위선암에서 matrix metalloproteinase 및 type Ⅳ collagen 의 발현양상:림프절 전이 및 예후와의 관계

정은선 가톨릭대학교 대학원 1998 국내석사

PET분해효소의 활성분석방법 개발에 관한 연구

박태서 서울대학교 대학원 2022 국내석사

Based on the petrochemical industry, various polymer materials have been synthesized since the 1950s. And today, the problem of plastic waste disposal is emerging all over the world. Among them, polyethylene terephthalate(PET) based on ester bonds occupies the largest proportion in plastic waste. Recently, starting with IsPETase, PET degradation enzymes such as leaf-branch compost cutinase(LCC) have been discovered in nature, and a method for decomposing PET waste is attracting attention. Furthermore, in terms of enzymatic engineering of these enzymes, studies to improve PET degradation-activity, which degrade PET to terephthalic acid(TPA), are also in the spotlight. However, to date, high throughput screening(HTS) for PET degradation-activity comparison of PET degradation enzyme mutants has not been reported properly. In this study, two activity analysis method and screening system based on Escherichia coli, which will be the basis for the development of HTS for the comparison of PET degradation-activity in the future, were attempted. First, TPA was synthesized with a fluorescent substance, 2-hydroxyterephthalate(HOTP), and then detected in vitro. It uses the principle that a hydroxyl radical combines with TPA to become HOTP. Hydroxyl radicals were generated under biological pH conditions using the Fe(II)-EDTA fenton reaction. To remove the autofluorescence reaction of LB media, a liquid extraction method using ethyl acetate(EA) was applied. A calibration curve with a positive correlation between TPA concentration and HOTP fluorescence intensity was obtained. The LOD value was measured to be 316.85 uM. To confirm the effectiveness of the HOTP screening assay, cross-validation was attempted by quantifying the fluorescence value of HOTP and TPA through HPLC. Under actual reaction conditions(70°C), the PET film was decomposed with His-tag purified LCC_wild-type and mutant-type enzymes. Among the comparative groups, it was confirmed that the quantitative value of TPA through HPLC was also high as the intensity of HOTP fluorescence value increased. This screening assay was named HOTPEA screening. Second, it was attempted to confirm TPA in vivo through a biosensor using an activator and green fluorescent protein(GFP) on E.coli. As an activator, the hca operon transcriptional activator(HcaR) protein derived from the E.coli K12 genome was selected. Through Gibson assembly, an HcaR_4HBA biosensor using 4-hydroxybenzoic acid(4HBA) as a ligand was constructed. To change the substrate specificity for 4HBA to TPA, the protein tertiary crystal structure of HcaR_4HBA was predicted using homology modeling, and the mutation location for the ligand binding site was selected through docking simulation. Through reaction conditions and strain optimization, a fold change in GFP fluorescence value of about 6 times compared to the control was obtained for 1 mM 4HBA. Although the constructed HcaR_4HBA biosensor was analyzed through FACS, it was confirmed that there was a problem in the expression of GFP and HcaR. In terms of genetic engineering, there was a need to adjust the promoter responsible for hcaR expression. Therefore, using the T7 promoter, HcaR was overexpressed at various IPTG concentrations. In this case, when IPTG was not added, it