Earticle

인공지능 기술의 발달로 자율주행, 의료 영상 분석 등 다양한 분야에서 인공지능이 널리 활용되고 있지만, 이와 동 시에 적대적 공격에 대한 위협이 점차 증가하고 있다. 적대적 공격은 사람에게는 감지되지 않는 미세한 변화를 통해 인공지능 모델의 예측을 왜곡시키며, 특히 인공지능의 신뢰성이 중요한 분야에서 심각한 문제를 초래할 수 있다. 이 러한 문제를 해결하기 위해, 본 연구에서는 샘플링 기반 일관성 탐지 기법을 활용한 새로운 적대적 공격 탐지 방안 을 제안한다. 입력된 영상의 주변부를 샘플링하여 분류한 결과의 일관성을 확인하고 이를 통해 입력된 영상의 공격 여부를 판단한다. MNIST 데이터 세트에서의 실험을 통해 제안한 탐지 기법이 FGSM, BIM, PGD, DeepFool, C&W 등 다양한 적대적 공격을 성공적으로 탐지할 수 있음을 실험하였고, 특히 C&W 공격에 대해 98%의 탐지 성능을 확인하였다.

Advances in AI technology are enabling its widespread use in various fields, such as autonomous driving and medical image analysis. However, they are also increasing the threat of adversarial attacks. Adversarial attacks subtly distort the predictions of AI models in ways undetectable to humans, potentially leading to severe consequences, particularly in fields where AI reliability is essential. To address these issues, this research proposes a new method for detecting adversarial attacks using a sampling-based consistency detection technique. By sampling around the input image, the consistency of the classification results is checked to determine whether the input image has been attacked. Experimental results on the MNIST dataset show that the proposed detection method can effectively detect adversarial attacks, achieving a 98% detection rate for C&W attacks.

본 연구는 자동차주행거리 추정과정에서 발생할 수 있는 오차발생 원인을 규명하였다. 그리고 각 원인이 자동차주행거리 추정 정확 도에 미치는 영향을 오차율로 정량화하여 효율적인 자동차주행거리 추정방안을 제시하였다. 이를 위한 연구과정은 다음과 같다. 첫째, 시범조사 지역을 대상으로 자동차주행거리 추정 방법론의 정확도를 검증하기 위한 자동차주행거리 관측 자료를 구축하였다. 둘째, 자 동 차주행거리 추정 오차발생 원인은 표본크기, 표본추출방법, 단위구간 설정방법의 오류로 구분하였다. 그리고 각 원인에 따른 자동차주 행거리 추정오차를 최소화하기 위한 다양한 방법론을 설정하였다. 셋째, 각 방법론에 의한 자동차주행거리 추정 오차율을 비교분석 하 였다. 마지막으로 Toy-Network를 구축하여 지역특성을 고려한 자동차주행거리 추정방안을 제시하였다. 본 연구는 실험 계획적 접근방법을 통하여 효율적인 자동차주행거리 추정방안을 제시하였으며, 추정 정확도 검증을 위하여 자동차 주행거리 관측 자료를 활용했다는 점에서 의의를 갖는다. 또한 본 연구에서 제시한 자료수준과 지역특성을 고려한 자동차주행거리 추 정 방안은 향후 지역별 자동차주행거리 추정에 기여할 것으로 판단된다.

This study identified the causes of errors that could take place in the estimation process of vehicle miles traveled and quantified the effects of each of those causes on the estimation accuracy of vehicle miles traveled via error rate to propose an efficient way to estimate vehicle miles traveled. The study proceeded as follows: first, the study established survey data of vehicle miles traveled in the pilot test areas to test the accuracy of a method to estimate vehicle miles traveled. Second, the causes of errors with the estimation of vehicle miles traveled were categorized into errors with the sample size, sampling methods, and homogeneous link setting methods. In addition, many different methodologies were set to minimize errors with the estimation of vehicle miles traveled according to each of the causes. Third, error rates of estimation of vehicle miles traveled were compared and analyzed according to each of the methodologies. Finally, a toy network was established to propose a way of estimating vehicle miles traveled by taking the local characteristics into consideration. The study finds its significance in that it proposed an efficient way to estimate vehicle miles traveled through an experiment and planning approach and made use of survey data of vehicle miles traveled to test estimation accuracy. The proposed way of estimating vehicle miles traveled by taking into account the local characteristics will make a contribution to the estimation of vehicle miles traveled by the areas in future along with the level of data offered in the study.

