제조 공정 결함 탐지를 위한 MixMatch 기반 준지도학습 성능 분석

김예준,정예은,김용수 한국산업경영시스템학회 2023 한국산업경영시스템학회지 Vol.46 No.4

Recently, there has been an increasing attempt to replace defect detection inspections in the manufacturing industry using deep learning techniques. However, obtaining substantial high-quality labeled data to enhance the performance of deep learning models entails economic and temporal constraints. As a solution for this problem, semi-supervised learning, using a limited amount of labeled data, has been gaining traction. This study assesses the effectiveness of semi-supervised learning in the defect detection process of manufacturing using the MixMatch algorithm. The MixMatch algorithm incorporates three dominant paradigms in the semi-supervised field: Consistency regularization, Entropy minimization, and Generic regularization. The performance of semi-supervised learning based on the MixMatch algorithm was compared with that of supervised learning using defect image data from the metal casting process. For the experiments, the ratio of labeled data was adjusted to 5%, 10%, 25%, and 50% of the total data. At a labeled data ratio of 5%, semi-supervised learning achieved a classification accuracy of 90.19%, outperforming supervised learning by approximately 22%p. At a 10% ratio, it surpassed supervised learning by around 8%p, achieving a 92.89% accuracy. These results demonstrate that semi-supervised learning can achieve significant outcomes even with a very limited amount of labeled data, suggesting its invaluable application in real-world research and industrial settings where labeled data is limited.

KNN-AdaBoost 모델과 AdaBoost-SVM 모델을 통한 준교사학습 비교

유회중(Hoe Joong You),권영만(Young Man Kwon) 한국IT마케팅학회 2014 한국IT마케팅학회 학술대회 Vol.2014 No.1

The supervised-learning and unsupervised-learning when analyzed in the case of the one without as there are output information is combined data, the error rate is not easily trust that a comparison learning history. Therefore I suggest the semi-supervised- learning that gave me that were mixed compared to history unsupervised-learning and supervised-learning. In addition, the semi-supervised learning that gave are mutually comparing model that combines the model and AdaBoost and SVM bound the KNN algorithm and AdaBoost algorithm, to check the semi-supervised learning of the error rate is low binding model.

머신러닝 기술을 이용한 사이버위협 대응 방안에 관한 연구

이광형(Kwang Hyoung Lee),정용훈(Young Hoon Jung) 한국산학기술학회 2023 한국산학기술학회논문지 Vol.24 No.10

최근 IoT, ICT 환경에서 정교하게 진화하고 있는 사이버위협은 증가하고 있으며, 이에 보안 운영 환경이 감당하기 힘든 수준으로 복잡해지고, 이로 인해 분석할 데이터는 증가하고 있다. 또한 보안 전문 인력 부족 및 성숙도 부족에 따른 휴먼에러가 증가하고 있다. 본 논문에서는 머신러닝 비지도학습과 준지도학습 모델을 이용하여 정교하게 진화하는 사이버위협에 대응할 수 있도록 하였다. 본 논문에서 사용된 비지도학습 모델은 세션정보(L4), 프로토콜 헤더정보(L7), 파일 등의 정보를 수집하고, 이를 위협정보와 비교하여 유사한 위협을 매핑 및 라벨링하여 이상행위를 탐지하는데 사용하였다. 클러스터링 기술을 통해 모델링을 수행하고, 생성된 모델은 재학습을 통해 모델 업데이트하여 분석속도가 향상될 수 있으며, 생성된 모델은 준지도학습에 재학습하여 모델을 업데이트할 수 있도록 하였다. 준지도학습 모델은 비지도 학습 모델의 시간별 클러스터링 기반 탐지기법은 평소와 다른 행위를 탐지하는데 유용하고, 공격유형별 지도학습 기반탐지 기법은 네트워크 기반 행위가 특정 공격에 해당하는지 구별하여 탐지하는데 유용하다. 준지도학습은 지도학습 모델과 비지도학습 모델을 적절히 혼합하여 탐지 정확도와 노이즈(탐지)를 줄일 수 있는 장점이 있다. Cyber threats are evolving more sophisticatedly in the IoT and ICT environments. As a result, the security operating environment is becoming unmanageably complex, and the data to be analyzed is increasing. In addition, human errors are increasing due to a lack of security professionals and maturity. This study used unsupervised and semi-supervised machine learning models to respond to sophisticated cyber threats. The unsupervised learning model used in this paper collects information, such as session information (L4), protocol header information (L7), and files, and compares this with threat information to map and label similar threats and detect abnormal behavior. Modeling is performed through clustering technology, and the analysis speed can be improved by updating the generated model through re-learning. The generated model can be re-trained through semi-supervised learning to update the model. The temporal clustering-based detection technique of the unsupervised learning model in the semi-supervised learning model is useful for detecting behavior that is different from usual, and the supervised learning-based detection technique for each attack type is useful for distinguishing and detecting whether network-based behavior corresponds to a specific attack. Semi-supervised learning can reduce the detection accuracy and noise (detection) by appropriately mixing supervised learning models and unsupervised learning models.