was confirmed that the soluble expression of HcaR occurred well. Through this, it was confirmed that as the strength of the promoter was weakened and the soluble expression of HcaR increased, it was advantageous for the fluorescence reaction of the biosensor. Therefore, a new biosensor was constructed using the E.coli constitutive promoter with a weaker strength than the T7 promoter. Although the fluorescence response was not better than that of the existing biosensor using the promoter derived from the E.coli K12 genome, the J23100 promoter was used among the constitutive promoters and the best results were obtained. In addition, due to the ease of promoter engineering, as a follow-up study, the need to increase the fluorescence response of the biosensor was confirmed through combination with RBS of various strengths based on the J23100 promoter cassette. Attempts to develop two activity analysis method and screening system for TPA detection are expected to contribute to eco-friendly PET treatment by helping to study mutations in PET degradation enzymes in the future. 석유화학산업을 기반으로 1950년대부터 다양한 폴리머 물질들이 합성되었다. 그리고 오늘날, 플라스틱 폐기물 처리문제가 전 세계적으로 대두되고 있다. 이 중, 에스터 결합을 기반으로 한 polyethylene terephthalate(PET)는 플라스틱 폐기물에서 가장 큰 비중을 차지하고 있다. 최근 IsPETase를 시작으로 leaf-branch compost cutinase(LCC)와 같은 PET분해효소들이 자연에서 발견되고 있으며, 이를 통해 폐 PET를 처리하는 방법이 주목을 받고 있다. 더 나아가 해당 효소들을 효소공학 측면에서, PET를 terephthalic acid(TPA)까지 분해하는 PET분해-activity를 향상하려는 연구들 또한 각광 받고 있다. 하지만 현재까지 PET분해효소 돌연변이들의 PET분해-activity 비교를 위한, High throughput screening(HTS)은 제대로 보고되지 않은 상태이다. 본 연구에서는 추후 PET분해-activity 비교를 위한 HTS 개발의 기초가 될, Escherichia coli 기반의 2가지 활성분석방법 및 스크리닝 시스템을 시도하였다. 첫째, TPA를 형광물질인 2-hydroxyterephthalate(HOTP)로 합성해 in vitro 상에서 검출하고자 하였다. hydroxyl radical이 TPA와 결합하여, HOTP가 되는 원리를 이용한다. Fe(II)-EDTA 펜톤반응을 이용해, 생물학적 pH조건에서 hydroxyl radical을 생성하였다. LB 배지의 autofluorescence 반응을 제거하기 위해, ethyl acetate(EA)를 사용한 액액 추출법을 적용하였다. TPA 농도와 HOTP 형광세기에 대한 양의 상관관계를 갖는 검량선을 얻을 수 있었다. LOD값은 316.85 uM로 확인되었다. HOTP screening assay의 유효성 확인을 위해, HOTP의 형광값과 HPLC를 통한 TPA 정량을 통해 교차검증 하고자 하였다. 실제 반응조건(70℃)에서, His-tag 정제한 LCC_야생형 및 돌연변이형 효소로 PET 필름을 분해하였다. 비교군 사이에서, HOTP 형광값 세기가 커짐에 따라 HPLC를 통한 TPA 정량값 또한 높게 나옴을 확인하였다. 해당 screening assay를 HOTPEA 스크리닝이라고 명명하였다. 둘째, TPA를 E.coli 상에서 활성인자와 green fluorescent protein(GFP)를 이용한 바이오센서를 통해 in vivo로 확인하고자 하였다. 활성인자로는 E.coli K12 genome 유래의 hca operon transcriptional activator(HcaR) 단백질을 선정하였다. 깁슨 어셈블리를 통해, 4-hydroxybenzoic acid(4HBA)를 리간드로 사용하는 HcaR_4HBA 바이오센서를 구축하였다. 4HBA에 대한 기질특이성을 TPA로 변화시키고자 HcaR_4HBA의 단백질 3차 결정구조를 homology modeling을 이용해 예측하였고, ligand binding site에 대한 돌연변이 위치는 docking 시뮬레이션을 통해 선정하였다. 반응조건 및 균주 최적화를 통하여, 1 mM 4HBA에 대해서, 대조군 대비 6배 정도의 GFP 형광값의 fold change를 얻었다. 구축한 HcaR_4HBA 바이오센서를 FACS를 통해 분석해보았지만, GFP와 HcaR 발현에 문제가 있음을 확인하였다. 유전공학 측면에서, hcaR 발현을 담당하는 프로모터를 조정해야할 필요성이 요구되었다. 그렇기에 T7 프로모터를 이용하여, HcaR을 다양한 IPTG 농도에서 과발현해보았다. 이 경우 IPTG를 넣어주지 않았을 때, HcaR의 soluble expression이 잘 일어남을 확인하였다. 이를 통해 프로모터의 세기가 약해짐에 따라 그리고 HcaR의 soluble expression의 증가함에 따라 바이오센서의 형광반응에 유리함을 확인하였다. 그렇기에 T7 프로모터보다 약한 세기의 E.coli constitutive 프로모터를 이용하여 새롭게 바이오센서를 구축하였다. E.coli K12 genome 유래의 프로모터를 사용한 기존의 바이오센서보다 더 좋은 형광반응을 얻진 못하였지만, constitutive 프로모터 중에선 J23100 프로모터를 사용한 경우에 가장 나은 결과를 얻었다. 또한 프로모터 엔지니어링에 대한 용이성 때문에, 후속연구로는 J23100 프로모터 cassette를 기반으로 다양한 세기의 RBS와의 조합을 통해, 바이오센서의 형광반응을 증가시킬 필요성이 확인되었다. TPA 검출을 위한 2가지 활성분석방법 및 스크리닝 시스템 개발시도는, 앞으로 PET분해효소의 돌연변이 연구에 도움이 되어 친환경적 PET 처리에 기여할 것으로 사료된다.