다른 산업과 달리 병원/의료 산업에서는 건강 보험료 심사 평가라는 독특한 검증 과정이 필수적으로 있게 된다. 건강 보험료 심사 평가는 병원의 수익 문제 뿐 아니라 적정한 진료행위를 하는 병원이라는 이미지와도 맞물려 매우 중요한 분야이며, 특히 대형 종합병원일수록 이 부분에 많은 심사관련 인력들을 투입하여, 병원의 수익과 명예를 위해서 업무를 수행하고 있다. 본 논문은 이러한 건강보험료 청구 심사 과정에서, 사전에 수많은 진료 청구 건 중 심사 평가에서 삭감이 될 수 있는 진료 청구 건을 데이터 마이닝을 통해서 발견하여, 사전의 대비를 철저히 하고자 하는 한 국내 대형 종합병원의 사례를 소개하고자 한다. 데이터 마이닝을 적용함에 있어, 주요한 문제점 중 하나는 바로 지도학습 기법을 적용하기에 곤란한 데이터 불균형 문제가 발생하는 것이다. 이런 불균형 문제를 해소하고, 비교 조건 중에 가장 효율적인 삭감 예상 진료 건 탐지 모델을 만들어 내기 위하여, 데이터 불균형 문제의 기본 해법인 Sampling과 오분류 비용의 다양한 혼합적인 적용을 통하여, 적합한 조건을 가지는 의사결정 나무 모델을 도출하였다.

In medical industry, health insurance bill audit is unique and essential process in general hospitals. The health insurance bill audit process is very important because not only for hospital's profit but also hospital's reputation. Particularly, at the large general hospitals many related workers including analysts, nurses, and etc. have engaged in the health insurance bill audit process. This paper introduces a case of health insurance bill audit for finding reducible health insurance bill cases using decision tree induction techniques at a large general hospital in Korea. When supervised learning methods had been tried to be applied, one of major problems was data imbalance problem in the health insurance bill audit data. In other words, there were many normal(passing) cases and relatively small number of reduction cases in a bill audit dataset. To resolve the problem, in this study, well-known methods for imbalanced data sets including over sampling of rare cases, under sampling of major cases, and adjusting the misclassification cost are combined in several ways to find appropriate decision trees that satisfy required conditions in health insurance bill audit situation.

본 연구는 Gibbs sampler와 MTDFREML 방법에 의해서 한우 도체형질의 유전력 및 유전(공)분산을 단형질 및 다형질 개체모형을 가지고 추정하고 비교 하였다. 배장근단면적(longissimus dorsi area), 등지방 두께(backfat thickness), 상강도(marbling score)의 유전 모수를 추정하였으며, 분석에 이용된 자료는 총 1,941두 이고, 혈연계수를 구하기 위한 혈통 자료는 23,058두를 이용하였다. 도체형질에 대한 유전력 추정 시 단형질과 다형질 개체모형에 의한 편차는 크게 나타나지 않았다. 단형질 개체모형에서 Gibbs sampler 방법을 이용한 추정에서는 LDA, BF 및 MS에서 각각 0.52, 0.59 및 0.42로서 고도의 유전력을 보였다. MTDFREML 방법을 통한 추정 시에는 LDA 0.41, BF 0.52로서 고도의 유전력을 보였으며, MS는 0.32로서 중도의 유전력을 보였다. 분석 방법에 의한 유전력 추정은 Gibbs sampler에 의한 방법이 MTDFREML에 의한 방법에 비해서 0.1정도 높게 추정되었다. MTDFREML 방법과 Gibbs Sampler 방법에 의한 도체 형질간의 유전상관은 LDA와 BF, MS 간에는 모두 부의 상관을 보였고, BF과 MS에서는 정의 상관을 보였다. MTDFREML과 Gibbs Sampler에서 이들의 분석 방법 간에 육종가 추정치에 대한 상관계수는 LDA와 BF에서는 0.989 이상의 높은 추정치를 보였으나, MS에 대해서는 이보다 다소 낮은 0.985를 나타내었다. 그리하여, 상강도(marbling score)와 같은 범주형 자료에 대한 유전분석은 기존의 선형의 정규분포를 가정한 REML방법에 의한 것 보다 범주형 모형을 설정하여 Gibbs sampling algorithm을 응용한 분석방법이 더 적합할 것으로 사료된다.

