논문 상세보기

한국인터넷정보학회> 인터넷정보학회논문지> 내장형 인공지능 프로세서를 위한 성능 분석기

KCI등재

내장형 인공지능 프로세서를 위한 성능 분석기

Performance Analyzer for Embedded AI Processor

황동현 ( Dong Hyun Hwang ) , 윤영현 ( Young Hyun Yoon ) , 한창엽 ( Chang Yeop Han ) , 이승은 ( Seung Eun Lee )
  • : 한국인터넷정보학회
  • : 인터넷정보학회논문지 21권5호
  • : 연속간행물
  • : 2020년 10월
  • : 149-157(9pages)

DOI


목차

1. 서 론
2. KNN과 RBF
3. 인공지능 프로세서 성능 분석기
4. 실 험
5. 결론 및 향후 연구과제
참고문헌(Reference)

키워드 보기


초록 보기

최근 인공지능에 대한 관심이 높아짐에 따라 인공지능 프로세서를 하드웨어로 구현하는 연구가 활발히 진행되고 있다. 하지만 인공지능 프로세서는 기존에 기능 검증을 위한 프로세서 시뮬레이션 외에 애플리케이션 단계에서 인공지능 프로세서가 해당 애플리케이션에 적합한지에 대한 성능 검증이 추가로 필요하다. 본 논문에서는 인공지능 프로세서를 활용한 애플리케이션 성능 검증과 프로세서의 한계점을 탐색할 수 있는 내장형 인공지능 프로세서를 위한 성능 분석기를 제안한다. 본 논문은 내장형 인공지능 프로세서를 위한 성능 분석기를 구현하기 위하여 기존에 구현된 인공지능 프로세서의 구조를 분석하고 이를 기반으로 인공지능 프로세서를 모사하는 내장형 인공지능 프로세서를 위한 성능 분석기를 구현한다. 내장형 인공지능 프로세서를 위한 성능 분석기를 활용해 이미지 인식, 음성 인식 애플리케이션에서 인공지능 프로세서의 성능 분석 및 한계점을 탐색하고, 제한된 메모리 크기 안에서 인공지능프로세서의 구조를 최적화한다.
Recently, as interest in artificial intelligence has increased, many studies have been conducted to implement AI processors. However, the AI processor requires functional verification as well as performance verification on whether the AI processor is suitable for the application. In this paper, We propose an AI processor performance analyzer that can verify the application performance and explore the limitations of the processor. By Using the performance analyzer, we explore the limitations of the AI processor and optimize the AI model to fit an AI processor in image recognition and speech recognition applications.

UCI(KEPA)

간행물정보

  • : 공학분야  > 기타(공학)
  • : KCI등재
  • :
  • : 격월
  • : 1598-0170
  • : 2287-1136
  • : 학술지
  • : 연속간행물
  • : 2000-2020
  • : 1590


저작권 안내

한국학술정보㈜의 모든 학술 자료는 각 학회 및 기관과 저작권 계약을 통해 제공하고 있습니다.

이에 본 자료를 상업적 이용, 무단 배포 등 불법적으로 이용할 시에는 저작권법 및 관계법령에 따른 책임을 질 수 있습니다.

발행기관 최신논문
| | | | 다운로드

1오프 폴리시 강화학습에서 몬테 칼로와 시간차 학습의 균형을 사용한 적은 샘플 복잡도

저자 : 김차영 ( Chayoung Kim ) , 박서희 ( Seohee Park ) , 이우식 ( Woosik Lee )

발행기관 : 한국인터넷정보학회 간행물 : 인터넷정보학회논문지 21권 5호 발행 연도 : 2020 페이지 : pp. 1-7 (7 pages)

다운로드

(기관인증 필요)