컴포넌트기반 사이버강의형 원격교육을 위한 저작도구 DELTOC의 설계 및 구현

김창근 전남대학교 대학원 2002 국내박사

최근 급변하는 교육정보화 추세에 부응하여 학습자가 시간과 공간의 제약을 받지 않고 학습할 수 있는 웹기반 원격강의시스템을 이용한 원격교육이 보편화되고 있다. 이는 새로운 교육 패러다임으로 기존의 면대면 교육형태에서 벗어나 시간과 공간의 지배를 받지 않고 학습할 수 있는 학습체제로 변화하고 있음을 의미한다. 인터넷을 이용한 원격교육은 자기주도적 학습, 수준별 학습, 구성주의적 협동학습을 강조하는 최근의 교육학습체제에 적합한 방법이다. 이에 부응하여 최근 교육인적자원부에서는 7차 교육과정에 정보통신기술(Information Communication Technology)을 이용한 새로운 교수 학습방법을 적용하고 있다. 사이버강의형 원격교육을 위해 요구되는 원격강의 컨텐츠들을 컴포넌트화하여 개발·촬용할 수 있고, 웹기반 원격강의시스템 개발시에 요구되는 컴포넌트들을 조립하여 재사용토록 지원해 주는 교육용 저작도구가 요구되고 있다. 하지만 기존의 사이버강의를 위한 저작도구의 경우 개발된 강의컨텐츠들이 컴포넌트기반이 아니며, 또한 개인적인 수준에서 개발 활용되고 있다. 관리와 중복개발 및 시간과 비용의 문제를 안고 있으므로 재사용과 신뢰성, 확장성에 어려움이 있다. 컴포넌트 컨테이너인 웹브라우저에서 시간과 공간의 지배를 받지 않고 활용할 수 있는 사이버강의 컨텐츠들을 컴포넌트화하여 활용할 수 있도록 해주며, 사이버강의형 원격교육시스템 개발을 위한 컴포넌트들을 지원해주는 저작도구의 부재를 들 수 있다. 또한 사이버강의형 저작도구의 경우 대부분 저작모드와 학습모드의 통합환경 인터페이스를 지원하지 않고 있으며, 설계 미숙과 표준차된 플랫폼이 없이 서비스를 제공하는 웹환경으로 인하여 개발자들의 작업이 방대해지고 있다. 본 논문에서 제안한 저작도구 DELTOC(DistancE Lecture Tool based On Component system)은 기존의 웹기반 원격교육시스템의 6가지 유형 즉, 사이버강의형, WBI/HTML형, 학습지/시험형, Q&A/자료실형, 사이버스쿨형, 포털형 중에서 사이버강의형 원격교육시스템에서 요구된 원격강의자료를 저작 및 촬용할 수 있도록 컴포넌트 기반 개발 방법을 적용하여 개발하였다. DELTOC은 다음과 같은 사항을 지원한다. 첫째, 사이버강의형 원격강의컨텐츠를 제작하여 독립환경 및 분산환경인 인터넷 웹상에서 시간과 공간의 제한을 받지 않고 직접 강의컨텐츠를 실행 시켜준다. 컴포넌트간의 데이터/정보를 교환 및 상호작용에 대한 명세를 규정한 COM 이론에 근거한 OCX를 Active-X 컨트롤로 제작한 것이 웹강의 실행 컴포넌트이다. 이를 통해서 시간과 공간의 지배를 벗어나 플러그인 없이 컴포넌트 컨테이너인 웹브라우저에서 직접 강의컨텐츠를 실행시켜 준다. 둘째, 웹에서 사이버강의형 원격교육시스템 즉, 원격강의시스템을 구축할 때 DELTOC에서 지원하는 컴포넌트들을 조립하여 재사용할 수 있다. 원격 강의시스템 구축에 필요한 컴포넌트들의 재사용성, 사용 편리성, 이식성으로 인하여 많은 시간과 비용을 절약할 수 있도록 개발된 강의저작도구이다. 이는 장차 보편차될 컴포넌트 기반 사이버강의형 원격교육을 위한 저작도구이며 교육용 저작도구의 모델을 제공한다. (이후원문누락) Recently, the web-based distance education using distance-lecture system becomes generalized so that learners can study without being restricted to time and space. It means that the new educational paradigm is being more widely used as new instruction method without face-to-face instruction through the internet in various education fields. Internet is used in distance education. Distance education is focused on self-centered, differentiated, and constructive learning. That is appropriate to the recent teaching and learning system. The Ministry of Education and Human Resources Development has applied the teaching and learning method using ICT to the 7th curriculum. It is necessary that the component-based developing methodology should be applied to the distance-lecture system by assembling components using the educational authoring tool. However, lecture contents developed by the traditional authoring tool for cyber lecture is far from Component-Based Software Development. So it is difficult for learner to use lecture contents applied component. Because They are being individually developed. They are very difficult for reuse, and the management of components. In additon to, the degree of faith is relatively low. The web browser is a component container for running cyber-lecture content without being restricted to time and space. But authoring tool for distance education of cyber-lecture type is not component-based software development. It is not user interface of IDE(Integrated Development Environment) for author and learner. It is difficult for developer to service the lecture content on the web. Because the used design skills are not sufficient and these platforms are not standardized. This paper shows DELTOC(DistancE Lecture Tool based On Component system), a authoring tool for cyber lecture. Unlike the current object methodology, a component-based methodology is applied to lecture-system development using DELTOC, which broadens the scope of cyber lectures, eliminating all time and space barriers. Distance-education system is classified as the type of cyber lecture, WBI/HTML, learning worksheet/test, Q&A/DATA, cyber school, and portal. DETOC is the type of cyber lecture. DELTOC has applied the type of cyber lecture using component-based development methodology. DELTOC is different from object oriented or traditional methodology. That can be used on real time or off line. DELTOC is a authoring tool for the following. First, this tool runs lecture contents in SA(Stand Alone) and distributive environment based on the web without being restricted to time and space. That runs lecture contents using OEX's technology based on COM theory of Active-X control. DELTOC is useful for lecture contents on the web browser. Second, DELTOC provides components to construct distance-lecture system. That can make us reuse, easy to use, transplant components. Because DELTOC componenets can be used to construct distance-education system of cyber-lecture type. In conclusion, DELTOC saves time and cost because that has the advantages of reuse and easy to use, transplantation of DELTOC components. DELTOC is the authoring tool for distance lecture through the component-based development as well as offers the model of the educational authoring tool.

