• 정보교육학회논문지
• /
• 제26권4호
• /
• pp.273-284
• /
• 2022

본고의 목표는 데이터 편향 인식 교육에서 기계학습 플랫폼의 사용을 제안하는 것이다. 학습자들이 인공지능 데이터 및 시스템을 다루거나 인공지능윤리 요소 중 데이터 편향에 의한 피해를 방지하고자 할 때 인지할 수 있는 역량을 배양할 수 있다. 구체적으로, 머신러닝포키즈를 활용해 데이터편향 학습을 하는 방법을 AI야구심판 사례를 통해 제시한다. 학습자는 구체적 주제선정, 선행연구 검토, 기계학습 플랫폼에서 편향/비편향 데이터의 입력 및 테스트 데이터 구성, 기계학습의 결과 비교, 결과를 통해 얻을 수 있는 데이터 편향에 대한 함의를 제시한다. 이러한 과정을 통해서 학습자는 인공지능 데이터 편향이 최소화되어야 한다는 점과 데이터 수집 및 선정이 사회에 미치는 영향을 체험적으로 배울 수 있다. 이 학습방법은 문제기반의 자기주도 학습의 용이성, 코딩교육과의 결합가능성, 그리고 인문사회적 주제와 인공지능 리터러시와 결합을 추동한다는 의의를 가진다.

• 대한한방소아과학회지
• /
• 제37권1호
• /
• pp.1-14
• /
• 2023

Objectives This study aimed to evaluate the effect of Samryungbaekchul-san on childhood anorexia. Methods We searched 11 English, Korean, Chinese, and Japanese databases for studies published up to May 30, 2022. Randomized controlled trials (RCTs) assessing the effect of Samryungbaekchul-san on childhood anorexia were included. In the meta-analysis, relative risk (RR) and 95% confidence intervals (CI) were indicated as dichotomous variables, and mean difference (MD) or standardized mean difference (SMD) and 95% CIs were indicated as continuous variables. Results We included 12 RCTs with 1345 participants. The Samryungbaekchul-san treatment group had a significantly higher total effective rate (TER) than that of the western medicine control group (RR 1.42, 95% CI 1.23-1.64, I² = 0%). The combined Samryungbaekchul-san and western medicine treatment group had significantly higher TER (RR 1.31, 95% CI 1.23 to 1.40, I² = 0%) and levels of neuropeptide Y (SMD 0.93, 95% CI 0.47-1.39, I² = 70%) and ghrelin (SMD 1.45, 95% CI 1.14-1.76, I² = 0%) than those of the western medicine alone group. Additionally, leptin levels were significantly lower in the combined treatment group (SMD -1.19, 95% CI -1.88 to -0.51, I² = 86%) compared with the western medicine alone group, although statistical heterogeneity was substantial. Conclusions Samryungbaekchul-san may be effective for childhood anorexia. However, owing to limitations such as high clinical heterogeneity between the studies, unclear risks of biases, and insufficient reports of adverse events and follow-ups, well-designed RCTs with a low risk of bias are needed in the future.

• 한국수자원학회:학술대회논문집
• /
• 한국수자원학회 2023년도 학술발표회
• /
• pp.161-161
• /
• 2023

Bias correction of values is a necessary step in downscaling coarse and systematically biased global climate models for use in local climate change impact studies. In addition to univariate bias correction methods, many multivariate methods which correct multiple variables jointly - each with their own mathematical designs - have been developed recently. While some literature have focused on the inter-comparison of these multivariate bias correction methods, none have focused extensively on the effect of diverse configurations (i.e., different combinations of input variables to be corrected) of climate variables, particularly high-dimensional ones, on the ability of the different methods to remove biases in uni- and multivariate statistics. This study evaluates the impact of three configurations (inter-variable, inter-spatial, and full dimensional dependence configurations) on four state-of-the-art multivariate bias correction methods in a national-scale domain over South Korea using a gridded approach. An inter-comparison framework evaluating the performance of the different combinations of configurations and bias correction methods in adjusting various climate variable statistics was created. Precipitation, maximum, and minimum temperatures were corrected across 306 high-resolution (0.2°) grid cells and were evaluated. Results show improvements in most methods in correcting various statistics when implementing high-dimensional configurations. However, some instabilities were observed, likely tied to the mathematical designs of the methods, informing that some multivariate bias correction methods are incompatible with high-dimensional configurations highlighting the potential for further improvements in the field, as well as the importance of proper selection of the correction method specific to the needs of the user.