The objective of this study was to compare of genetic parameter estimates on carcass traits of Hanwoo(Korean Cattle) according to modeling with Gibbs sampler and MTDFREML. The data set consisted of 1,941 cattle records with 23,058 animals in pedigree files at Hanwoo Improvement Center. The variance and covariance among carcass traits were estimated via Gibbs sampler and MTDFREML algorithms. The carcass traits considered in this study were longissimus dorsi area, backfat thickness, and marbling score. Genetic parameter estimates using Gibbs sampler and MTDFREML from single-trait analysis were similar with those from multiple-trait analysis. The estimated heritabilities using Gibbs sampler were .52～.54, .54 ～.59, and .42～.44 for carcass traits. The estimated heritabilities using MTDFREML were .41, .52～.53, and .31～.32 for carcass traits. The estimated genetic correlation using Gibbs sampler and MTDFREML of LDA between BF and MS were negatively correlated as .34～.36, .23～.37. Otherwise, genetic correlation between BF and MS was positive genetic correlation as .36～.44. The correlations of breeding value for marbling score between via MTDFREML and via Gibbs sampler were 0.989, 0.996 and 0.985 for LDA, BF and MS respectively.

본 연구는 한우의 도체형질들에 대한 유전적 변이를 분석방법에 따라 어떠한 차이가 있는지를 알아보고자 한우 후대검정자료를 이용하여 분석하였다. 분석에 이용된 도체성적 관련 자료는 가축개량사업소와 한우개량단지에서 1996년부터 2001년까지 태어난 후대검정우 1526두로부터 조사었다. 한우에 있어서 주요 개량형질인 육량과 육질에 영향하는 도체중, 도체율, 배최장근단면적, 등지방두께, 근내지방도를 대상으로 유전력과 유전상관을 추정하였다. 유전모수 추정에 있어서 REML 분석법과 Bayesian추론방법으로써 Gibbs Sampling 분석법을 사용하였는데 각각의 모수들에 대한 Gibbs Sampler는 100,000회 실시하였고 초기 1,000회는 모수의 사후분포에 대한 부정확성으로 간주하여 모수의 사후분포특성 규명에서 제외하였다. 각각의 형질들에 대한 유전변이는 이러한 두 가지 분석방법을 상호 비교 함으로써 최적의 통계분석방법을 모색하였다.도체형질에 대한 유전력 추정은 REML 방법을 통한 추정에서는 근내지방도가 0.51로 고도의 유전력을 보였고, 도체율이 0.25로 중도의 유전력이 추정되었다. Gibbs Sampling 방법을 통한 추정에서도 근내지방도가 0.54로 고도의 유전력을 도체율에서 0.25로 중도의 유전력을 보였다. REML 분석방법과 Gibbs Sampling 분석방법에서의 유전력은 다소 큰 차이는 보이지 않았으나, 대체로 Gibbs Sampling 방법으로 분석한 유전력 추정치가 높게 추정되었다.그리고, 유전상관분석에서는 REML 방법을 통한 분석에서 도체중과 배최장근단면적에서 0.651로 높은 정의 유전상관을 나타내었고, 배최장근단면적과 등지방두께에서 -0.139로 부의 유전상관을 나타내었다. Gibbs Sampling 방법에서는 도체중과 도체율, 배최장근단면적에서 각각 0.814, 0.695으로 높은 정의 상관을 나타내었고, 배최장근단면적과 등지방두께에서 -0.126으로 부의상관을 나타내었다. Gibbs Sampling 방법을 통한 분석에서 특정 형질간 유전상관이 높게 나타났으며, 다소 큰 차이를 보이지 않았다.REML 분석방법과 Bayesian Inference를 바탕으로 한 Gibbs Sampling 분석에서 모수 추정은 큰 차이를 보이지 않았다. 따라서 방대한 현장정보를 활용하여 보다 정확한 모수 추정을 수행하기 위해서는 분석모형에 대한 계수행렬의 역행렬 계산을 통한 REML 방법의 한계성을 극복할 수 있는 방법으로써 Gibbs Sampling 분석 방법이 가능할 것으로 사료된다.

