Earticle

어노테이션(annotation)은 문서에서 개인의 의견, 정리, 요약 등을 표현하기 위한 주석을 의미한다. 따라서 전자문서에서도 어노테이션은 중요하게 사용되며 특히 전자 잉크(digital inking)릉 이용한 이동 단말기 환경에서 효과적으로 사용된다. 그러나 기존 연구에서는 휴대용 단말기 환경의 단점인 적은 디스플레이 공간을 전혀 고려하지 않기 때문에 어노테이션 작성 및 활용이 매우 불편하다. 따라서 본 논문에서는 전자펜과 이동식 단말기 환경을 고려한 어노테이션 모델 및 시스템을 제안한다. 제안 어노테이션 모델은 다양한 컨텍스트(context)를 고려하고 이에 기반한 어노테이션 마크업 언어를 정의한다. 본 모델은 다양한 어노테이션 타입 및 의미(semantic) 모델, 펜 기반 어노테이션의 자동 인식 및 영역 보정 기능 등을 고려하며, 이것을 기반으로 CAML(Context-based Annotation Markup Language)를 정의한다. 또한 본 모델을 이용하여 XML 기반의 전자책문서 및 단말기 환경을 고려한 어노테이션 시스템을 구현하고 그 활용 가능성에 대하여 살펴본다. 본 연구의 결과는 eLearning, Cyber-Class, IETM(Interactive Electronic Technical Manuals) 에서 적절히 응용 가능하다.

For the accurate creation of annotation information in a free-form annotation environment, the ambiguity that arises in the analysis stage between the geometric information and annotations needs to be resolved. Therefore, this This paper identifies, analyzes, and proposes presents solutions methods for the ambiguity that can occur between free-form marking and various contexts in XML-based annotation environment. The proposed method is based on context which includes various textual and structure information between free-form marking and annotated part. The proposed method show that the annotated portions areas included in the free-form marking information are more accurate, achieving more accurate exchange results amongst multiple users in a heterogeneous document environment. This study can be effectively applied to eLearning, Cyber-Class, and IETM

<順天金氏墓出土簡札(Suncheon KimFamily’s grave-excavated Ganchal)> is a very important data in Korean language history that was excavated in 1977 and reported to the academic world the following year. Nevertheless, this data was out of interest for a long time because of the prejudice about manuscript of Korean language academic world. And after late 1990s, it started to be noted in earnest. Many deciphering work about this data was done and there was a annotation work, too. It made many deciphering errors be corrected and the misinterpreted errors were caught. But the decipherment is unfinished and the meaning interpretation is incomplete. This article is to find the deciphering errors and edit Cho Hang-bum(1998) comparing it with the achievements accomplished after Cho Hang-bum(1998), and to correct the misinterpreted errors on the basis of the edited decipher and existing study. There can be ‘the error of the letter decipher’ and ‘the error of punctuating’ in the deciphering error. In Cho Hang-Bum(1998), there are many errors in these two aspects, especially in ‘the letter decipher.’ This may be caused by simple mistake, but it could come from carelessness of the order of grammar, calligraphic style and comparison of letter. The error of ‘punctuating’ was caused mainly by the ambiguity of the meaning of the word or the whole sentence. The error of misinterpretation can be found mainly in a rare word, the name of a certain area and special Chinese. Because the annotation work for this data is not so active, there are not so much edited content in the meaning interpretation. Most of the words treated as ‘the unidentified’ in Cho Hang-bum(1998) is still unidentified. They are usually used in spoken language, in special area or life words related to the period, so it is not easy to solve this problem. It is essential and urgent to reduce ‘the unidentified words’ for the appropriate decipherment of the Korean old vernacular letters.