선택적 레이저 용융 공정을 이용한 Al-Si계 합금분말 3차원 조형체 제조 및 특성연구

신기훈 울산대학교 2018 국내석사

Selective laser melting (SLM) is one of the additive manufacturing technologies, which allows the fabrication of 3D parts through the use of focused laser energy to powder beds directly from user defined 3D CAD models. Generally, this technology is considered as one of the most promising manufacturing techniques for metallic materials due to its ability to produce complex geometries with unique design and high accuracy that can be used not only for the prototyping step but also for small series productions. During the process, the metallic powders are completely molten by highly localized laser energy, which enables the production of individual parts with high density and excellent mechanical properties. Therefore, the laser energy is a critical factor that plays a decisive role in the porosity and microstructural development in SLM process. In this study the densification behavior and microstructural development of AlSi10Mg parts fabricated by selective laser melting (SLM) were investigated with variation of the applied laser scan speed, laser power, overlap ratio and heat treatment. The starting materials of AlSi10Mg powder particles were prepared using a gas atomization process showing uniform size distribution with high flowability. The atmosphere in the chamber in which the process proceeds is performed in an Ar atmosphere. The laser power was changed to 90W, 180W, 270W and 360W, and the laser speed varied from 100mm/s to 4000mm/s. In order to confirm the microstructures and mechanical properties of the laser beam overlap, the specimens were fabricated with 10%, 30%, 50%, and 70% Generally, in the selective laser melting process, Mg2Si appears in 3D fabrication of AlSi10Mg alloy, but XRD analysis showed no peaks other than Al and Si peaks. The changes of microstructure and relative density with the variation of laser power and scan speed were proportional to the size and number of pores. The relative density increased as the laser scan speed increased, but decreased again after a certain period. The microstructure was also densified as the relative density, and then the pore was again generated. At the same laser speed, relative density increased with increasing laser power. We can confirm that the laser output energy is important as the density decreases after a certain period. Under all conditions, the microstructure of AlSi10Mg alloy was confirmed that Si was segregated around Al and appeared dendrite. This dendrite shape appears because the selective laser melting process is a rapid solidification. The analysis of the characteristics of 3D parts produced by selective laser melting was also confirmed by heat treatment condition. First, as-built state, Si was segregated and microstructure was observed. It was confirmed that Si coalesced in the annealing. Especially, it was confirmed that the size of furnace cooling Si precipitate is smaller than water cooling. The elongation was highest for furnace cooling specimens, while yield strength and UTS were highest for as-built specimens. Vickers hardness value was as high as 142.41Hv in the hardness value of the as-built specimen, similar to the tensile property, and the second highest in 81.5Hv of the sample subjected to age hardening treatment after water cooling.73Hv for water cooling and 52Hv for furnace cooling. Analysis of specimens prepared under each condition showed that optimum conditions were found..

