Earticle

본 논문에서는 거리기반 비선형 매니폴드 학습인 ISOMap에 노이즈에 강건한 거리함수를 결합한 새로운 차원축소 방법을 제안한다. ISOMap은 주성분분석이나 선형판별분석과 같은 선형 부분공간 분석법과 는 달리 데이터가 가지 는 비선형 구조를 고려한 그래프 기반 거리함수를 사용함으로써 고차원 입력공간에서 데이 터가 형성하는 비선형의 저차원 매니폴드를 찾아준다. 그러나 기본적인 ISOMap은 클래스 정보를 사용하지 않는 비 교사 학습을 수행할 뿐 아니라, 새로운 데이터에 대한 매핑함수를 제공해 주지 않으므로 패턴인식에 쉽게 적용되기 어렵고 학습 데이터가 충분히 조밀하지 않은 경우에는 노이즈에 강건하지 못한 문제점을 가지고 있다. 본 논문에 서는 이러한 문제점을 해결하기 위하여 클래스 정보를 활용한 교사학습을 수행하면서, 클래스 내 데이터의 변형 정 보를 활용하여 노이즈에 강건한 거리함수를 새롭게 정의함으로써 패턴인식에 효과적인 새로운 비선형 매니폴드 학습법을 제안한다. 제안하 는 방법을 노이즈를 추가한 벤치마크 데이터에 적용하여 그 성능을 확인하였다.

In this paper, we propose a novel dimension reduction method by modifying ISOMap with a noise-tolerant distance function. The ISOMap is the well known distance-based nonlinear manifold learning method. Unlike the linear subspace analysis methods such as PCA and LDA, it can find a low dimensional manifold that can represent nonlinear structure of data set by using graph-based distance function. However, the original ISOMap is not appropriate for patten classification problems because it does unsupervised learning without classification information and is likely to be sensitive to noises when the data set is not sufficiently dense. In order to solve the problems, we propose a new supervised manifold learning method, which uses a new distance function that is robust to input noise. The performance of the proposed method is confirmed by the experiments using the benchmark data with additional noises.

Many deep learning approaches have been studied for image classification in computer vision. However, there are not enough data to generate accurate models in medical fields, and many datasets are not annotated. This study presents a new method that can use both unlabeled and labeled data. The proposed method is applied to classify cervix images into normal versus cancerous, and we demonstrate the results. First, we use a patch self-supervised learning for training the global context of the image using an unlabeled image dataset. Second, we generate a classifier model by using the transferred knowledge from self-supervised learning. We also apply attention learning to capture the local features of the image. The combined method provides better performance than state-of-the-art approaches in accuracy and sensitivity.

In this study, we present a novel approach for enhancing chest X-ray image classification (normal, Covid-19, edema, mass nodules, and pneumothorax) by combining contrastive learning and machine learning algorithms. A vast amount of unlabeled data was leveraged to learn representations so that data efficiency is improved as a means of addressing the limited availability of labeled data in X-ray images. Our approach involves training classification algorithms using the extracted features from a linear fine-tuned Momentum Contrast (MoCo) model. The MoCo architecture with a Resnet34, Resnet50, or Resnet101 backbone is trained to learn features from unlabeled data. Instead of only fine-tuning the linear classifier layer on the MoCopretrained model, we propose training nonlinear classifiers as substitutes for softmax in deep networks. The empirical results show that while the linear fine-tuned ImageNet-pretrained models achieved the highest accuracy of only 82.9% and the linear fine-tuned MoCo-pretrained models an increased highest accuracy of 84.8%, our proposed method offered a significant improvement and achieved the highest accuracy of 87.9%.

Part of speech (POS) tagging is an indispensable part of major NLP models. Its progress can be perceived on number of languages around the globe especially with respect to European languages. But considering Indian Languages, it has not got a major breakthrough due lack of supporting tools and resources. Particularly for Odia language it has not marked its dominancy yet. With a motive to make the language Odia fit into different NLP operations, this paper makes an attempt to develop a POS tagger for the said language on a HMM (Hidden Markov Model) platform. The tagger judiciously considers bigram HMM with dynamic Viterbi algorithm to give an output annotated text with maximum accuracy. The model is experimented on a corpus belonging to tourism domain accounting to a size of approximately 0.2 million tokens. With the proportion of training and testing as 3:1, the proposed model exhibits satisfactory result irrespective of limited training size.