이 논문은 방대한 분량의 자료를 소장하고 있는 규장각한국학연구원의 자료 가운데 한국음악학 관련 자료가 영인‧번역된 현황을 정리하고, 해제의 양상을 살펴보기 위한 것이다. 지금까지 한국음악학 연구 자료의 영인과 번역을 통해 적극적인 사료 공유를 시도하며 연구의 저변을 확대시켜 왔는데 이는 연구기초자료의 확보 측면에서 한국음악학 연구와는 별도로 평가되어야 한다. 한국음악학계의 영인·번역의 흐름을 살펴보면 간행주체와 대상에 따라 4시기로 나눌 수 있다. 제 1시기(1933~1959)는 자료 발굴기로 중요하고 기본적인 자료들이 영인되었으며, 인접학문 분야에서도 기초사료의 영인이 이루어졌다. 제 2시기(1960~1979)는 음악자료의 기틀마련기로 고악보 중심의 총서 발간이 이루어졌고, 제 3시기(1980~1999)는 자료의 정착발전기로 이 사업을 이어받아 국립국악원에서 80년대에는 영인본 총서를, 90년대에는 번역본 총서를 발간해내기 시작하면서 본격적으로 많은 한국음악학 자료들이 보급되었다. 이를 중심으로 자료가 양적‧질적으로 심화되고 연구도 진전되었다. 제 4시기(2000~현재)는 성과물 축적을 통한 학제간 연구기로 여러 학술단체의 번역총서가 등장하기 시작하였다. 특히 의궤분야 연구가 심화되었고, 비음악학 분야 자료가 보다 중요해졌다. 영인‧번역서에 딸린 해제는 연구동향과 자료 특성에 따라 의궤, 악서, 악보로 나누어서 살펴볼 수 있었다. 처음에는 자료의 소개와 연구 소재를 제시하는 것에 그쳤으나 한국음악학의 연구동향과 밀접한 관계를 가지며 해제 자체가 연구성과가 되거나 향후 연구 과제를 제시해주기도 하였으며, 연향관련 의궤가 가장 활발하게 간행되고 있었다. 앞으로 아직 영인되지 않은 의궤, 이미 간행되었으나 품질이 낮거나 판본이 다른 자료, 고악보의 문자정보 등 자료의 사각지대를 잘 살펴 자료를 확보해나가야 한다. 또한 디지털 서비스의 확대로 출판과 색인의 수고를 덜 수 있으므로 이를 적극적으로 활용해야 한다.

This article is to understand the current situation of the photoprint‧translation‧bibliographic explanation about Korean musicology of the massive materials housed in Kyujanggak Institute for Korean Studies. The base of Korean Musicology study has expanded through publishing of phoroprint and translation of Korean musicology materials, trying to sharing resources of Korean Musicology. Therefore it needs to be reviewed in terms of securing the basic research materials separately from the studies on Korean Musicology. In the first period(1933~1959) which was an exploring era of resources and primary musical materials were photoprinted as well as related studies materials. In the second period(1960~1979) which was paving the way of Korean musicology, photoprint of the series of old music scores tried in priority. The third period(1980~1999) was when the materials were established. For that the National Gukak Center published the series of the photoprint of Korean musicology materials in the 1980s and the series of the translation in the 1990s thereby many Korean Musicology materials have been in earnest supplied, which are more intensive quantitatively as well as qualitatively and advanced in the studies. In the fourth(2000~) period that is interdisciplinary research through building up the results, existing data are update many academic societies publish the series of translated korean musical historic materials and non-musicology materials are becoming more importance, especially deepening study on Uigwae. The annotations in the photoprinted and translated books or the website of the Kyujanggak Institute for Korean Studies give a brief information about the source material to star with. However they are getting closely connected with research trends, suggesting a research subject, also that is the result in itself. Looking into Uigwae, old score, music theory book, these days Uigwae related royal court banquet are most active published as photoprint and translation. From now on it's necessary to watch the development of the work and the blind spot of photoprint and translation: resource materials not yet to photoprint and translate, already published version in lower-quality, different version and the text information of the old music scores. Also we can expand active-using digitalized services to save trouble of printing or index of materials.