A survey on talking face generation

설쌍인 신라대학교 일반대학원 2026 국내석사

대화형 얼굴 생성(Talking Face Generation) 기술은 음성, 영상, 텍스트 등의 다중 모달 입력으로부터 인물의 입모양, 표정, 제스처를 자연스럽게 합성하는 인공지능 기반 생성 기술로, 디지털 휴먼 연구의 핵심 분야로 부상하고 있다. 본 논문은 해당 기술의 연구 동향을 체계적으로 분석하고, 최근 등장한 대표적 생성 패러다임과 모델을 종합적으로 고찰하였다. 특히 2D 이미지 기반의 Wav2Lip, 3D 기하 기반의 AD-NeRF, 가우시안 표현 기반의 LAM(Large Avatar Model), 통합 확산 모델 기반의 EchoMimic V3, 제스처 확장형 모델 EMO2 등을 중심으로, 각 접근법의 합성 경로, 구조적 특징, 성능 지표를 비교·분석하였다. 또한 VFHQ, HDTF, MOSEI, AVSpeech 등 주요 데이터셋과 평가 지표를 정리하여, 기술적 발전 흐름과 한계점을 함께 제시하였다. 분석 결과, 최근 연구는 효율적 3D 표현·다중모달 통합·전신 협동 생성 방향으로 발전하고 있으며, 실시간 상호작용 및 감정 일관성, 다언어 일반화, 윤리적 거버넌스 등 다양한 연구 과제가 남아 있음을 확인하였다. 본 연구는 디지털 휴먼 기술의 현황과 향후 발전 방향을 종합적으로 제시함으로써, 관련 학문 및 산업 분야에서의 응용 확대에 기여하고자 한다. Talking Face Generation is an artificial intelligence–based generative technology that synthesizes natural lip movements, facial expressions, and gestures of human characters from multimodal inputs such as speech, images, and text, and has emerged as a core research area in digital human studies. This paper systematically analyzes recent research trends in TFG and provides a comprehensive review of representative generative paradigms and models. In particular, we examine and compare approaches including the 2D image–based Wav2Lip, the 3D geometry–based AD-NeRF, the Gaussian representation–based Large Avatar Model, the unified diffusion model–based EchoMimic V3, and the gesture-augmented model EMO2, focusing on their synthesis pipelines, architectural characteristics, and performance metrics. Major datasets and evaluation benchmarks, such as VFHQ, HDTF, MOSEI, and AVSpeech, are also summarized to highlight both the technological progress and the remaining limitations. Our analysis indicates that recent advances are converging toward efficient 3D representations, multimodal integration, and full-body coordinated generation. At the same time, several open challenges remain, including real-time interactive generation, emotional consistency, multilingual generalization, and ethical governance. By presenting a comprehensive overview of the current landscape and future directions of digital human technologies, this study aims to facilitate broader applications in both academic research and industrial practice

Research on Software Defects Predict Methods Based on Bayesian Network

Yuyang Liu 전남대학교 대학원 2009 국내석사

Software defect prediction is an important process in software quality management process. To get Acquire of which files/modules are most likely to contain the largest numbers of defects always in a large software system have significant values to any project developers. For a large scale software system development, testing is also a hard work. Software defect prediction technique gives testers a guide to design the test process, of which file/module are probably to present failures when running or which file/module would include the most density of defects. To accomplish this, in this thesis, we first review past papers to under software defect prediction technique, then analysis software design metrics, which are deem to efficiently predict software quality in the early age of software development, then analysis these metrics through a Bayesian Network method. The probability computation based on Bayesian has a strong theory basis and is concise to compute and easy to understand. In this model how to choose factors can thin and thick on the basis of taking into account technique and management status of team, it can choose data flexible according as expert experience. During software development it is helpful to obtain early estimates of the defect density of software components. Such estimates identify fault-prone areas of code requiring further testing.