초록보기

강화학습에서 근사함수로써 사용되는 딥 인공 신경망은 이론적으로도 실제와 같은 근접한 결과를 나타낸다. 다양한 실질적인 성공 사례에서 시간차 학습(TD) 은 몬테-칼로 학습(MC) 보다 더 나은 결과를 보여주고 있다. 하지만, 일부 선행 연구 중에서 리워드가 매우 드문드문 발생하는 환경이거나, 딜레이가 생기는 경우, MC 가 TD 보다 더 나음을 보여주고 있다. 또한, 에이전트가 환경으로부터 받는 정보가 부분적일 때에, MC가 TD보다 우수함을 나타낸다. 이러한 환경들은 대부분 5-스텝 큐-러닝이나 20-스텝 큐-러닝으로 볼 수 있는데, 이러한 환경들은 성능-퇴보를 낮추는데 도움 되는 긴 롤-아웃 없이도 실험이 계속 진행될 수 있는 환경들이다. 즉, 긴롤-아웃에 상관없는 노이지가 있는 네트웍이 대표적인데, 이때에는 TD 보다는 시간적 에러에 견고한 MC 이거나 MC와 거의 동일한 학습이 더 나은 결과를 보여주고 있다. 이러한 해당 선행 연구들은 TD가 MC보다 낫다고 하는 기존의 통념에 위배되는 것이다. 다시 말하면, 해당 연구들은 TD만의 사용이 아니라, MC와 TD의 병합된 사용이 더 나음을 이론적이기 보다 경험적 예시로써 보여주고 있다. 따라서, 본 연구에서는 선행 연구들에서 보여준 결과를 바탕으로 하고, 해당 연구들에서 사용했던 특별한 리워드에 의한 복잡한 함수 없이, MC와 TD의 밸런스를 랜덤하게 맞추는 좀 더 간단한 방법으로 MC와 TD를 병합하고자 한다. 본 연구의 MC와 TD의 랜덤병합에 의한 DQN과 TD-학습만을 사용한 이미 잘 알려진 DQN과 비교하여, 본 연구에서 제안한 MC와 TD의 랜덤 병합이 우수한 학습방법임을 OpenAI Gym의 시뮬레이션을 통하여 증명하였다.

2상호접속료인가, 망 이용대가인가? - ISP-CP간 망 연결 대가 분쟁 중심으로 -

저자 : 조대근 ( Dae-keun Cho )

발행기관 : 한국인터넷정보학회 간행물 : 인터넷정보학회논문지 21권 5호 발행 연도 : 2020 페이지 : pp. 9-20 (12 pages)

다운로드

(기관인증 필요)

초록보기

본 연구는 넷플릭스-SK브로드밴드간 대가 분쟁에서 나타난 망의 연결행위와 그 대가에 대한 용어 혼란을 국내·외 법령을 통해 재정립해보고 있다. 양측 분쟁 당사자, 학계, 언론 등에서 망 이용 및 제공에 따른 금전적 반대급부를 “(상호)접속료” 또는 “망 이용대가” 등의 용어를 통일성 없이 사용하고 있고, 경우에 따라서는 전략적 목적에 따라 혼용하고 있다. 동일한 현상에 대해 서로 다른 용어를 사용하는 것(또는 그 반대도 동일)은 문제에 대한 통일된 접근, 생산적이고 합리적인 논의, 더 나아가 분쟁 해결을 어렵게 한다는 점에서 이 연구는 의의를 가진다. 이에 본 연구는 망 이용 및 연결과 관련된 용어 즉 “이용”, “접근(Access)”, “상호접속(Interconnection)”과 그에 따른 반대급부로서의 비용 관련 용어를 상호 비교/분석하여 (상호)접속료와 망 이용대가를 구분하여 사용할 것을 제언하고, 향후 ICT 부문 이슈 해결에 단초로서 기능할 수 있도록 하고 있다. 본 연구 결과 넷플릭스-SK브로드밴드간 망 연결/이용에 따라 수수하는 금전적 반대급부는 망 이용대가(Access fee) 또는 (소매)요금이며, 네트워크를 보유/운용하는 기간통신사업자간(ISP)간 연결에서 발생하는 수수료에 한정하여 “상호접속료(Interconnection fee)”라는 용어로 통일하여 사용할 것을 제안하고 있다.