Real-time emotion recognition technology is required for human-robot interaction. There is a method using the Somatic Nervous System (SNS). The PPG signal, which is easy to acquire data and can be used by dividing it into pulse units, is easy for real-time emotion recognition. However, each pulse label is the most important factor in segmenting and using the short-term 1-second to 3-second PPG signal. However, DEAP and MAHNOB-HCI are public bio-signal data, but DEAP Dataset only provides Self-Assessment Labeling for emotion-inducing videos of 60 seconds. The MAHNOB-HCI Database provides Annotation Labeling in which the observer sees the frontal image of the subject's face and measures emotion in real-time. Annotation labeling is effective to split the labels of PPG signals by pulse units. As a result of the experiment, Arousalbased positives 2415/neutral 3884/negative 4201, Valence-based positives 3570/neutral 2835/negative 4095.labels were drived. Annotation labeling allows us to check the emotions of participants in various distributions for 60 seconds. Analyzing the label setting and data preprocessing process contributes to improving short-term real-time emotion recognition.

전자펜을 이용한 문서교정 시스템에서 정확한 교정결과를 보장하기 위해서는 문서 교정자가 드로잉한 교정부호와 문서내용간의 영역 모호성(ambiguity)을 해결하여야 한다. 한편 교정의 대상이 되는 전자문서가 HTML/XML과 같은 경우 교정된 문서구조가 반드시 기 정의된 DTD를 위배하지 않아야 한다. 본 논문에서는 펜 기반의 교정시스템에서 교정부호(마킹)와 대상문서간의 모호성 문제를 최소화하기 위한 기법을 제안한다. 제안 인터페이스에서는 모호성 문제를 최소화하기 위하여 교정부호와 문서간의 컨텍스트(Context)를 반영하였으며 동시에 대상문서의 문서 구조를 유지하기 위한 방법을 제공한다. 그 결과 본 논문에서 제안한 교정 인터페이스는 기존 교정시스템에 비하여 보다 정확한 영역정보를 포함할 수 있으며, 교정부호 입력에 따른 구조문서 변경시에도 원본문서의 DTD에 따르는 문서구조를 유지할 수 있다.

To produce accurate editing results, the ambiguity of editing scopes related to marked correction signs should be solved. Proofreading the web document modifies the document structures, and the modified structures should be robustly valid for the defined DTD. This paper presents a pen-based proof-reading interface in the XML document. In the proposed interface, correction signs are free-drawn, and the editing scopes are recognized and revised based on the contexts of the document to minimize the ambiguity of the editing scopes. The proposed interface provides both implicit and explicit modification methods for document structures. As a result, the editing scopes processed in the proposed interface are more accurate, and the document structures are maintained valid for DTD after the editing.

The usage of virtual reality (VR) in healthcare field has been gaining attention lately. The main use cases revolve around medical imaging and clinical skill training. Healthcare professionals have found great benefits in these cases when done in VR. While medical imaging on the desktop has lots of available software with various tools, VR versions are mostly stripped-down with only basic tools. One of the many tool groups significantly missing is annotation. In this paper, we survey the current situation of medical imaging software both on the desktop and in the VR environment. We will discuss general information on medical imaging and provide examples of both desktop and VR applications. We will also discuss the current status of annotation in VR, the problems that need to be overcome and possible solutions for them. The findings of this paper should help developers of future medical image annotation tools in choosing which problem they want to tackle and possible methods. The findings will be used to help in our future work of developing annotation tools.

Comparing online and paper-based environment for report submission and correction, the former supersedes to the latter, since (1) the turn-around time becomes shorter, (2) teaching opportunity increases, and (3) as a consequence, the student's achievement level becomes higher in the online environment. In this paper, we propose an annotation tool GABA for PDF document in order to reduce correction time by the teachers and to facilitate instruction to students. In a usual class, the same or similar assignments are given to the students. Then it is often the case that many students make similar mistakes. A teacher can register and classify common correction comments to GABA. Report correction time becomes significantly shorter by reusing the registered comments. GABA also provides various support functions in order to assist efficient checking of numerous report files such as (1) sorting of frequently-used comments, (2) similarity-based file sorting, and (3) cross tabulation of comments using category and weight.

