• Proceedings of the Korean Society of Computer Information Conference
• /
• 2022.07a
• /
• pp.29-30
• /
• 2022

Curriculum learning은 딥러닝의 성능을 향상시키기 위해 사람의 학습 과정과 유사하게 일종의 'curriculum'을 도입해 모델을 학습시키는 방법이다. 대부분의 연구는 학습 데이터 중 개별 샘플의 난이도를 기반으로 점진적으로 모델을 학습시키는 방안에 중점을 두고 있다. 그러나, coarse-to-fine 메카니즘은 데이터의 난이도보다 학습에 사용되는 class의 유사도가 더욱 중요하다고 주장하며, 여러 난이도의 auxiliary task를 차례로 학습하는 방법을 제안했다. 그러나, 이 방법은 혼동행렬 기반으로 class의 유사성을 판단해 auxiliary task를 생성함으로 다중 레이블 분류에는 적용하기 어렵다는 한계점이 있다. 따라서, 본 논문에서는 multi-label 환경에서 multi-class와 binary task를 생성하는 방법을 제안해 coarse-to-fine 메카니즘 적용을 위한 방안을 제시하고, 그 결과를 분석한다.

• Journal of the Korean Society of Marine Environment & Safety
• /
• v.26 no.2
• /
• pp.139-148
• /
• 2020

The most important factor affecting the berthing energy generated when a ship berths is the berthing velocity. Thus, an accident may occur if the berthing velocity is extremely high. Several ship features influence the determination of the berthing velocity. However, previous studies have mostly focused on the size of the vessel. Therefore, the aim of this study is to analyze various features that influence berthing velocity and determine their respective importance. The data used in the analysis was based on the berthing velocity of a ship on a jetty in Korea. Using the collected data, machine learning classification algorithms were compared and analyzed, such as decision tree, random forest, logistic regression, and perceptron. As an algorithm evaluation method, indexes according to the confusion matrix were used. Consequently, perceptron demonstrated the best performance, and the feature importance was in the following order: DWT, jetty number, and state. Hence, when berthing a ship, the berthing velocity should be determined in consideration of various features, such as the size of the ship, position of the jetty, and loading condition of the cargo.

• Korean Journal of Cognitive Science
• /
• v.19 no.2
• /
• pp.163-175
• /
• 2008

Twelve Japanese students living in South Korea have been examined for their perceptual identification of an initial consonant in Korean syllables with or without a white noise. A confusion matrix was then subject to analyses of additive clustering, individual difference scaling, and probability of information transmission, the results of which were also compared to those of South Koreans. The Japanese in the present experiment confused /다/and/타/ most frequently, followed by /가/ and /카/, /자, 차, 짜/, /타/ and /따/, and so on. The results of additive clustering analysis of the Japanese significantly differed from those of the South Koreans. Individual difference scaling revealed dimensions of sonorant, aspiration and coronal. While South Koreans showed binary values on aspiration and tenseness dimensions, the Japanese did continuous values on such dimensions. An information transmission probability analysis revealed that the Japanese participants could not perceive very well such larynx features as tenseness and aspiration compared to the South Korean participants. The former group, however, perceived very well place of articulation features such as labial and coronal. The present results suggest that an approach dealing with structures of base representations is important in understanding the phonological categories of languages.

• The Korean Journal of Applied Statistics
• /
• v.27 no.5
• /
• pp.773-786
• /
• 2014

We propose the multiplication of false rates (MFR) which is a classification accuracy criteria and an area type of rectangle from ROC curve. Optimal threshold obtained using MFR is compared with other criteria in terms of classification performance. Their optimal thresholds for various distribution functions are also found; consequently, some properties and advantages of MFR are discussed by comparing FNR and FPR corresponding to optimal thresholds. Based on general cost function, cost ratios of optimal thresholds are computed using various classification criteria. The cost ratios for cost curves are observed so that the advantages of MFR are explored. Furthermore, the de nition of MFR is extended to multi-dimensional ROC analysis and the relations of classification criteria are also discussed.

• The Korean Journal of Applied Statistics
• /
• v.32 no.2
• /
• pp.187-198
• /
• 2019

Extensive literature exists on how to estimate optimal thresholds based on various accuracy measures using receiver operating characteristic (ROC) and cumulative accuracy profile (CAP) curves. This paper now proposes an alternative measure to represented the specific partial area under the ROC and CAP curves. The relationship between ROC and CAP functions is examined using differential equations of the new defined partial area under curves. In addition, the relationship with the optimal thresholds under conditions of various accuracy measures for the ROC and CAP functions is also derived. We assume there are two kinds of distribution functions composing the mixed distribution as various normal distributions before finding the optimal thresholds. Corresponding type 1 and 2 errors are also explored and discussed under various conditions for accuracy measures.