Locating text data in images automatically has been a challenging task. In this approach, we build a three stage system for text detection purpose. This system utilizes tensor voting and Completed Local Binary Pattern (CLBP) to classify text and non-text regions. While tensor voting generates the text line information, which is very useful for localizing candidate text regions, the Nearest Neighbor classifier trained on discriminative features obtained by the CLBP-based operator is used to refine the results. The whole algorithm is implemented in MATLAB and applied to all images of ICDAR 2011 Robust Reading Competition data set. Experiments show the promising performance of this method.

Deepfake detection is increasingly critical due to the rise of manipulated media. Existing methods often require extensive datasets and struggle with interpretability issues. To address these issues, this study introduces a novel one-class approach for detecting and localizing deepfake artifacts in videos, using authentic images to generate manipulated data for training. By integrating segmentation and leveraging convolutional neural networks with visual transformers, the method predicts both the presence and location of the generated manipulations. Experiments on seven deepfake datasets and emerging diffusion-based manipulations show that our approach consistently outperforms existing methods, demonstrating superior accuracy and localization capabilities.

Background Frailty is an important condition to detect in its early stages to prevent progression to more severe stages in older adults. Age-related declines in physical performance are strongly associated with frailty. Purpose This study aims to develop a frailty classification model by comparing the performance of machine learning models based on physical performance measures in community-dwelling older adults. Study design A cross-sectional study Methods Physical performance data were collected from older adults aged ≥65 years. Frailty classification models were developed using logistic regression, support vector machine (SVM), K-nearest neighbors (KNN), decision tree, and random forest. Clinical features including short physical performance battery, single-leg stance, SARC-F, body mass index, and mini-mental state examination (MMSE) were used as input variables for model development. The performance of each model was evaluated using accuracy, sensitivity, specificity, precision, F1-score, and area under the receiver operating characteristic curve (AUC). Permutation feature importance was employed to identify key predictors of frailty. Results The KNN model demonstrated the highest classification performance, achieving an accuracy of 0.93, an F1-score of 0.95, and an AUC of 0.86, indicating its suitability for frailty assessment. The logistic regression model achieved an accuracy of 0.86, an F1-score of 0.89, and an AUC of 0.98. The random forest model showed similar results, with an accuracy of 0.86, an F1-score of 0.88, and an AUC of 0.96. The SVM model recorded an accuracy of 0.79, an F1-score of 0.84, and an AUC of 0.80. The decision tree model showed the lowest performance, with an accuracy of 0.71, an F1-score of 0.78, and an AUC of 0.64. Feature importance analysis revealed that MMSE and SARC-F were the most influential predictors in the KNN model. Conclusions This study demonstrates that KNN is well-suited for identifying subtle variations in physical function that contribute to frailty. The results highlight its potential for clinical implementation in automated frailty screening. Feature importance analysis provides insight into key predictors, supporting personalized assessment strategies. However, due to the small sample size, further research is needed to assess the generalizability of frailty classification models in larger populations.

In this study, a supervised deep learning approach is employed to construct a model that learns the relationship between input data and the corresponding output labels. The objective is to enable the model to generate accurate predictions for unseen data. Furthermore, we propose a method for semantically segmenting and tracking real-time drivable road areas to visually display them. To achieve this objective, the feasibility and performance of real-time applications using you only look once v8 (YOLOv8) architecture and mask regional-based convolutional neural network (Mask R-CNN) algorithm for learning are compared. Drivable road area segmentation involves delineating the present drivable zone across several lanes in the direction of vehicle travel, necessitating the detection of the status of the preceding vehicle. The identification rate for area detection is generally high during both day and night under clear conditions. Furthermore, the model derived from the YOLOv8 architecture demonstrated superior performance in visually extracting drivable area segmentation compared with the Mask R-CNN-based model.

Deep reinforcement learning (DRL) technology has been actively studied with the recent advances in deep learning. As a result, the researchers are continuously improving performance and expanding the applications. However, recent literature reports that the performance of DRL is sensitive to the various design choices, e.g., the neural network initialization. Accordingly, it makes DRL hard to obtain a stable performance, which degrades reproducibility. Therefore, we propose a supervised pre-training method for both policy and value networks to improve stability. We pre-train the policy network to maximize the initial entropy and pre-train the value network to bias the distribution to a specific value. The experiments are conducted on tasks with discrete action space where it is hard to control the initial entropy. Through the experiments, the effectiveness of the proposed method in terms of stability and performance is validated.