3Empirical Comparison of Deep Learning Networks on Backbone Method of Human Pose Estimation

저자 : Beanbonyka Rim , Junseob Kim , Yoo-joo Choi , Min Hong

발행기관 : 한국인터넷정보학회 간행물 : 인터넷정보학회논문지 21권 5호 발행 연도 : 2020 페이지 : pp. 21-29 (9 pages)

다운로드

(기관인증 필요)

초록보기

Accurate estimation of human pose relies on backbone method in which its role is to extract feature map. Up to dated, the method of backbone feature extraction is conducted by the plain convolutional neural networks named by CNN and the residual neural networks named by Resnet, both of which have various architectures and performances. The CNN family network such as VGG which is well-known as a multiple stacked hidden layers architecture of deep learning methods, is base and simple while Resnet which is a bottleneck layers architecture yields fewer parameters and outperform. They have achieved inspired results as a backbone network in human pose estimation. However, they were used then followed by different pose estimation networks named by pose parsing module. Therefore, in this paper, we present a comparison between the plain CNN family network (VGG) and bottleneck network (Resnet) as a backbone method in the same pose parsing module. We investigate their performances such as number of parameters, loss score, precision and recall. We experiment them in the bottom-up method of human pose estimation system by adapted the pose parsing module of openpose. Our experimental results show that the backbone method using VGG network outperforms the Resent network with fewer parameter, lower loss score and higher accuracy of precision and recall.

4The Design and Implementation of Mobile Application Solution for Forest Fire based on Drone Photography and Amazon Web Service (AWS)

저자 : Si-eun Choi , Jong-ho Bang

발행기관 : 한국인터넷정보학회 간행물 : 인터넷정보학회논문지 21권 5호 발행 연도 : 2020 페이지 : pp. 31-37 (7 pages)

다운로드

(기관인증 필요)

초록보기

Last year's Goseong-Sokcho forest fires have highlighted the limitations of extinguishing work for night-time forest fire and the importance of quick identification for information on the spread of forest fire. However, it is not easy to find services that take into account the characteristics of forest fires, as most existing disaster-related mobile applications and research assume various disaster situations rather than just forest fire situations. Therefore, a system that can provide information quickly is needed, taking into account the characteristics of forest fires and the limitations of extinguishing work. In this paper, we propose evacuation route guidance services that bypass areas where fire has already spread, supplement existing methods of extinguishing work, and provide general information on forest fire situations in real time, by putting drones into forest fire situations. It has been implemented to automate image analysis using the Rekognition service of Amazon Web Service (AWS), and the results of fire detection and the T Map API guide the evacuation path. It is expected that the results of this paper will allow efficient and rapid rescue and extinguishing work to be carried out, and further reduce the damage of human life caused by forest fires.

5Improved Slow Charge Scheme for non-communication Electric Vehiclesby Predicting Charge Demand

저자 : Tae Uk Chang , Young Su Ryu , Ki Won Kwon , Jong Ho Paik

발행기관 : 한국인터넷정보학회 간행물 : 인터넷정보학회논문지 21권 5호 발행 연도 : 2020 페이지 : pp. 39-48 (10 pages)

다운로드

(기관인증 필요)

초록보기

Recently, the study and development of environment-friendly energy technique have increased in worldwide due to environmental pollution and energy resources problems. In vehicle industry, the development of electric vehicle(EV) is now on progress, and also, many other governments support the study and development and make an effort for EV to become widely available. In addition, though they strive to construct the EV infra such as a charge station for EV, the techniques related to managing charge demand and peak power are not enough. The standard of EV communication has been already established as ISO/IEC 15118, however, most of implemented EVs and EV charge stations do not support any communication between each of them. In this paper, an improved slow charge scheme for non-communication EVs is proposed and designed by using predicting charge demand. The proposed scheme consists of distributed charge model and charge demand prediction. The distributed charge model is designed to manage to distribute charge power depending on available charge power and charge demand. The charge demand prediction is designed to be used in the distributed charge model. The proposed scheme is based on the collected data which were from EV slow charge station in business building during the past 1 year. The system-level simulation results show that the waiting time of EV and the charge fee of the proposed scheme are better than those of the conventional scheme.

6A method of generating virtual shadow dataset of buildings for the shadow detection and removal

저자 : Kangjik Kim , Junchul Chun

발행기관 : 한국인터넷정보학회 간행물 : 인터넷정보학회논문지 21권 5호 발행 연도 : 2020 페이지 : pp. 49-56 (8 pages)

다운로드

(기관인증 필요)