Background: Heat shock proteins play an important role in protection from stress stimuli and metabolic insults in almost all organisms. Methods: In this study, computational tools were used to deeply analyse the physicochemical characteristics and, using homology modelling, reliably predict the tertiary structure of the blunt snout bream (Ma-) Hsp70 and Hsc70 proteins. Derived three-dimensional models were then used to predict the function of the proteins. Results: Previously published predictions regarding the protein length, molecular weight, theoretical isoelectric point and total number of positive and negative residues were corroborated. Among the new findings are: the extinction coefficient (33725/33350 and 35090/34840 - Ma-Hsp70/ Ma-Hsc70, respectively), instability index (33.68/35.56 ? both stable), aliphatic index (83.44/80.23 ? both very stable), half-life estimates (both relatively stable), grand average of hydropathicity (?0.431/-0.473 ? both hydrophilic) and amino acid composition (alanine-lysine-glycine/glycine-lysine-aspartic acid were the most abundant, no disulphide bonds, the N-terminal of both proteins was methionine). Homology modelling was performed by SWISS-MODEL program and the proposed model was evaluated as highly reliable based on PROCHECK’s Ramachandran plot, ERRAT, PROVE, Verify 3D, ProQ and ProSA analyses. Conclusions: The research revealed a high structural similarity to Hsp70 and Hsc70 proteins from several taxonomically distant animal species, corroborating a remarkably high level of evolutionary conservation among the members of this protein family. Functional annotation based on structural similarity provides a reliable additional indirect evidence for a high level of functional conservation of these two genes/proteins in blunt snout bream, but it is not sensitive enough to functionally distinguish the two isoforms.

Protein-translated mRNA analysis has been extensively used to determine the function of various traits in animals. The non-coding RNA (ncRNA), which was known to be non-functional because it was not encoded as a protein, was re-examined as it was studied to actually function. One of the ncRNAs, long non-coding RNA (lncRNA), is known to have a function of regulating mRNA expression, and its importance is emerging. Therefore, lncRNAs are currently being used to understand the traits of various animals as well as human diseases. However, studies on lncRNA annotation and its functions are still lacking in most animals except humans and mice. lncRNAs have unique characteristics of lncRNAs and interact with mRNA through various mechanisms. In order to make lncRNA annotations in animals in the future, it is essential to understand the characteristics of lncRNAs and the mechanisms by which lncRNAs function. In addition, this will allow lncRNAs to be used for a wider variety of traits in a wider range of animals, and it is expected that integrated analysis using other biological information will be possible.

According to previous studies, the form of a city mentioned in Kaogongji(考工記) of Zhouli(周禮) does not exist in reality. Only Beijing during Ming(明) and Qing(淸) Dynasties is discussed as an example, making it lose its worth as a theory. But of all the annotation of Zhouli throughout the 2,000 years before the modern era, core theory related to capital construction had never been stated from the aspect of the present day. Such discussion can be found depicted in Yingzaofashi(營造法式), a specialized book about architectural technology. Unlike what is known until now, the principle of capital construction has a link to the theory of Fengshui(風水), in that it implies the logic of 'Yi(易)'.

As a result of the rapid growth in the elderly population in Korea, it is expected that there will be problems in providing housing for use by the elderly. Although the government of Korea is seeking to revitalise the welfare housing system for the elderly, on the whole, private and public supplies have been reached to those in high-income and low-income households. Therefore, regardless of their economic status, it is necessary to identify the crucial housing issues in terms of design and research that is in line with the government’s policies. This study recognises the current conditions of the residential environment for the elderly demographic in Korea, and expands its physical architectural concept to the wider context of the healthy environment. The purpose of this study is to derive EBD factors from the elderly based on Evidence-based Design methodology researches, which are Healthcare at Home General Annotation and WELL Certification. This study will yield findings that aid in elderly housing renovation project(s) and new-build project(s).