소셜미디어의 발달로 인하여 즉각적인 소통이 활발해졌지만, 혐오표현이 유발하는 차별행위가 늘어남에 따라 혐오표현을 필터링하는 연구의 필요성이 제기되고 있다. 혐오표현은 다양한 카테고리로 구분되지만, 카테고리별로 균형 잡힌 데이터셋을 구축하기에는 어려움이 존재한다. 따라서 본 연구에서는 데이터 증강을 적용하여 혐오표현 분류 성능을 향상시킨 모델을 제시한다. Easy data augmentation techniques를 적용하여 최소 규모의 카테고리 데이터를 증강하였다. Kcbert-base 모델에 focal loss와 supervised contrastive learning을 적용하여, 동일 카테고리의 문장 유사도는 높이고, 다른 카테고리와의 문장 유사도는 낮추면서 모델을 학습시켰다. 실험 결과 증강과 focal loss를 적용하지 않은 모델에 비해 easy data augmentation techniques와 focal loss, supervised contrastive learning을 적용한 모델의 평균 정확도는 1.4%, macro f1-score는 4.4% 우수한 것을 확인하였다.

Anomaly recognition in visual and audio data has gained increasing significance in computer vision, as it plays a crucial role in protecting human lives and property. In this work, we developed a semi-supervised multimodal framework for anomaly recognition that combines audio and visual data for better performance. The proposed framework employs a hybrid network consisting of a convolutional neural network, Bi-Directional Long Short-Term Memory, a multi-head attention module, and a fully connected layer for anomalous pattern recognition. We created a novel real-time visual-audio anomaly recognition dataset and evaluated our framework on it, achieving promising results.

Although living organisms differ in shape and size, all are fundamentally structured by genetic sequences. Interpreting these sequences helps explain how organisms function. With the advancement of AI, significant breakthroughs have been made in protein sequencing and understanding protein function. However, there is still room for improvement, as data-intensive models require a substantial amount of protein sequences, many of which are not publicly available or lack quality. In this paper, we present a semi-supervised learning scheme to address the shortage of high-quality training data necessary for training protein language models. We demonstrate that this approach enhances the model's capability to classify toxic fungi protein sequences.

Researchers and healthcare professionals have paid attention to artificial intelligence (AI) in the healthcare industry. The application of effective and efficient AI techniques can help to increase the precision of medical decision. AI offers great advances for countries currently struggling with complex healthcare systems and a physician shortage. In previous research in the field of health management, big data analytics in healthcare has been a significant subject from multidisciplinary aspects. To detect the onset of disease in the large-scale data, examining image and video sources and social media data is required. The literature represents various AI applications for healthcare services as well as an unexplored area of medical research emphasizes medical decision-making, patient diagnostics data, and network of health services. This study presents an innovative datadriven approach with knowledge-based analysis utilizing AI techniques. The proposed method develops a comprehensive approach including machine learning methods and its application to healthcare big data analysis. The significance of identifying and covering the key AI applications for healthcare is suggested. The result highlights that organization may greatly benefit from the use of this technology in healthcare operations with AI-based solutions to provide different treatment options and personalized therapies. With the applications of AI techniques, the general effectiveness of hospitals and healthcare systems may be increased, and overall healthcare costs may be reduced.

Recently, due to the recent significant advances in machine learning and deep learning, it is being utilized in many fields. However, real-world data in the medical field significantly degrades the performance of machine learning algorithms due to problems that are heavily skewed to specific states or that the distribution of data is unbalanced. Therefore, this study solves the problem of not being learned by converting the dependent variable into a regression problem that predicts using a new dependent variable by pseudo labeling. Also, this study present ensemble methods to improve the performance of the model and prevent overfitting.