초록보기

Detecting shadows in images and restoring or removing them was a very challenging task in computer vision. Traditional researches used color information, edges, and thresholds to detect shadows, but there were errors such as not considering the penumbra area of shadow or even detecting a black area that is not a shadow. Deep learning has been successful in various fields of computer vision, and research on applying deep learning has started in the field of shadow detection and removal. However, it was very difficult and time-consuming to collect data for network learning, and there were many limited conditions for shooting. In particular, it was more difficult to obtain shadow data from buildings and satellite images, which hindered the progress of the research. In this paper, we propose a method for generating shadow data from buildings and satellites using Unity3D. In the virtual Unity space, 3D objects existing in the real world were placed, and shadows were generated using lights effects to shoot. Through this, it is possible to get all three types of images (shadow-free, shadow image, shadow mask) necessary for shadow detection and removal when training deep learning networks. The method proposed in this paper contributes to helping the progress of the research by providing big data in the field of building or satellite shadow detection and removal research, which is difficult for learning deep learning networks due to the absence of data. And this can be a suboptimal method. We believe that we have contributed in that we can apply virtual data to test deep learning networks before applying real data.

7A Feature-Based Malicious Executable Detection Approach Using Transfer Learning

저자 : Yue Zhang , Hyun-hoyang , Ning Gao

발행기관 : 한국인터넷정보학회 간행물 : 인터넷정보학회논문지 21권 5호 발행 연도 : 2020 페이지 : pp. 57-65 (9 pages)

다운로드

(기관인증 필요)

초록보기

At present, the existing virus recognition systems usually use signature approach to detect malicious executable files, but these methods often fail to detect new and invisible malware. At the same time, some methods try to use more general features to detect malware, and achieve some success. Moreover, machine learning-based approaches are applied to detect malware, which depend on features extracted from malicious codes. However, the different distribution of features oftraining and testing datasets also impacts the effectiveness of the detection models. And the generation oflabeled datasets need to spend a significant amount time, which degrades the performance of the learning method. In this paper, we use transfer learning to detect new and previously unseen malware. We first extract the features of Portable Executable (PE) files, then combine transfer learning training model with KNN approachto detect the new and unseen malware. We also evaluate the detection performance of a classifier in terms of precision, recall, F1, and so on. The experimental results demonstrate that proposed method with high detection rates andcan be anticipated to carry out as well in the real-world environment.

8CNN-based Gesture Recognition using Motion History Image

저자 : Youjin Koh , Taewon Kim , Min Hong , Yoo-joo Choi

발행기관 : 한국인터넷정보학회 간행물 : 인터넷정보학회논문지 21권 5호 발행 연도 : 2020 페이지 : pp. 67-73 (7 pages)

다운로드

(기관인증 필요)

초록보기

In this paper, we present a CNN-based gesture recognition approach which reduces the memory burden of input data. Most of the neural network-based gesture recognition methods have used a sequence of frame images as input data, which cause a memory burden problem. We use a motion history image in order to define a meaningful gesture. The motion history image is a grayscale image into which the temporal motion information is collapsed by synthesizing silhouette images of a user during the period of one meaningful gesture. In this paper, we first summarize the previous traditional approaches and neural network-based approaches for gesture recognition. Then we explain the data preprocessing procedure for making the motion history image and the neural network architecture with three convolution layers for recognizing the meaningful gestures. In the experiments, we trained five types of gestures, namely those for charging power, shooting left,shooting right, kicking left, and kicking right. The accuracy of gesture recognition was measured by adjusting the number of filters in each layer in the proposed network. We use a grayscale image with 240 x 320 resolution which defines one meaningful gesture and achieved a gesture recognition accuracy of 98.24%.

9Consumers attitude towards Internet banking services in an underdeveloped country: A case of Pokhara, Nepal

저자 : Deepanjal Shrestha , Tan Wenan , Neesha Rajkarnikar , Seung Ryul Jeong

발행기관 : 한국인터넷정보학회 간행물 : 인터넷정보학회논문지 21권 5호 발행 연도 : 2020 페이지 : pp. 75-85 (11 pages)

다운로드

(기관인증 필요)

초록보기

The application of Internet technology has created enormous impact on banking sector with the implementation of many techno-oriented services like Internet banking, EFT, branchless banking, Automated Clearing House (ACH) transactions etc. Study of customer's attitude in terms of trust, perceived risk and ease of use of a particular technology is as an important parameter for acceptance or rejection of a technology. To explore the customers'attitude for Internet banking this research is undertaken. The research is carried out in Pokhara valley which is the second largest city and tourism capital of Nepal. The study employs descriptive research design with stratified sampling procedure for eight top commercial banks. A set of 25 customers is taken from each selected 8 banks making a sample size of 200 respondents. A fixed set of question related to demographic factors is provided personally or by visiting the location of the customers of Internet banking service and collected accordingly. Reliability test is performed using Cronbach's alpha and data is analyzed using inferential statistics to present the results of the study. This study provides knowledge on the current scenario of Internet banking and helps banks in cost saving, mass customization, product innovation, improved marketing and communication. This study is very important for financial institutions like banks, government agencies and business houses to understand the perception of customers towards Internet banking and technology as a whole. The study also supplements the gap in literature on technology and banking in Nepal and serves as an important knowledge base.