FedGC: 준지도 연합학습을 위한 글로벌 일관성 정규화

정구본,최동완 한국정보과학회 2022 정보과학회논문지 Vol.49 No.12

Recently, in the field of artificial intelligence, methods of learning neural network models in distributed environments that use sufficient data and hardware have been actively studied. Among them, federated learning, which guarantees privacy preservation without sharing data, has been a dominant scheme. However, existing federated learning methods assume supervised learning using only labeled data. Since labeling costs are incurred for supervised learning, the assumption that only label data exists in the clients is unrealistic. Therefore, this study proposes a federated semi-supervised learning method using both labeled data and unlabeled data, considering a more realistic situation where only labeled data exists on the server and unlabeled data on the client. We designed a loss function considering consistency regularization between the output distributions of the server and client models and analyzed how to adjust the influence of consistency regularization. The proposed method improved the performance of existing semi-supervised learning methods in federated learning settings, and through additional experiments, we analyzed the influence of the loss term and verified the validity of the proposed method. 최근 인공지능 분야에서는 충분한 데이터와 하드웨어를 이용하기 위해 분산 환경에서의 신경망 모델 학습 방법이 활발히 연구되어지고 있다. 그중 데이터 공유 없이 프라이버시를 보장하는 연합학습이 대두되고 있지만, 기존 연합학습 방법들은 레이블 데이터만 이용하는 지도학습을 가정한다. 지도학습을 위해서는 레이블 비용이 발생한다는 점에서 클라이언트에 레이블 데이터만 존재하는 가정은 비현실적이다. 따라서 본 논문은 서버에 레이블 데이터가 있고 클라이언트에 레이블이 없는 데이터만 존재하는 현실적인 상황을 가정하여, 레이블 데이터와 레이블이 없는 데이터 모두를 사용한 준지도 연합학습 방법을 제안한다. 논문에서는 서버와 클라이언트 모델의 일관성 정규화를 고려한 손실함수를 설계하며, 일관성 정규화의 영향력을 조절하는 방안에 대해 분석한다. 제안된 방법은 연합학습 환경에서 기존 준지도 학습 방법의 성능을 개선하였으며, 추가적인 실험을 통해 손실항의 영향력을 분석하고 제안된 방법의 타당성을 검증한다.