정보통신기술(ICT) 고도화에 따라 PDF, MS Office, HWP 파일로 대표되는 전자 문서형 파일의 활용이 많아졌고, 공격자들은 이 상황을 놓치지 않고 문서형 악성코드를 이메일과 메신저를 통해 전달하여 감염시키는 피해사례가 많아졌다. 이러한 피해를 막고자 AI를 사용한 악성코드 탐지 연구가 진행되고 있으나, PDF나 MS-Office와 같이 전 세계적으로 활용성이 높은 전자 문서형 파일에 비해 주로 국내에서만 활용되는 HWP(한글 워드 프로세서) 문서 파일은 양질의 정상 또는 악성 데이터가 부족하여 지속되는 공격에 강건한 모델 생성에 한계점이 존재한다. 이러한 한계점을 해결하기 위해 기존 수집된 데이터를 변형하여 학습 데이터 규모를 늘리는 데이터 증강 방식이 제안 되었으나, 증강된 데이터의 유용성을 평가하지 않아 불확실한 데이터를 모델 학습에 활용할 가능성이 있다. 본 논문에서는 HWP 악성코드 탐지에 있어 데이터의 유용성을 정량화하고 이에 기반하여 학습에 유용한 증강 데이터만을 활용하여 기존보다 우수한 성능의 AI 모델을 학습하는 준지도학습 기법을 제안한다.

With the advancement of information and communication technology (ICT), the use of electronic document types such as PDF, MS Office, and HWP files has increased. Such trend has led the cyber attackers increasingly try to spread malicious documents through e-mails and messengers. To counter such attacks, AI-based methodologies have been actively employed in order to detect malicious document files. The main challenge in detecting malicious HWP(Hangul Word Processor) files is the lack of quality dataset due to its usage is limited in Korea, compared to PDF and MS-Office files that are highly being utilized worldwide. To address this limitation, data augmentation have been proposed to diversify training data by transforming existing dataset, but as the usefulness of the augmented data is not evaluated, augmented data could end up harming model's performance. In this paper, we propose an effective semi-supervised learning technique in detecting malicious HWP document files, which improves overall AI model performance via quantifying the utility of augmented data and filtering out useless training data.

본 연구는 기온, 강수량, 풍향, 풍속, 습도, 운량, 일조, 일사 등 시간별 기상 데이터를 이용하여, 일사 및 일조 그리고 태양광 발전예측을 하였다. 지도학습에서 입출력패턴은 예측에서 가장 중요한 요소이지만 인간이 직접 결정해야하기 때문에, 반복적인 실험에 의해 결정해야 한다. 본 연구는 일사 및 일조 예측을 위하여 4가지 모델의 입출력 패턴을 제안하였다. 또한, 예측된 일조 및 일사 데이터와 전라남도 영암 태양광 발전소의 발전량 데이터를 사용하여 태양광 발전량을 예측하였다. 실험결과 일조 및 일사 예측에서 모델 4가 가장 예측결과가 우수했으며, 모델 1에 비해 일조의 RMSE는 1.5배 정도 그리고 일사의 RMSE는 3배 정도 오차가 줄었다. 태양광 발전예측 실험결과 일조 및 일사와 마찬가지로 모델 4가 가장 예측결과가 좋았으며, 모델 1 보다 RMSE가 2.7배 정도 오차가 줄었다.

This study predicts solar radiation, solar radiation, and solar power generation using hourly weather data such as temperature, precipitation, wind direction, wind speed, humidity, cloudiness, sunshine and solar radiation. I/O pattern in supervised learning is the most important factor in prediction, but it must be determined by repeated experiments because humans have to decide. This study proposed four input and output patterns for solar and sunrise prediction. In addition, we predicted solar power generation using the predicted solar and solar radiation data and power generation data of Youngam solar power plant in Jeollanamdo. As a experiment result, the model 4 showed the best prediction results in the sunshine and solar radiation prediction, and the RMSE of sunshine was 1.5 times and the sunshine RMSE was 3 times less than that of model 1. As a experiment result of solar power generation prediction, the best prediction result was obtained for model 4 as well as sunshine and solar radiation, and the RMSE was reduced by 2.7 times less than that of model 1.

3D open-field farm model developed from UAV (Unmanned Aerial Vehicle) data could make crop monitoring easier, also could be an important dataset for various fields like remote sensing or precision agriculture. It is essential to separate crops from the non-crop area because labeling in a manual way is extremely laborious and not appropriate for continuous monitoring. We, therefore, made a 3D open-field farm model based on UAV images and developed a crop segmentation model using a supervised machine learning algorithm. We compared performances from various models using different data features like color or geographic coordinates, and two supervised learning algorithms which are SVM (Support Vector Machine) and KNN (K-Nearest Neighbors). The best approach was trained with 2-dimensional data, ExGR (Excess of Green minus Excess of Red) and z coordinate value, using KNN algorithm, whose accuracy, precision, recall, F1 score was 97.85, 96.51, 88.54, 92.35% respectively. Also, we compared our model performance with similar previous work. Our approach showed slightly better accuracy, and it detected the actual crop better than the previous approach, while it also classified actual non-crop points (e.g. weeds) as crops.