10Determinants of Continuance Intention in Mobile Payment Services: Based on the IS Success Model

저자 : Viyada Itthiphone , Donghyuk Jo , Chulhwan Kwon

발행기관 : 한국인터넷정보학회 간행물 : 인터넷정보학회논문지 21권 5호 발행 연도 : 2020 페이지 : pp. 87-95 (9 pages)

다운로드

(기관인증 필요)

초록보기

Highly competitive environment has been forcing e-commerce industries to seek strategies to achieve competitive advantages. Mobile payment is a kind of service that allows mobile phone user to easily and conveniently initiate payments and transfer funds using their mobile phone anytime, and anywhere. This study is designed to identify factors that affect the intention of continued use of mobile payment services between users in Korea and Laos.
As a result, first, in the case of Korean consumers, system quality, information quality and service quality were shown to have a positive effect on trust and satisfaction. In addition, trust and satisfaction were shown to have a positive effect on continuance intention. Second, in the case of Laotian consumers, system quality and service quality were shown to have a positive effect on trust, and system quality and information quality were shown to have a positive effect on satisfaction. In addition, trust and satisfaction were shown to have a positive effect on continuance intention.
The study has its implications by analyzing factors affecting the continuance intention with the comparison of the customers from a developed nation and a developing nation, providing a direction of development for developing competitive advantages for those in development. For the developed, the study provides a guideline of what to modify and supplement in cases of entering the markets of developing nations.

12
주제별 간행물
간행물명 수록권호

한국감성과학회 추계학술대회
2020권 0호 ~ 2020권 0호

KCI등재 SCI SCOUPUS

KSII Transactions on Internet and Information Systems (TIIS)
15권 1호 ~ 15권 1호

KCI후보

한국화상학회지
26권 3호 ~ 26권 3호

KCI등재

인터넷정보학회논문지
21권 6호 ~ 21권 6호

KCI등재

한국안전학회지(구 산업안전학회지)
35권 6호 ~ 35권 6호

KCI등재

한국산업융합학회 논문집
23권 6호 ~ 23권 6호

KCI등재

공학기술논문지
13권 4호 ~ 13권 4호

복합신소재구조학회지
11권 4호 ~ 11권 4호

KCI등재

감성과학
23권 4호 ~ 23권 4호

KCI등재

복합신소재구조학회논문집
11권 6호 ~ 11권 6호

KCI등재 SCI SCOUPUS

KSII Transactions on Internet and Information Systems (TIIS)
14권 12호 ~ 14권 12호

KCI등재 SCI SCOUPUS

KSII Transactions on Internet and Information Systems (TIIS)
14권 11호 ~ 14권 11호

KCI등재 SCI SCOUPUS

KSII Transactions on Internet and Information Systems (TIIS)
14권 10호 ~ 14권 10호

KCI등재

한국산업융합학회 논문집
23권 5호 ~ 23권 5호

KCI등재

인터넷정보학회논문지
21권 5호 ~ 21권 5호

KCI등재

복합신소재구조학회논문집
11권 5호 ~ 11권 5호

KCI등재

한국안전학회지(구 산업안전학회지)
35권 5호 ~ 35권 5호

KCI등재

감성과학
23권 3호 ~ 23권 3호

KCI등재 SCI SCOUPUS

KSII Transactions on Internet and Information Systems (TIIS)
14권 9호 ~ 14권 9호

KCI후보

한국화상학회지
26권 2호 ~ 26권 2호
발행기관 최신논문
자료제공: 네이버학술정보
발행기관 최신논문
자료제공: 네이버학술정보

내가 찾은 최근 검색어

최근 열람 자료

맞춤 논문

보관함

내 보관함
공유한 보관함

1:1문의

닫기