Even human genome has been sequenced completely, we have very limited idea of all characterized genes at the protein level. Therefore, several attempts have been taken to map proteome of human chromosome. Since sex chromosomes determined the sex of individual, it important to study sex chromosome-encoded proteins. Here, we studied human sex chromosome- encoded proteins of the immune system, protein pathways, protein-protein interactions, and diseases association using several bioinformatics tools. We retrieved 30 proteins (X chromosome 28; Y chromosome 7; both 5) from the recent NCBI human genome annotation, based on their association with immune systems (immunological and inflammation pathways) by Pathway studio program. Searching of these proteins in the Human Proteome Project, including neXtProt, PeptideAtlas, and the Human Protein Atlas showed that all proteins were also identified in several cells and tissues of body. Proteins were further investigated using Pathway Studio and STRING programs. Pathway Studio and STRING programs showed 15 and 25 X chromosome-encoded proteins were interacted. All Y chromosome-encoded proteins showed interactions using STRING program, however no interaction was found using Pathway studio. In addition, difference in human sex chromosome-encoded proteins of the immune system showed an indirect relationship with the occurrence of some diseases in a sex specific manner.

The recognition of crop diseases and pests based on images is one of the required techniques to identify damage due to diseases and pests and to take efficient management and prior actions. With the development of deep learning technology, the image recognition of diseases and pests using deep learning exhibited excellent performance and plays an important role in managing and controlling diseases and pests of crops. However, research on image recognition of diseases and pests in crops is facing difficulties due to the lack of large-scale datasets about diseases and pests. To solve this problem, this study proposes a system to manage and annotate the images of diseases and pests that can efficiently manage collected disease and pest images to generate high-quality disease and pest datasets. The proposed disease and pest image management and annotation system can collect disease and pest images uploaded from various sources and create statistics of them. It can also provide image inspection and user-friendly annotation functions.

The gap between the increasing amount of raw sequence data and qualified biological information is widening: it is not possible to provide qualified annotation for new genomes with the same speed at which sequences are generated -- especially with next generation sequencing technologies. Additionally, managing the huge number ofavailable bioinformatics algorithms represents a challenge of its own. Thus, a systemwhich provides maximum annotation quality with approved methods in a minimum of time is absolutely critical as the basis for the R&D pipeline in industrial biotechnology. Also using those various genome annotation, comprehensive knowledge of a fungal’s biosynthetic capacities is the indispensable basis of its usefulness in the biotechnology research. Comparative genomics is on the rise as a potent tool in molecular biology. The publicly available genomes from yeasts and several fungal species are used to comparative genomics method for discovery of functional information in our research. Also we had integrated information from different data sources to dynamically build semantic networks representing up-to-date knowledge, which can then be mined and analyzed. Through those methods, we could inspect the common features and uniques between yeasts and fungal species in the new aspect.