다가오는 지능정보사회의 핵심이 될 인공지능(AI) 교육의 필요성이 대두되면서, 국가적 차원에서도 교육과정 에 인공지능 관련 내용을 포함하는 등 관심을 집중시키고 있다. 본 연구에서는 지도학습 중심의 머신러닝을 통 해 생활 속 문제를 해결하는 과정에서 학생들의 창의적 문제해결력을 신장시키기 위해 PASPA 교육 프로그램을 제시하였으며, 학습의 효과를 높이기 위해 피지컬 컴퓨팅 도구인 마이크로비트(Micro:bit)를 활용하였다. PASPA 교육 프로그램에 적용된 교수 학습 과정은 문제 인식(Problem Recoginition), 해결 방법 논의(Argument), 데이터 기준 세우기(Setting data standard), 프로그래밍(Programming), 적용 및 평가(Application and evaluation)의 5단 계로 이루어진다. 본 교육 프로그램을 학생들에게 적용한 결과 창의적 문제해결력의 향상을 확인할 수 있었으며, 세부 영역에서는 특정 영역의 지식·사고, 비판적·논리적 사고 영역에서 유의한 차이를 보임이 확인되었다.

As the need for artificial intelligence (AI) education, which will become the core of the upcoming intelligent information society rises, the national level is also focusing attention by including artificial intelligence-related content in the curriculum. In this study, the PASPA education program was presented to enhance students' creative problem-solving ability in the process of solving problems in daily life through supervised machine learning. And Micro:bit, a physical computing tool, was used to enhance the learning effect. The teaching and learning process applied to the PASPA education program consists of five steps: Problem Recoginition, Argument, Setting data standard, Programming, Application and evaluation. As a result of applying this educational program to students, it was confirmed that the creative problem-solving ability improved, and it was confirmed that there was a significant difference in knowledge and thinking in specific areas and critical and logical thinking in detailed areas.

기존의 e스포츠 하이라이트 생성은 주로 사전에 정의된 인게임 이벤트에 기반한 규칙 기반 시스템에 의존해 왔으나, 경기의 세밀한 전개 흐름을 충분히 반영하지 못한다. 본 논문에서는 전문 편집자의 의사결정을 학습하는 지도 학습기 반 하이라이트 탐지 프레임워크를 제안한다. 본 연구는 방송 화면에서 직접 획득 가능한 인게임 수치를 기반으로 시 계열 데이터를 정제하여 학습에 활용함으로써, 특정 중계 환경에 종속되지 않고 일반 사용자 영상에도 동일한 방식으 로 적용할 수 있는 확장성을 확보하였다. 광학 문자 인식을 통해 추출한 인게임 시간 시계열 데이터를 정제하고, 이 상치 보정과 보간을 수행하여 신뢰도 높은 프레임 단위 라벨을 구축하였다. 제안된 방법은 정제된 게임플레이 정보를 활용해 팀 승률과 플레이어 기여도를 예측하고, 이를 통합하여 하이라이트 후보 프레임을 식별한다. 실험 결과, 전문 편집 장면의 특성을 효과적으로 학습하여 높은 정밀도와 재현율을 달성하였으며, 제안된 접근법은 e스포츠 하이라이 트 자동화의 품질과 실용성을 향상시킬 수 있음을 확인하였다.

Conventional esports highlight generation has relied mainly on rule-based systems driven by predefined in-game events, which limits their ability to capture the fine-grained progression of gameplay. This study proposes a supervised learning–based highlight detection framework that models the decision-making patterns of professional editors. The framework leverages in-game numerical indicators directly obtainable from broadcast video and refines them into reliable time-series inputs, thereby ensuring applicability not only to professional broadcast footage but also to gameplay videos produced by general users without dependence on specific production environments. Using optical character recognition, we extract and preprocess frame-aligned in-game time-series data, applying outlier correction and interpolation to construct high-fidelity frame-level labels. The refined game-derived information is then used to estimate team win probability and player contribution, which are integrated to identify candidate highlight frames. Experimental results demonstrate that the proposed method effectively learns the characteristics of expert-edited highlights, achieving high precision and recall. These findings confirm that the framework improves the quality and practicality of automated esports highlight generation.