The aims of this study were to estimate genetic parameters for carcass traits on Hanwoo(Korean Native Cattle) and to compare two different statistical algorithms for estimating genetic parameters. Data obtained from 1526 steers at Hanwoo Improvement Center and Hanwoo Improvement Complex Area from 1996 to 2001 were used for the analyses. The carcass traits considered in these studies were carcass weight, dressing percent, eye muscle area, backfat thickness, and marbling score. Estimated genetic parameters using EM-REML algorithm were compared to those by Bayesian inference via Gibbs Sampling to find out statistical properties. The estimated heritabilities of carcass traits by REML method were 0.28, 0.25, 0.35, 0.39 and 0.51, respectively and those by Gibbs Sampling method were 0.29, 0.25, 0.40, 0.42 and 0.54, respectively. This estimates were not significantly different, even though the estimated heritabilities by Gibbs Sampling method were higher than ones by REML method.Since the estimated statistics by REML method and Gibbs Sampling method were not significantly different in this study, it is inferred that both mothods could be efficiently applied for the analysis of carcass traits of cattle. However, further studies are demanded to define an optimal statistical method for handling large scale performance data.

목적 : 본 연구는 work sampling 기법과 단면분석법을 이용하여 제조업 종사 근로자의 근골격계질환 노출 위험수준을 알아보고자 하였다. 연구방법 : 이를 위해 자동차부품 제조업 종사 근로자 29명의 근골격계질환 증상을 자기기입식 설문지로 조사하여, 그 정도를 기준1(KOSHA), 기준2(NIOSH), 기준3(인천노동과학연구소)으로 구분하였다. 그리고 개별 근로자의 작업과정을 고성능카메라로 촬영한 후 OWAS, RULA, REBA를 사용하여 work sampling 기법(1초당 1컷)과 단면분석법으로 근골격계질환 위험수준을 분석하였다. 결과 : 신체부위 중 1부위 이상 통증이 있다고 호소한 근로자는 기준1 24명(82.8%), 기준2 22명(75.9%), 그리고 기준3 19명(65.5%)이었다. 인간공학적 측정도구를 이용한 근골격계질환 노출위험수준은 OWAS는 단면분석-조치단계 4(근골격계에 매우 심한 해를 끼침), work sampling-조치단계 2(근골격계에 약간의 해를 끼침)인 근로자가 3명이었다. 또, RULA는 단면분석-조치단계 4(정밀조사와 즉각적인 개선필요), work sampling-조치단계 3(계속적 관찰과 빠른 작업개선 요함)인 근로자가 2명이었고, REBA에서는 단면분석-조치단계 3(위험 높음), work sampling-조치단계 2(위험 보통)인 근로자가 3명으로 분석되었다. 단면분석과 work sampling 기법에서 고위험군과 저위험군의 일치도는 OWAS Kappa=.4423, RULA Kappa=.4938, 그리고 REBA Kappa=.5193이었다. 결론 : Work sampling 기법을 이용하여 근로자의 근골격계질환 노출위험수준을 객관적으로 분석할 수 있었으며, 결과를 토대로 근로자 작업환경의 개선이 우선적으로 필요한 작업자를 선정하는데 신뢰할만한 방법임이 확인되었다.