준지도학습 기반 반도체 공정 이상 상태 감지 및 분류

이용호,최정은,홍상진,Lee, Yong Ho,Choi, Jeong Eun,Hong, Sang Jeen 한국반도체디스플레이기술학회 2020 반도체디스플레이기술학회지 Vol.19 No.4

With miniaturization of semiconductor, the manufacturing process become more complex, and undetected small changes in the state of the equipment have unexpectedly changed the process results. Fault detection classification (FDC) system that conducts more active data analysis is feasible to achieve more precise manufacturing process control with advanced machine learning method. However, applying machine learning, especially in supervised learning criteria, requires an arduous data labeling process for the construction of machine learning data. In this paper, we propose a semi-supervised learning to minimize the data labeling work for the data preprocessing. We employed equipment status variable identification (SVID) data and optical emission spectroscopy data (OES) in silicon etch with SF6/O2/Ar gas mixture, and the result shows as high as 95.2% of labeling accuracy with the suggested semi-supervised learning algorithm.

지도학습과 강화학습을 이용한 준능동 중간층면진시스템의 최적설계

강주원,김현수,Kang, Joo-Won,Kim, Hyun-Su 한국공간구조학회 2021 한국공간구조학회지 Vol.21 No.4

A mid-story isolation system was proposed for seismic response reduction of high-rise buildings and presented good control performance. Control performance of a mid-story isolation system was enhanced by introducing semi-active control devices into isolation systems. Seismic response reduction capacity of a semi-active mid-story isolation system mainly depends on effect of control algorithm. AI(Artificial Intelligence)-based control algorithm was developed for control of a semi-active mid-story isolation system in this study. For this research, an practical structure of Shiodome Sumitomo building in Japan which has a mid-story isolation system was used as an example structure. An MR (magnetorheological) damper was used to make a semi-active mid-story isolation system in example model. In numerical simulation, seismic response prediction model was generated by one of supervised learning model, i.e. an RNN (Recurrent Neural Network). Deep Q-network (DQN) out of reinforcement learning algorithms was employed to develop control algorithm The numerical simulation results presented that the DQN algorithm can effectively control a semi-active mid-story isolation system resulting in successful reduction of seismic responses.

Semi-supervised Multi-view Manifold Discriminant Intact Space Learning

( Lu Han ),( Fei Wu ),( Xiao-yuan Jing ) 한국인터넷정보학회 2018 KSII Transactions on Internet and Information Syst Vol.12 No.9

Semi-supervised multi-view latent space learning is gaining considerable popularity recently in many machine learning applications due to the high cost and difficulty to obtain the large amount of label information of data. Although some semi-supervised multi-view latent space learning methods have been presented, there is still much space for improvement: 1) How to learn latent discriminant intact feature representations by employing data of multiple views; 2) How to exploit the manifold structure of both labeled and unlabeled point in the learned latent intact space effectively. To address the above issues, we propose an approach called semi-supervised multi-view manifold discriminant intact space learning (SM2DIS) for image classification in this paper. SM2DIS aims to seek a manifold discriminant intact space for data of different views by making use of both the discriminant information of labeled data and the manifold structure of both labeled and unlabeled data. Experimental results on MNIST, COIL-20, Multi-PIE, and Caltech-101 databases demonstrate the effectiveness and robustness of our proposed approach.