• 대기
• /
• 제33권4호
• /
• pp.423-440
• /
• 2023

The updated version of Global Ocean Data Assimilation and Prediction System (GODAPS) in the NIMS/KMA (National Institute of Meteorological Sciences/Korea Meteorological Administration), which has been in operation since December 2021, is being introduced. This technical note on GODAPS2 describes main progress and updates to the previous version of GODAPS, a software tool for the operating system, and its improvements. GODAPS2 is based on Forecasting Ocean Assimilation Model (FOAM) vn14.1, instead of previous version, FOAM vn13. The southern limit of the model domain has been extended from 77°S to 85°S, allowing the modelling of the circulation under ice shelves in Antarctica. The adoption of non-linear free surface and variable volume layers, the update of vertical mixing parameterization, and the adjustment of isopycnal diffusion coefficient for the ocean model decrease the model biases. For the sea-ice model, four vertical ice layers and an additional snow layer on top of the ice layers are being used instead of previous single ice and snow layers. The changes for data assimilation include the updated treatment for background error covariance, a newly added bias scheme combined with observation bias, the application of a new bias correction for sea level anomaly, an extension of the assimilation window from 1 day to 2 days, and separate assimilations for ocean and sea-ice. For comparison, we present the difference between GODAPS and GODAPS2. The verification results show that GODAPS2 yields an overall improved simulation compared to GODAPS.

• 한국임상약학회지
• /
• 제33권3호
• /
• pp.195-201
• /
• 2023

Background: Hematocrit is usually measured from venous blood collected by invasive venipuncture. This study was performed to determine hematocrit accurately and precisely using minimally invasive volumetric absorptive microsampling (VAMS) technique. Such technique is to be applied to determining hematocrit in various clinical settings for the care, including therapeutic drug monitoring, of neonatal or epileptic patients, or patients with high risk of infection or bleeding. Methods: The study was performed using 31 VAMS samples obtained from 21 pancreatic cancer patients. Hematocrit was determined using the values of potassium concentrations obtained from blood in VAMS tips (Hct_VAMS). Hct_VAMS was compared with hematocrit measured from blood collected by venipuncture (Hct_VP). The accuracy and precision of Hct_VAMS in comparison to Hct_VP were evaluated using Bland-Altman plot, Deming regression and mountain plot. Results: Bland-Altman plot displayed a random scattering pattern of the differences between Hct_VAMS and Hct_VP with the mean bias of -0.010 and the 95% limit of agreement ranging from -0.063 to 0.044. Deming regression for Hct_VAMS and Hct_VP line demonstrated very small proportional and constant biases of 1.04 and -0.003, respectively. Mountain plot exhibited a narrow and symmetrical distribution of the differences with their median of -0.011 and central 95% range from -0.049 to 0.033. Conclusion: Hematocrit was accurately and precisely determined using less invasive VAMS technique. Such technique appears to be applicable to determining hematocrit in situations that venipuncture is not favorable or possible.

• 한국심리학회지:법
• /
• 제11권1호
• /
• pp.63-87
• /
• 2020

판단과 의사결정 연구에서 사람들이 판단 대상과 무관한 정보에 의해 영향을 받아, 예측 가능한 방향으로 편향이 일어날 수 있다. 이런 편향은 법적 판단에서도 나타난다는 연구도 많다. 그 중 하나는 피고인의 성격증거에 의해 유도된 편향이다. 본 연구에서는 성격증거로 편향을 유도한 다음, 판단자의 사고를 촉진하는 활동을 통해 편향을 감소시키는 방안을 모색하였다. 실험 1에서는 대학생 121명을 대상으로 하여 토론, 반사실적 사고와 토론, 그리고 반사실적 사고와 동료평가를 한 경우로 나누어 어떤 방법이 피고인의 성격증거로 유도된 편향을 줄이는데 효과가 있는지를 알아보았다. 연구결과 탈편향 활동을 한 집단은 통제 집단보다 유의미하게 편향이 줄어들었지만, 세 가지 다른 탈편향 활동을 한 집단들 간에는 감소량에서 차이가 없었다. 동일한 설계와 절차로 일반인 125명을 대상으로 실시한 실험 2에서는, 대학생 집단과는 달리, 반사실적사고와 토론을 병행한 집단에서만 유의미하게 편향이 줄어들었다. 종합 논의에서는 대학생과 일반인 간에 왜 이런 차이가 나타났는지에 대한 탐색과 연구의 한계점, 그리고 향후 연구방향에 대해 다루었다. 본 연구는 탈편향 전략이 피고인의 성격증거로 인해 발생할 수 있는 오판을 축소시킬 수 있음을 확인하였다는 점에서 의의가 있다.