인터페이스 명세 기반 EJB 골격코드 생성

박기창 전남대학교 대학원 2003 국내석사

컴포넌트 기반 소프트웨어 개발에서 하나의 컴포넌트는 크게 사용자 관점과 개발자 관점으로 나누어 진다. 이 두 관점은 하나의 컴포넌트를 공유하게 되는데 이는 컴포넌트의 인터페이스를 명세하는 언어인 IDL을 통해 가능하다. 대부분의 분산 컴포넌트 아키텍처는 고유의 IDL을 통해 컴포넌트의 기능적 명세를 바탕으로 컴포넌트를 개발하게 된다. 현존하는 컴포넌트 모델에는 CCM, DCOM, EJB가 있다. C0RBA의 경우 C0RBA IDL이, DCOM의 경우 DCOM IDL이 존재한다. 하지만 EJB는 고유의 IDL이 존재하지 않는다. 이는 컴포넌트 개발 시 사용자 관점을 유지하기 어렵고 단지 EJB 스펙에 정의된 컴포넌트 계약기반으로 컴포넌트를 작성하게 되므로 컴포넌트 개발자와 사용자의 계약관계를 유지하는데 어려움을 유발한다. 본 논문에서는 자바 언어와 유사한 RMI기반 인터페이스 명세를 기반으로 EJB 컴포넌트의 인터페이스를 정의하고 IDL명세 기반 EJB 골격코드를 생성하는 방안에 대해서 연구한다. A Component in CBSD(Component Based Software Development) is largely divided into user view and developer view. These two view points share the concept of component by IDL(Interface Definition Language). Most of distributed component architectures provide unique IDL to users and developers. There are CCM, DCOM and EJB in component model existing today. In the case of CCM, CORBA IDL is used and in the case of DCOM, DCOM IDL is used. But there is not peculiar IDL for EJB. So EJB has a difficulty to present users view exactly. In this thesis, we specify interface of EJB component in RMI-eanbled Java Interface similar to Java language and study on the method to generate EJB skeleton code based on IDL specification.

Rational design for enzyme engineering of CYP153 family and its application to production of ω-hydroxy palmitic acid

정은옥 Seoul National University 2016 국내박사

In this study, the ω–specific hydroxylation of fatty acids using cytochrome P450 monooxygenase (CYPs) was investigated. Among bacterial CYPs in CYP153 family which reported as fatty acid ω-hydroxylase, CYP153As from Marinobacter aquaeolei VT8 (CYP153A33), Alcanivorax borkumensis SK2 (CYP153A13) and Gordonia alkanivorans (CYP153A35) were selected, and compared their specific activities and product yields of ω-hydroxy palmitic acid based on whole-cell reactions toward palmitic acid. Using CamAB as redox partner, CYP153A35 and CYP153A13 showed the highest product yields of ω-hydroxy palmitic acid by whole-cell and in vitro reactions, respectively. To investigate electron transfer system for CYP153A35, artificial self-sufficient CYP153A35-BMR was constructed by fusing it to the reductase domain of CYP102A1 (i.e. BM3) from Bacillus megaterium, and its catalytic activity was compared with CYP153A35 and CamAB system. Unlike the expectations, the system with CamAB resulted 1.5 fold higher yield of ω-hydroxy palmitic acid than that using A35-BMR in whole-cell reaction, whereas the electron coupling efficiency of CYP153A35-BM3 reductase was 4 times higher than that of CYP153A35 and CamAB system. Furthermore, various CamAB expression systems according to gene arrangements of the three proteins and promoter strength in their gene expression were compared in terms of product yields and productivities. Tricistronic expression of the three proteins in the order of camB, cyp153A35 and camA, i.e. A35-AB2 construct, showed the highest product yield from 5 mM of palmitic acid within 9 h in batch reaction system owing to the concentration of CamB, which is the rate limiting factor for the activity of CYP153A35. However, in fed-batch reaction system, A35-AB1 construct, which expressed the three proteins individually using three T7 promoters, resulted the highest product yield of 17.0 mM (4.6 g/L) of ω-hydroxy palmitic acid from 20 mM (5.1 g/L) of palmitic acid in 30 h. For the improvement of hydroxylation activity of CYP153A35, the structures of CYP153A35 were predicted by homology modelling, and the major cavities and the amino acid interacting with the fatty acid were revealed by CAVER 3.0. In order to screen mutants, a powerful high-throughput screening assay was developed, which used Purpald to sense formaldehyde produced as a by-product during O-dealkylation reaction. Saturation mutagenesis on 19 amino acids was performed and D131S mutant showing 281.4 min-1mM-1 of catalytic constant which was more than 17 times higher value than that of wild-type (16.5 min-1mM-1). To optimize the linker sequence between fatty acid ω-hydroxylase (CYP153A33) and reductase domain of CYP102A1, repeated flexible or rigid sequence are designed randomly and screened. The best mutant, EAAAK-(GGGGS)3-EAAAK, showed the 50% higher specific activity than native BM3 linker, although poor expression level in E.coli.

