• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.11a
• /
• pp.675-677
• /
• 2023

본 연구는 병리학 이미지 분석에서 자주 발생하는 문제 중 하나인 레이블링 불일치 문제를 해결하고자 준지도학습(semi-supervised learning) 기법을 적용하였다. 기존의 병리 진단 과정은 정확한 판정 및 치료를 위해 전문가의 판단을 필요로 한다. 이로 인해, 시간이 매우 많이 소모되며 전문가의 피로도가 증가한다. 최근 이를 해결하고자 지도학습(supervised learning) 기법을 사용하여 업무의 피로도를 감소시키고자 하는 연구가 진행되고 있다. 하지만 병리 이미지 데이터에 대한 접근이 어렵고, 병변의 위치를 레이블링 하는 부분에서 많은 비용이 발생한다. 또한 암 병변의 스펙트럼적 특성으로 인해 레이블링 과정 속에서 레이블링 불일치 문제가 발생할 가능성이 높다. 이러한 문제를 극복하기 위해, 우리는 제한된 레이블 된 데이터와 많은 양의 레이블 되지 않은 데이터를 활용하는 준지도학습 방법론을 제안한다. 이 제안하는 방법은 필요한 수동 레이블링 작업량을 줄여, 병리학자들에게 보다 효과적인 진단 도구를 제공할 것으로 예상된다.

• Journal of Korean Tunnelling and Underground Space Association
• /
• v.21 no.3
• /
• pp.419-432
• /
• 2019

Most of deep learning model training was proceeded by supervised learning, which is to train labeling data composed by inputs and corresponding outputs. Labeling data was directly generated manually, so labeling accuracy of data is relatively high. However, it requires heavy efforts in securing data because of cost and time. Additionally, the main goal of supervised learning is to improve detection performance for 'True Positive' data but not to reduce occurrence of 'False Positive' data. In this paper, the occurrence of unpredictable 'False Positive' appears by trained modes with labeling data and 'True Positive' data in monitoring of deep learning-based CCTV accident detection system, which is under operation at a tunnel monitoring center. Those types of 'False Positive' to 'fire' or 'person' objects were frequently taking place for lights of working vehicle, reflecting sunlight at tunnel entrance, long black feature which occurs to the part of lane or car, etc. To solve this problem, a deep learning model was developed by simultaneously training the 'False Positive' data generated in the field and the labeling data. As a result, in comparison with the model that was trained only by the existing labeling data, the re-inference performance with respect to the labeling data was improved. In addition, re-inference of the 'False Positive' data shows that the number of 'False Positive' for the persons were more reduced in case of training model including many 'False Positive' data. By training of the 'False Positive' data, the capability of field application of the deep learning model was improved automatically.

• Annual Conference on Human and Language Technology
• /
• 2022.10a
• /
• pp.262-267
• /
• 2022

이커머스 리뷰와 같은 특정 도메인의 경우, 텍스트 표현벡터 학습을 위한 양질의 오픈 학습 데이터를 구하기 어렵다. 또한 사람이 수동으로 검수하며 학습데이터를 만드는 경우, 많은 시간과 비용을 소모하게 된다. 따라서 본 논문에서는 수동으로 검수된 데이터없이 양질의 텍스트 표현벡터를 만들 수 있도록 두 단계의 대조 학습 시스템을 제안한다. 이 두 단계 대조 학습 시스템은 레이블링 된 학습데이터가 필요하지 않은 자기지도 학습 단계와 리뷰의 특성을 고려한 자동 레이블링 기반의 지도 학습 단계로 구성된다. 또한 노이즈에 강한 오류함수와 한국어에 유효한 데이터 증강 기법을 적용한다. 그 결과 스피어먼 상관 계수 기반의 성능 평가를 통해, 베이스 모델과 비교하여 성능을 14.03 향상하였다.

• The Journal of the Korea Contents Association
• /
• v.14 no.1
• /
• pp.14-23
• /
• 2014

There have been many researches for XML as the international standard to store and exchange data on the internet. Among these research fields, we focus on the techniques labeling the nodes of the XML tree that is required for querying the structural information. A labeling scheme assigns the unique label to the nodes and supports the queries for the structural information such as Ancestor-Descendant and Parent-Child relationships. In this paper, we propose a labeling scheme using vector representation where the assigned labels are not altered although XML documents are changed dynamically. Our labeling scheme reduces the storage requirement for the labels of the XML tree and provides the efficient query by using the fixed-length labels with a short size. Result of performance evaluation shows that our labeling scheme is superior to the previous approaches.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.25 no.6
• /
• pp.585-591
• /
• 2015

In transcription for spoken language research, labeling is a work linking text-represented utterance to recorded speech. Most existing labeling tools have been working manually. Semi-automatic labeling we are proposing consists of automation module and manual adjustment module. Automation module extracts voice boundaries utilizing G.Saha's algorithm, and predicts utterance boundaries using the number and length of utterance which established utterance text. For maintaining existing manual tool's accuracy, we provide manual adjustment user interface revising the auto-labeling utterance boundaries. The implemented tool of our semi-automatic algorithm speed up to 27% than existing manual labeling tools.