• 한국전자통신학회논문지
• /
• 제18권3호
• /
• pp.449-458
• /
• 2023

많은 사람들이 수화는 청각 장애가 있고 말을 할 수 없는 사람들을 위한 것이라고 생각하지만 물론 그들과 대화하고 싶은 사람들에게 필요하다. ASL(: American Sign Language) 알파벳 인식에서 가장 큰 문제 중 하나는 높은 클래스 간 유사성과 높은 클래스 내 분산이다. 본 논문에서는 이 두 가지 문제점을 극복할 수 있는 유사도 학습을 수행하여 이미지 간의 클래스 간 유사도와 클래스 내 분산을 줄이는 아키텍처를 제안하였다. 제안된 아키텍처는 매개변수(가중치 및 편향)를 공유하는 이중으로 구성된 동일한 컨벌루션 신경망으로 구성하고 또한 이 경로를 통해 유사도 학습과 분산을 줄이는 Keras API를 적용하였다. 이중 동종 CNN을 사용한 유사성 학습 결과는 두 클래스의 좋지 않은 결과를 포함하지 않음으로써 클래스 간 유사성과 변동성을 줄임으로서 정확도가 개선된 결과를 나타내고 있다.

• 대기
• /
• 제32권1호
• /
• pp.71-84
• /
• 2022

This paper describes verification results for the ocean analysis field produced by the Nucleus for European Modelling of the Ocean (NEMO) of the Korea Meteorological Administration (KMA) against observed Argo floats and drift buoys over the western Pacific Ocean and the equatorial Pacific during 2020~2021. This is confirmed by a comparison of the verification for the newly updated version of the HYbrid Coordinate Ocean Model/Navy Coupled Ocean Data Assimilation (HYCOM/NCODA) against same observations. NEMO shows that the vertical ocean temperature is much closer to the Argo floats than HYCOM for most seasons in terms of bias and root mean square error. On the other hand, there are overall considerable cold biases for HYCOM, which may be due to the more rapid decreasing temperature at the shallow thermocline in HYCOM. Conclusion demonstrated that the NEMO analysis for ocean temperature is more reliable than the analysis produced by the latest version of HYCOM as well as by the out-of-date HYCOM applied to the precedent study. The surface ocean current produced by NEMO also shows 14% closer to the AOML (Atlantic Oceanographic and Meteorological Laboratory) in situ drift buoys observations than HYCOM over the western Pacific Ocean. Over the equatorial Pacific, however, HYCOM shows slightly closer to AOML observation than NEMO in some seasons. Overall, this study suggests that the resulting information may be used to promote more use of NEMO analysis.

• 천문학회보
• /
• 제45권1호
• /
• pp.38.2-38.2
• /
• 2020

Horizon Run 5 (HR5) is a cosmological hydrodynamical simulation which captures the properties of the Universe on a Gpc scale while achieving a resolution of 1 kpc. This enormous dynamic range allows us to simultaneously capture the physics of the cosmic web on very large scales and account for the formation and evolution of dwarf galaxies on much smaller scales. Inside the simulation box. we zoom-in on a high-resolution cuboid region with a volume of 1049 × 114 × 114 Mpc3. The subgrid physics chosen to model galaxy formation includes radiative heating/cooling, reionization, star formation, supernova feedback, chemical evolution tracking the enrichment of oxygen and iron, the growth of supermassive black holes and feedback from active galactic nuclei (AGN) in the form of a dual jet-heating mode. For this simulation we implemented a hybrid MPI-OpenMP version of the RAMSES code, specifically targeted for modern many-core many thread parallel architectures. For the post-processing, we extended the Friends-of-Friend (FoF) algorithm and developed a new galaxy finder to analyse the large outputs of HR5. The simulation successfully reproduces many observations, such as the cosmic star formation history, connectivity of galaxy distribution and stellar mass functions. The simulation also indicates that hydrodynamical effects on small scales impact galaxy clustering up to very large scales near and beyond the baryonic acoustic oscillation (BAO) scale. Hence, caution should be taken when using that scale as a cosmic standard ruler: one needs to carefully understand the corresponding biases. The simulation is expected to be an invaluable asset for the interpretation of upcoming deep surveys of the Universe.

• Steel and Composite Structures
• /
• 제45권2호
• /
• pp.159-173
• /
• 2022

This paper proposes a hybrid machine-learning model, which is called DANN-IP, that combines a deep artificial neural network (DANN) and an interior-point (IP) algorithm in order to improve the prediction capacity on the patch loading resistance of steel plate girders. For this purpose, 394 steel plate girders that were subjected to patch loading were tested in order to construct the DANN-IP model. Firstly, several DANN models were developed in order to establish the relationship between the patch loading resistance and the web panel length, the web height, the web thickness, the flange width, the flange thickness, the applied load length, the web yield strength, and the flange yield strength of steel plate girders. Accordingly, the best DANN model was chosen based on three performance indices, which included the R^2, RMSE, and a20-index. The IP algorithm was then adopted to optimize the weights and biases of the DANN model in order to establish the hybrid DANN-IP model. The results obtained from the proposed DANN-IP model were compared with of the results from the DANN model and the existing empirical formulas. The comparison showed that the proposed DANN-IP model achieved the best accuracy with an R^2 of 0.996, an RMSE of 23.260 kN, and an a20-index of 0.891. Finally, a Graphical User Interface (GUI) tool was developed in order to effectively use the proposed DANN-IP model for practical applications.