5G 네트워크에서 지도 기계 학습을 이용한 자원 할당 시스템에 관한 연구

박상면 숭실대학교 대학원 2020 국내박사

최근, 기존 무선 네트워크를 이용한 서비스를 제공하는 산업들이 급증하고 있다. 이러한 서비스들은 기존 무선 네트워크의 효율성 손실 및 과부하와 같은 문제를 야기한다. 게다가 모바일 통신에 최적화 된 기존 무선 네트워크는 다양한 산업에서 요구하는 서비스를 제공하기 어렵다. 이러한 문제를 해결하기 위해, 하나의 네트워크를 통해 다양한 산업에서 요구하는 서비스를 제공할 수 있는 네트워크 슬라이싱 기술에 관한 연구가 진행 중이다. 네트워크 슬라이싱 기술을 이용한 기존 자원 할당 방안은 자원 할당만 발생하는 소규모 네트워크에는 적합하지만, 자원 할당 해제도 발생하는 대규모 네트워크에서는 적합하지 않다. 그러므로 네트워크 규모와 환경에 상관없이 효율적으로 자원 할당에 대한 연구가 필요하다. 본 논문에서는 5G 네트워크에서 지도 기계 학습을 이용한 자원 할당 시스템을 제안한다. 슬라이스의 요청에 적합한 물리 서버를 선택하기 위해 제안 시스템은 다양한 산업에서 요청한 서비스들의 패턴을 추출, 학습, 검증할 수 있는 지도 기계 학습을 이용한다. 또한 제안 시스템의 효율성은 processing time, balancing, learning rate simulation을 통해 증명한다. 기존 연구들과 제안 시스템의 비교 분석에 대한 실험은 제안 시스템에 최적인 resource allocation과 resource deallocation이 동시에 발생하는 환경에서 진행한다. 그 결과, 제안 시스템의 processing time은 기존 방안 대비 약 36.53%가 감소하였고 cluster를 적용한 방안 대비 약 15.81%가 감소하였다. Balancing simulation에서 제안 시스템의 balancing은 기존 방안 대비 약 24.95%가 증가하였고 cluster를 적용한 방안 대비 약 11.92%가 증가하였다. 이러한 결과는 서버의 사용률을 이용하는 기존 연구들보다 지도 기계 학습을 이용하는 제안 시스템이 서버의 balancing을 유지하면서 슬라이스 할당하기 위해 요구하는 시간이 적은 것을 보여 제안 시스템의 효율성을 입증한다. 또한 제안 시스템에서 설정한 지도 기계 학습의 learning rate에 대한 실험에서 learning rate에 따른 processing time, balancing, 실제값과 예측값의 차이, MSE, 상관계수를 비교분석하였다. 그 결과, 제안 시스템에서 설정한 learning rate가 최적임을 입증하였다. Recently, the size of industries providing intelligent services using existing wireless networks has been rapidly increased. Newly adapted services amplify various problems of the existing wireless networks such as loss of efficiency and network overload. In addition, existing wireless networks that provide optimized services for mobile communication make it difficult to provide services required by various industries. In order to solve these problems, investigation on network slicing technology that can provide services required by various industries through a single network has been conducted. Existing resource allocation methods using network slicing technology are suitable for small network with only resource allocation. But it is not suitable for large network with resource deallocation. Therefore, regardless the network size and environment, schemes about efficient resource allocation are needed. In this dissertation a new resource allocation system using supervised machine learning in 5G networks is proposed. In order to select a physical server suitable physical server for the slice request, the proposed system uses supervised machine learning to extract, learn, and verify the patterns of services required by various industries. Also, the effectiveness of the proposed system is confirmed through processing time, balancing, and learning rate simulation. The comparative analysis simulation between the existing resource allocation method and the proposed system is performed in resource allocation and resource deallocation environment. As a result, the processing time of the proposed system is reduced by 36.53% compared to the existing resource allocation method and by 15.81% compared to the resource allocation method using the cluster. Also, the balancing of the proposed system is increased by 24.95% compared to the existing resource allocation method and by 15.81% compared to the resource allocation method using the cluster. Result of the comparative analysis simulation between the existing resource allocation method and the proposed system is confirmed that the proposed system is efficient. In addition, learning rate simulation about processing time, balancing, difference between actual and predicted values, MSE, and correlation coefficient according to the learning rate is performed. As a result, the learning rate set by the proposed system has been proved to be optimal.