웹 검색을 위한 확장 가능 준지도 선호도 학습

김계현(Kye-Hyeon Kim),최승진(Seungjin Choi) 한국정보과학회 2011 정보과학회 컴퓨팅의 실제 논문지 Vol.17 No.4

본 논문은 웹 검색에서 사용자의 검색 기록과 웹 문서간의 연관 관계를 동시에 이용하여 적합한 랭킹 함수를 학습하는 방법을 소개한다. 제안하는 방법은 그래프 기반의 준지도 학습(semi-supervised learning) 기법을 선호도 학습(preference learning)에 적용한 기계학습 알고리즘으로, 그래프의 가중치 행렬(weight matrix)을 직접적으로 계산할 필요가 없는 matrix-free 알고리즘을 고안하여 대규모 데이터를 다룰 수 있도록 하였다. 또한 새로운 검색 기록들이 추가될 때마다 이미 학습된 랭킹 함수를 효율적으로 업데이트할 수 있도록 점진적(incremental) 학습 알고리즘을 개발하였다. Microsoft Research Asia에서 약 400만개 질의어에 대해 수집한 MSN Live Search의 검색 기록 데이터에 본 방법을 적용한 결과, 주어진 질의어에 적합함에도 Live Search에서 순위가 낮게 책정되었던 웹 페이지들의 검색 순위를 크게 향상시킴으로써(평균 11-20위 → 3-12위로 향상) 더욱 정확한 검색 결과를 산출하였으며, 이를 위해 질의어당 실시간으로 소요된 처리 시간은 불과 1.4밀리초였다. In this paper, we present a novel method for learning to rank, which is capable of semi-supervised learning by utilizing both click-through logs and the similarities between web pages simultaneously. To achieve web-scale semi-supervised learning, we develop a matrix-free algorithm that extracts latent features from a given set of web pages, where the huge similarity matrix of the web pages is not needed. Moreover, we present an incremental algorithm for our semi-supervised preference learning framework. Experiments on the Microsoft Live Search query log data show that our method effectively improves the ranks of relevant web pages of a given query, which are underestimated by Microsoft Live Search.

Simultaneous Kernel Learning and Label Imputation for Pattern Classification with Partially Labeled Data

Minyoung Kim 한국지능시스템학회 2017 INTERNATIONAL JOURNAL of FUZZY LOGIC and INTELLIGE Vol.17 No.1

The kernel function plays a central role in modern pattern classification for its ability to capture the inherent affinity structure of the underlying data manifold. While the kernel function can be chosen by human experts with domain knowledge, it is often more principled and promising to learn it directly from data. This idea of kernel learning has been studied considerably in machine learning and pattern recognition. However, most kernel learning algorithms assume fully supervised setups requiring expensive class label annotation for the training data. In this paper we consider kernel learning in the semi-supervised setup where only a fraction of data points need to be labeled. We propose two approaches: the first extends the idea of label propagation along the data similarity graph, in which we simultaneously learn the kernel and impute the labels of the unlabeled data. The second aims to minimize the dual loss in the support vector machines (SVM) classifier learning with respect to the kernel parameters and the missing labels. We provide reasonable and effective approximate solution methods for these optimization problems. These approaches exploit both labeled and unlabeled data in kernel leaning, where we empirically demonstrate the effectiveness on several benchmark datasets with partially labeled learning setups.

BeSCL: 그래프 노드 분류를 위한 개선된준지도 및 대조학습

김호승,김효준,최준수,이지형 한국지능시스템학회 2023 한국지능시스템학회논문지 Vol.33 No.4

There has been much research on semi-supervised and unsupervised learningmethods to address the problem of graph node classification. In particular, there is alot of interest in addressing situations where label information is scarce. To addressthis, we propose a new algorithm called BeSCL (Better Semi-supervised andContrastive Learning) that combines label-based data augmentation and contrastivelearning. BeSCL is designed for semi-supervised learning environments and appliesdata augmentation to the graph to create new graph data, which is then trainedusing contrastive learning in an unsupervised learning environment. Experimentalresults show that BeSCL exhibits robust performance regardless of the amount ofdata and outperforms existing node classification methods 그래프 노드 분류 문제를 해결하기 위해 많은 준지도 및 비지도 학습 방법들이 연구되고 있다. 특히 레이블 정보가 부족한 상황을 많이 해결하고자 하는데, 우리는 이를 해결하고자 레이블 정보를 이용한 데이터 증강과 대조 학습을 결합한 BeSCL(Better Semi-supervised andContrastive Learning) 이라는 새로운 알고리즘을 제안한다. BeSCL은 준지도 학습 환경에서 진행되며, 그래프에 데이터 증강을 적용하여, 새로운 그래프 데이터를 만들고 이를 비지도 학습 환경에서 이용되는 대조 학습을 이용하여 학습한다. 실험결과, BeSCL은 데이터 양에 관계 없이 강건한 성능을 보이며, 기존 노드 분류 방법론 대비 더 뛰어난 성능을 보인다.