Objective : Using a work sampling technique and cross-section analysis, this study was performed in order to analyze the risk level of musculoskeletal disorders for workers engaged in the manufacturing industry. Methods : Using a self-entering mode questionnaire, the study was conducted to survey the musculoskeletal disorders of 29 workers at an automative component factory. The extent of the musculoskeletal disorders of the workers was classified into basis 1(KOSHA), basis 2(NIOSH), and basis 3 (labor scientific research institute of Incheon university). And the risk levels of the musculoskeletal disorders were evaluated using OWAS, RULA, and REBA for each work posture, extracting 1 cut per 1 sec, based on the work sampling technique, after having photographed the work processes of the individual workers with a high performance camera. Results : The number of workers who complained of pain in one or more parts of their body were 24 persons (82.8%) in basis 1, 22 persons (75.9%) in basis 2, and 19 persons (65.5%) in basis 3. As a result of having analyzed the musculoskeletal disorders, OWAS was in risk level 4 in a cross-section analysis (inflicted with severe musculoskeletal disorders), and as for the work sampling technique, the workers in risk level 2 (inflicted with milder musculoskeletal disorders) were 3 persons. Also, the workers whose work sampling of risk level had been 3 (continuous observation and fast work improvement were required) while their cross-section analysis for RULA was at risk level 4 (precision survey and immediate improvement were required) were 2 persons, while in REBA, the workers whose work sampling risk level was 2 (risk was normal) while their cross-section analysis risk levels were 3 (risk was high) were analyzed as 3 persons. The consistencies of the work sampling and cross-section analysis in the high-risk group and the low-risk group are as follows; OWAS (Kappa=.4423), RULA (Kappa=.4938), and REBA (Kappa=.5193). Conclusion : The results showed that this study may be used to analyze the exposure risk level to the musculoskeletal disorders of workers using a work sampling technique, and it has been confirmed as a trusted method in selecting the workers who require a preferential improvement of their work environment.

In-band network telemetry (INT) is an emerging network monitoring framework based on a protocol-independent packet processor (P4). Network devices can be programmed with a domain-specific language to embed switch-internal states into data packets as they traverse through networks. However, the current P4-based INT does not support a sampling; thus, INT headers should be augmented for all incoming packets, which will incur high overhead in a large-scale network. In this paper, we propose a flexible sampling-based INT (FS-INT) scheme to address the aforementioned issues. Simulation results show that FS-INT can reduce the protocol overhead in a controlled manner while providing sufficiently high accuracy.

In-band network telemetry (INT) is one of the promising technologies for achieving fine-grained and real-time visibility into the network, however, it brings non-negligible traffic overhead due to the increased packet length. Several sampling methods have been introduced to reduce the INT overhead; although, the provision of satisfactory network visibility was not well investigated. In this paper, we introduce a utility-based INT sampling rate allocation scheme, namely uINT, that aims to reduce data redundancy while improving network visibility. We conducted experiments using the software switches and demonstrated that uINT reduces redundantly collected INT data by up to 15% compared to sINT.

Due to the existence of computer and development of its technology, anybody may make new music with the help of the synthesizer which is not expensive, and the problems about the digital sampling of sound recording have arisen in Korea. However, Korean courts have not decided any case because most issues about digital sampling have been settled before going to court. Therefore, this article introduces some U.S. cases and BGH cases to help Korean courts to decide in the future. Finally, this article argues the de minimis use test and fair use doctrine in U.S. courts should be applied to most digital sampling case to promote the progress of culture and Korean courts should not follow the BGH reasoning because, in most digital sampling cases, most people may not recognize the original music in the new music.

Purpose: To compare patient care multipliers estimated from subjective evaluation against work sampling (WS) techniques in genetic nursing activities. Methods: An observational WS technique was conducted from November to December 2019 with nine genetic nurses in a tertiary referral center in Malaysia. The WS activity instrument was devised, validated, and pilot tested. All care- and non-care-related activities were sampled at 10-minute intervals within 8 hours of working over 14 days, followed by a subjective evaluation of activities survey over the same period. Bonferroni correction was undertaken for multiple testing with a p value of 0.0025. Results: The two techniques produced significant differences in genetic nurses’ activities categorization. The WS showed that compared with subjective evaluation, direct care (19.3% vs. 45.0%; p < .001) was estimated to be significantly lower, and indirect care (40.4% vs. 25.6%; p < .001) and unit-related activity (28.5% vs. 16.9%; p < .001) were higher. Both techniques produced a similar proportion of time spent in other non-care activities (12.0%) but differed in genetic meetings and information-gathering activities. While the multipliers for patient face-to-face contact were significantly larger between WS (4.57) and the survey (1.94), the multipliers for patient care time were smaller between WS (1.47) and the survey (1.24), indicating that caution should be taken when multiplying for patient contact time compared to patient care activity to determine the cost of care provision. Conclusion: A considerable proportion of time spent away from the patient needs to be allocated to patient-related care time. Thus, estimating the paid cost solely based on direct time with patients considerably underestimates the cost per hour of nurses' care. It is recommended to employ ‘patientrelated activity’ instead of the ‘face-to-face contact’ multiplier because the former did not significantly differ from the one estimated using WS.