宋代 文人 ‘硯銘’ 硏究

원종필 전북대학교 일반대학원 2016 국내석사

벼루는 문방의 필수품으로 문인에게 많은 사랑을 받아왔다. 그러나 실제로 벼루 형태상의 정립은 漢代에 이르러 이루어졌고, 唐代에 이르러 여러 석재가 발견되면서 발전을 이루었다. 송대에 이르러서는 벼루를 수집하고 감상하는 문화가 유행하였다. 이에 따라 벼루에 대한 전문서가 등장하기도 하였으며, 벼루 소재의 문학작품의 창작도 매우 흥성하였다. 특히, 硯銘을 짓는 문화가 갑자기 성행했다. ≪사고전서≫에 수록된 작품 중 앞의 저록에 포함된 연명까지 헤아려 볼 때 송대 이전의 연명은 5~6편에 불과한데, 송대의 연명은 200편에 달한다. 본고에서는 이러한 특별한 문화 현상에 대해서 硯銘을 중심으로 전반적으로 조망하고, 송인들에게 벼루가 어떤 의미가 있었기에 이러한 현상이 발생하였는지를 고찰하고자 한다. 이를 위하여 송대 문인들의 벼루에 대한 관심에 대하여 조사하고, 연구의 주요 대상인 ‘銘’이라는 문체에 대한 분석을 진행하였다. 그 결과 송대 문인들은 벼루를 완상하고 즐기는 애완취미의 하나로 바라보았으며, 그 가운데서 특히 벼루를 교우관계의 매개체로 인식하였다는 점을 발견하였다. 또한 문인들은 벼루를 서예를 비롯한 모든 서사활동의 주요 도구로 사용함은 물론이고 거기에서 한 걸음 더 나아가 일상생활의 반려로 인식하였다는 점을 발견하였다. 이러한 인식아래, 송대의 문인들은 자신이 가지고 있는 도덕적 자아를 벼루에 의탁하여 표현한 경우가 매우 많았다. 이러한 인식으로 말미암아 송대의 벼루 애호 문화는 단순한 ‘玩物喪志’의 취미활동을 넘어서 송대 전체에 걸쳐 하나의 ‘문인 문화’로 유행할 수 있었다. 송대의 문인들이 벼루를 단순한 驚奇喜異의 수집품으로 바라보지 않고 도덕적 자아를 의탁할 수 있었던 배경에는 그들이 벼루라는 物을 스스로의 뜻(志)을 지키고 기르며 수련하는 거울이며, ‘游於藝’의 실현 수단으로 생각했기 때문이다. Inkstone has been unusually loved by men of letters as an essential article of the study. However, inkstone shape was actually formed during Han Dynasty and it was developed as its raw stone material was discovered during Tang Dynasty. From the beginning of Song dynasty, a culture of collecting and appreciating inkstone was popularized. Along with this trend, specialized books for inkstone were emerged and creation of literary works based on inkstone became extensively prospered. In particular, a culture of creating 'yan ming' was suddenly prospered. When observing 'yan ming' contained in <Siku Quanshu>, 'yan ming' before Song dynasty was not more than 5-6 pieces but that of Song dynasty reaches 200. In this study, this special cultural phenomenon will be generally explored based on 'yan ming' and how inkstone signified to writers of Song Dynasty will be considered, For this objective, an analysis for a writing style called 'Míng' as a major target of research was performed by surveying concern of writers of Song Dynasty over inkstone. As its result, it was revealed that writers of Song Dynasty regarded inkstone as one of appreciating and enjoying hobbies and they not only used inkstone as major tool of all the narrative activities but also further recognized it as a companion of daily life. Under this recognition writers of Song Dynasty expressed their moral ego by falling back on inkstone in many cases. Due to this recognition, inkstone-loving culture during Song dynasty could be popularized as 'writers' culture' throughout the Song dynasty beyond simple hobby activity of 'Wan wu sang zhi'. The reason why writers of Song dynasty could fall their moral ego on inkstone without regarding inkstone as collection of simple Gyeonggi Hee was that they regarded it as a mirror of keeping, cultivating and training self identity and means of realizing 'You Yu Yi'