• The Korean Journal of Applied Statistics
• /
• v.34 no.5
• /
• pp.807-822
• /
• 2021

Various accuracy measures that can be explained on the odds curve are discussed, and an alternative accuracy measure, the maximum square, is proposed based on the characteristics of the odds curve. Thresholds corresponding to these accuracy measures are obtained by considering various probability distribution functions and an illustrative example. Their characteristics are discussed while comparing many kinds of statistics measuring thresholds. Therefore, we can conclude that optimal thresholds could be explored from the odds curve, similar to the ROC curve, and that the maximum square measure can be used as a good accuracy measure that can improve the performance of the binary classification model.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2023.05a
• /
• pp.359-359
• /
• 2023

최근 기후변화로 인한 홍수 및 가뭄과 같은 자연재해가 증가함에 따라 이를 선제적으로 탐지 및 예방할 수 있는 해결책에 대한 필요성이 증가하고 있다. 이러한 수재해를 예방하기 위해서 하천, 저수지 등 가용수자원의 지속적인 모니터링은 필수적이다. SAR 위성 영상의 경우 주야간 및 기상상황에 상관없이 지속적인 수체 탐지가 가능하다. 일반적으로 SAR 기반 수체 탐지 시 송수신 방향이 동일한 편파(co-polarized) 영상을 사용한다. 하지만 co-polarized 영상의 경우 바람 및 강우에 민감하게 반응하여 수체 미탐지의 가능성이 존재한다. 한편 송수신 방향이 서로 다른 편파(cross-polarized) 영상은 강우 및 바람의 영향에 민감하지 않지만 식생에 민감하게 반응하여 수체의 오탐지율이 높다는 단점이 존재한다. 이에 SAR 영상의 편파 특성에 따라 수체 탐지의 정확도 차이가 발생하여 최적의 편파 영상 조합을 구성하는 것이 중요하다. 본 연구에서는 Sentinel-1 SAR 위성의 VV, VH, VV+VH 편파 영상과 머신러닝 알고리즘 중 하나인 SVM (support vector machine)을 활용하여 수체탐지를 수행하였다. 편파 영상 조합별 수체 탐지 결과의 검증을 위하여 혼동행렬 (confusion matrix) 기반 평가지수를 사용하였다. 각각의 수체탐지 결과의 비교 및 분석을 통하여 SAR 기반 수체 탐지를 위한 최적의 밴드 조합을 도출하였다. 본 연구결과를 바탕으로 차후 높은 시공간 해상도를 가진 SAR 영상의 활용이 가능하다면 수재해 및 수자원 관리의 효율성을 높일 수 있을 것으로 기대된다.

• Journal of the Korea Society of Computer and Information
• /
• v.28 no.8
• /
• pp.59-66
• /
• 2023

The Golden Cross is commonly seen as a buy signal in financial markets, but its reliability for predicting stock price movements is limited due to market volatility. This paper introduces a time-invariant approach that considers the Golden Cross as a singular event. Utilizing LSTM neural networks, we forecast significant stock price changes following a Golden Cross occurrence. By comparing our approach with traditional time series analysis and using a confusion matrix for classification, we demonstrate its effectiveness in predicting post-event stock price trends. To conclude, this study proposes a model with a precision of 83%. By utilizing the model, investors can alleviate potential losses, rather than making buy decisions under all circumstances following a Golden Cross event.

• Journal of the Korean Data and Information Science Society
• /
• v.22 no.2
• /
• pp.149-157
• /
• 2011

Undecided inference could be regarded as a missing data problem such as MARand MNAR. Under the assumption of MAR, undecided inference make use of logistic regression model. The probability of default for the undecided group is obtained with regression coefficient vectors for the decided group and compare with the probability of default for the decided group. And under the assumption of MNAR, undecide dinference make use of logistic regression model with additional feature random vector. Simulation results based on two kinds of real data are obtained and compared. It is found that the misclassification rates are not much different from the rate of rawdata under the assumption of MAR. However the misclassification rates under the assumption of MNAR are less than those under the assumption of MAR, and as the ratio of the undecided group is increasing, the misclassification rates is decreasing.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.12 no.1
• /
• pp.75-81
• /
• 2012

Road identification is the important task for extracting a road region from the high-resolution satellite images, when the road candidates is extracted by the pre-processing tasks using a binarization, noise removal, and color processing. Therefore, we propose a noble approach for identifying a road using the orientation-selective spatial filters, which is motivated by a computational model of neuron cells found in the primary visual cortex. In our approach, after the neuron cell typed spatial filters is applied to the isolated connected-labeling road candidate regions, proposed method identifies the region of perceiving the strong orientation feature with the real road region. To evaluate the effectiveness of the proposed method, the accuracy&error ratio in the confusion matrix was measured from road candidates including road and non-road class. As a result, the proposed method shows the more than 92% accuracy.