• Journal of the Korea Society of Computer and Information
• /
• v.16 no.4
• /
• pp.169-177
• /
• 2011

As demand for efficiency in handling dynamic XML data grows, new dynamic XML labeling schemes have been researched. The key idea of the dynamic XML labeling scheme is to find ancestor-descendent-sibling relationships and to minimize memory space to store total label, response time and range of relabeling incurred by update operations. The prime number labeling scheme is a representative scheme which supports dynamic XML documents. It determines the ancestor-descendant relationships between two elements by a simple divisibility test of labels. When a new element is inserted into the XML data using this scheme, it does not change the label values of existing nodes. However, since each prime number must be used exclusively, labels can become significantly large. Therefore, in this paper, we introduce a novel technique to effectively reduce the problem of label overflow. The suggested idea is based on tree decomposition. When label overflow occurs, the full tree is divided into several sub-trees, and nodes in each sub-tree are separately labeled. Through experiments, we show the effectiveness of our scheme.

• Journal of KIISE:Computing Practices and Letters
• /
• v.16 no.1
• /
• pp.130-134
• /
• 2010

When an XML document is updated in order to represent correctly the structural relationships of nodes in a document, the existing XML labeling schemes relabel nodes or use a labeling scheme that the label of a node has much information. However, the relabeling on large XML documents needs many labeling costs and the labeling scheme that the label of a node has much information requires many storage costs. Therefore, the existing labeling schemes degrade significantly query processing performance on dynamic XML documents. This paper proposes the bit-pattern labeling scheme that solves the problems of the existing schemes. The proposed labeling scheme outperforms the existing labeling schemes because the structural relationships of nodes are represented with a bit string.

• Journal of KIISE:Databases
• /
• v.36 no.3
• /
• pp.180-188
• /
• 2009

Recently, the needs for an efficient and secure access control method of dynamic XML data in a ubiquitous data streams environment have become an active research area. In this paper, we proposed an improved role-based prime number labeling scheme for an efficient and secure access control labeling method in dynamic XML data streams. And we point out the limitations of existing access control and labeling schemes for XML data assuming that documents are frequently updated. The improved labeling method where labels are encoded ancestor-descendant and sibling relationships between nodes but need not to be regenerated when the document is updated. Our improved role-based prime number labeling scheme supports an infinite number of updates and guarantees the arbitrary nodes insertion at arbitrary position of the XML tree without label collisions. Also we implemented an efficient access control using a role-based prime number labeling. Finally, we have shown that our approach is an efficient and secure through experiments.

• Journal of the Korea Society of Computer and Information
• /
• v.12 no.5
• /
• pp.65-72
• /
• 2007

Labeling schemes which don't consider about frequent update in dynamic XML documents need relabeling process to reflect the changed label information whenever the tree of XML document is update. There is disadvantage of considerable expenses in the dynamic XML document which can occurs frequent update. To solve this problem, we suggest prime number labeling scheme that doesn't need relabeling process. However the prime number labeling scheme does not consider that it needs to update the sibling order of nodes in the tree of XML document. This update process needs much costs because the most of the tree of XML document has to be researched and rewritten. In this paper, we propose the prime number labeling scheme with sibling order value that can maintain the sibling order without researching or rewriting the tree of XML documents.

• Journal of KIISE:Software and Applications
• /
• v.31 no.7
• /
• pp.922-928
• /
• 2004

We propose a new 3 dimensional labeling method based on slice information for the volume data. This method is named SIL (Slice Information based Labeling). Compare to the conventional algorithms, it has advantages that the use of memory is efficient and it Is possible to combine with a variety of 2 dimensional labeling algorithms for finding an appropriate labeling algorithm to its application. In this study, we applied SIL to confocal microscopy images of cervix cancer cell and compared the results of labeling. According to the measurement, we found that the speed of Sd combined with, CCCL (Contour based Connected Component Labeling) is almost 2 times higher than that of other methods. In conclusion, considering that the performance of labeling depends on a kind of image, we obtained that the proposed method provide better result for the confocal microscopy cell volume data.