중국인 학습자는 전 세계 영어 학습자 가운데 가장 큰 집단을 형성하고 있음에도 불구하고, 이들의 영어 작문을 대상으로 포괄적인 오류 주석을 제공하는 공개 코퍼스는 아직 충분히 구축되어 있지 않다. 이러한 자료적 공백은 중국인 영어 학습자의 오류 양상에 대한 체계적 연구를 제약할 뿐만 아니라, 실제 교육 현장에서 활용 가능한 피드백 자료의 개발 또한 제한한다. 본 연구는 이러한 한계를 보완하기 위해 신뢰도 높은 오류 주석을 산출하는 동시에 교육적 피드백을 생성할 수 있는 인간 참여형 AI 보조 프레임워크를 제안한다. 이 프레임워크는 문법, 어휘, 표기 및 형식의 세영역으로 구성된 간소화된 오류 분류 체계를 채택하며, 각 오류 주석은 오류 유형, 오류 범위, 수정안, 설명의 네 가지 항목으로 이루어진 최소 기록 구조에 따라 제시된다. 본 연구에서는 해당 프레임워크를 중국인 영어 학습자 10,000 편 작문 코퍼스(TECCL)에 적용하였으며, 주석 과정에는 중국인 영어 학습자 코퍼스(CLEC)에 기반한 문맥 내 학습 예시, 연쇄 사고 추론, 검색 증강 생성이 통합되었다. 100 편의 에세이를 대상으로 한 개념 증명 평가 결과, 주석자 간 일치도는 전체 Cohen’s κ = 0.78 에 도달하였고, 오류 범위의 정확 일치율은 85%였으며, 주석자 내 일치도는 90%를 초과하였다. 또한 164 개 문장을 분석한 결과, 총 187 개의 오류가 확인되었고, 그 분포는 문법 66.3%, 어휘 23.0%, 표기 및 형식 10.7%로 나타났다. 전문가 검토에서는 형태통사 및 어휘 오류에 대한 높은 정밀도가 확인되었으며, 주로 관용표현 또는 연어와 관련된 9 개의 누락 오류가 추가로 발견되었다(재현율 = 95.4%). 이러한 예비 분석 결과는 본 연구에서 제안한 프레임워크가 연구 간 비교 가능성을 유지하면서도 오류 주석 작업의 투명하고 확장 가능한 수행을 지원할 수 있음을 보여준다. 나아가 본 프레임워크는 교실 수업에 활용 가능한 오류 피드백 자료를 생성할 수 있다는 점에서 학습자 코퍼스 연구와 언어 교육 실천을 연결하는 방법론적 가능성을 지닌다.

Large-scale error annotation in learner corpora remains challenging, particularly for under-resourced L1 populations. Although Chinese learners represent the world’s largest group of English learners, publicly accessible corpora of their English writing with comprehensive error annotations are scarce, limiting research and pedagogical applications. This study introduces a human-in-the-loop, AI-assisted framework designed to deliver reliable annotations while generating pedagogical feedback. The framework employs a streamlined three-domain taxonomy (Grammar, Lexis, and Mechanics) and a four-field record structure (error type, span, correction, and explanation). Applied to the Ten-thousand English Compositions of Chinese Learners, the pipeline integrates in-context demonstrations from the Chinese Learner English Corpus, chain-of-thought reasoning, and retrieval-augmented generation. In a proof-of-concept evaluation of 100 essays, inter-annotator agreement reached Cohen’s κ = 0.78, with an exact-span match of 85%, and intra-annotator agreement exceeded 90%. Analysis of 164 sentences yielded 187 errors: Grammar 66.3%, Lexis 23.0%, and Mechanics 10.7%. Expert review confirmed high precision for morphosyntactic and lexical errors and identified nine omissions (recall = 95.4%), involving idiomaticity or collocation. These results suggest that the approach can support scaling while preserving research comparability and producing materials for classroom use. Although further validation with larger datasets is required, the framework shows potential for adaptation to other learner populations.

고품질 감성 분석 데이터셋은 형태론적으로 복잡한 교착어를 포함한 자연어 처리(NLP) 성능 향상에 핵심적이다. 본 연구는 이러한 언어적 특성을 고려하여 감성 주석의 일관성과 정확성을 향상시키기 위한 목적을 가진다. 이를 위해 감성 분류 를 다단계로 구성하는 계층적 감성 분류(Hierarchical Sentiment Voting, HSV) 방식과 불확실성이 높은 주석에 인간 검토 를 추가하는 인간 개입 기반(Human-in-the-Loop, HITL) 프레임워크를 결합한 하이브리드 주석 방식을 제안한다. 연구 범 위는 한국어를 포함한 교착어에 집중되며, 다양한 온라인 데이터(상품 후기, 영화 리뷰, 커뮤니티 댓글 등)로부터 150만 건 이 상의 텍스트를 수집하여 데이터셋을 구축하였다. 실험 결과, 제안된 방법은 주석자 간 일치도와 감성 분류 모델의 성능을 모 두 향상시켰으며, 특히 미세한 감성 차이를 요구하는 문맥에서 높은 안정성을 보였다. 본 연구는 교착어 기반 감성 데이터셋 구축에 있어 구조화된 인간 정제 주석이 중요함을 실증적으로 제시하며, 실무적으로는 정확한 감성 분석 모델 개발에 기여할 수 있는 데이터 구축 프레임워크로 활용될 수 있다.

