumers' inferences about the quality of the brand as a psychological factor underlying this effect. This paper provides evidence that star-marketing can sometimes signal management's misjudgment about resource allocation for a brand, which unduly invests in celebrity marketing at the expense of quality improvement for the product.