This study was performed to evaluate the microbial and temperature changes of boxed beef during transport and distribution under vacuum and modified atmosphere packaging (MAP), and to compare between excision and swab sampling for 15 days. The top round and striploin (quality grade 1) from Hanwoo steers at 2 days post-slaughter were obtained from a local meat processing plants and chilled at 4 ± 2°C in a cold room. The boxes were transported under refrigeration (4 ± 2°C) to the laboratory within half an hour. Vacuum and MAP packs were subsequently taken out from cool boxes, and microbiological examinations were carried out at 0, 6, 12, and 24 h of storage time. MAP was more effective than vacuum packaging for the inhibition of total aerobic, lactic acid bacteria and Pseudomonas (p < 0.05). Microbial loads of swab methods were slightly lower than those of excision ones (p < 0.05). The results of this study could be utilized by meat consumers in future studies as well as by manufacturers to determine the ideal storage conditions for cool boxed meat, thus ensuring reduced economic losses due to spoilage.

In this paper, we describe the design and evaluation of a second-generation front-end unit for wideband sampling radio receivers. The unit contains a surface acoustic wave (SAW) filter to protect the receiver from strong out-of-band signals, an RF limiter to protect both the filter and the receiver from physical damage due to strong signals, and a bias tee with a DC limiter to provide DC power to a masthead low-noise amplifier, if one is used. The unit allows receivers such as those of the universal software radio peripheral (USRP) N-series type to be effectively used in RF environments with weak signals and strong in-band and out-of-band interferences.

In this paper, we will consider the problem of multi-unmanned aerial vehicles (UAVs) deployment among users’ hotspots. UAVs carry reconfigurable intelligent surface (RIS) boards to strengthen millimeter wave (mmWave) coverage at these hotspots. UAVs should maximize hotspots’ sum data rates while minimizing their flights’ cost, i.e., hovering and flying energy consumptions. Moreover, collisions should be avoided among UAVs, i.e., one hotspot should be served by only one UAV at a time. In this paper, this problem is considered as a multi-player multi-armed bandit (MP-MAB) game with budget constraint and collision avoidance. In this context, budget constraint Thompson sampling (TS) with collision avoidance MAB algorithm (BTSCA-MAB) is proposed to efficiently implement the formulated MP-MAB game. Numerical analysis confirms the superior performance of the proposed BTSCA-MAB over other benchmarks.

Background: The Korean Peninsula is split between the South Korea and Democratic People’s Republic of Korea (DPRK, North Korea), with North Korea possessing nuclear weapons. There are rising concerns over legacy sites, areas contaminated primarily by nuclear activities. In North Korea, these sites lack regulatory oversight and are likely neglected even if inactive. Hence, it is vital to devise strategies to manage and assess these sites. Yongbyon, a prominent nuclear site in North Korea, has been proposed for environmental sampling to gauge contamination levels. The goal is to determine contamination by suggesting sample collection points in Yongbyon. Materials and Methods: Using recent data and satellite imagery from the International Atomic Energy Agency’s periodic Safety Measures Report, we selected legacy sites to assess their contamination levels with recent data and satellite imagery. We identified sampling locations using the Visual Sample Plan (VSP) to check nuclear contamination. From these results, we devised a scenario evaluating accessibility and the local environment. Results and Discussion: For Yongbyon in North Korea, this study created two sampling scenarios based on interior access feasibility. Given North Korea’s constraints, we finalized the scenario without interior access. The sampling areas include two steam lines, four vehicle paths, and five green zones outside the facility. Conclusion: Nuclear facility operations worldwide are concerning. When halted, these become legacy sites requiring contamination management. Despite North Korea’s limited recordkeeping and challenges in determining contamination, we devised a sampling scenario using VSP software, factoring in accessibility and movement paths.