High-quality sentiment analysis datasets are critical for enhancing the performance of natural language processing (NLP) systems, especially for morphologically complex agglutinative languages. This study aims to improve the consistency and accuracy of sentiment annotation by considering the unique linguistic characteristics of such languages. To this end, we propose a hybrid annotation framework that combines Hierarchical Sentiment Voting (HSV), which organizes sentiment classification into multiple levels, with a Human-in-the-Loop (HITL) mechanism that selectively applies human validation to low-confidence annotations. The study focuses on agglutinative languages, including Korean, and constructs a dataset of over 1.5 million text samples collected from various online sources such as product reviews, movie critiques, and community comments. Experimental results show that the proposed method significantly improves inter-annotator agreement and model performance, particularly in contexts requiring fine-grained sentiment distinctions. This research empirically demonstrates the importance of structured, human-refined annotation for building reliable sentiment datasets in agglutinative languages and presents a practical framework that can support the development of accurate sentiment analysis models.

클라우드 컴퓨팅이 널리 사용되면서, 데이터 유출에 대한 관심도 같이 증가하고 있다. 동형암호는 데이터를 암호화된 채로 클라우드 서버에서 연산을 수행함으로써 해당 문제를 해결할 수 있다. 그러나, 프로그램 전체를 동형암호로 연산하는 것은 큰 오버헤드를 가지고 있다. 프로그램의 일부분만 동형암호를 사용하는 것은 오버헤드를 줄일 수 있지만, 사용자가 직접 프로그램의 코드를 분할하는 것은 시간이 오래 걸리는 작업이고 또한 에러를 발생시킬 수 있다. 이 연구는 지시문을 활용하여 동형암호 프로그램의 코드를 분할하는 컴파일러인 Heapa를 제시하였다. 사용자가 프로그램에 클라우드 컴퓨팅 영역에 대한 코드를 지시문으로 삽입하면 Heapa는 클라우드 서버와 호스트사이의 통신 및 암호화를 적용시킨 계획을 세우고, 분할된 프로그램을 생성한다. Heapa는 영역 단위의 지시문뿐만 아니라 연산 단위의 지시문도 사용가능하여 프로그램을 더 세밀한 단계로 분할 가능하다. 이 연구에선 6개의 머신러닝 및 딥러닝 어플리케이션을 통해 컴파일러의 성능을 측정했으며, Heapa는 기존 동형암호를 활용한 클라우드 컴퓨팅보다 3.61배 개선된 성능을 보여주었다.

Despite its wide application, cloud computing raises privacy leakage concerns because users should send their private data to the cloud. Homomorphic encryption (HE) can resolve the concerns by allowing cloud servers to compute on encrypted data without decryption. However, due to the huge computation overhead of HE, simply executing an entire cloud program with HE causes significant computation. Manually partitioning the program and applying HE only to the partitioned program for the cloud can reduce the computation overhead. However, the manual code partitioning and HE-transformation are time-consuming and error-prone. This work proposes a new homomorphic encryption enabled annotation-guided code partitioning compiler, called Heapa, for privacy preserving cloud computing. Heapa allows programmers to annotate a program about the code region for cloud computing. Then, Heapa analyzes the annotated program, makes a partition plan with a variable list that requires communication and encryption, and generates a homomorphic encryptionenabled partitioned programs. Moreover, Heapa provides not only two region-level partitioning annotations, but also two instruction-level annotations, thus enabling a fine-grained partitioning and achieving better performance. For six machine learning and deep learning applications, Heapa achieves a 3.61 times geomean performance speedup compared to the non-partitioned cloud computing scheme.