• Proceedings of the Korea Water Resources Association Conference
• /
• 2023.05a
• /
• pp.450-450
• /
• 2023

인공지능 기법의 급격한 발달에 따라 다양한 분야에서 인공지능 기법을 활용하기 위한 노력이 이루어지고 있다. 재난은 발생하기 전에 다양한 전조 현상을 나타내나 수많은 정보 속에서 전조 증상을 정확히 인지하는 것은 매우 어렵다. 따라서 인공지능은 방대한 사전 정보의 해석을 통해 재난 발생의 전조를 신속 정확하게 감지하는데 최적의 기술이다. 최근 OpenAI의 딥러닝 기반의 언어모델인 GPT(Generative Pre-trained Transformer)의 성능이 기대 이상을 나타내면서 많은 분야에서 GPT에 대한 관심과 실험이 시작되고 있다. 본 실험에서는 GPT를 이용하여 재난 검색 및 해석의 특징을 검토하여 보았다. 정확한 재난 기록은 정확한 재난 예측을 위해 반드시 필요한 자료이나 부정확한 재난 기록은 그 기록이 비록 방대하더라도 오히려 예측의 신뢰도를 크게 떨어뜨린 수 있다. 따라서 비지도학습 기반의 대화형 인공지능을 재난 검색에 활용하기 위해서는 인공지능 기법의 인지 특성을 반드시 가늠해 봐야 한다. 향후 보다 많은 연구자가 이에 관심을 가진다면 보다 정확한 인공지능 기반의 재난 탐지 기술의 개발이 가능할 것으로 기대된다.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2019.05a
• /
• pp.160-161
• /
• 2019

As a method for modeling a three-dimensional object, there are a method using a 3D scanner, a method using a motion capture system, and a method using a Kinect system. Through this method, a portion that is not captured due to occlusion occurs in the process of creating a three-dimensional object. In order to implement a perfect three-dimensional object, it is necessary to arbitrarily fill the obscured part. There is a technique to fill the unexposed part by various image processing methods. In this study, we propose a method using GANs, which is the latest trend of unsupervised machine learning, as a method for more natural hole-filling.

• The Journal of Korean Institute of Next Generation Computing
• /
• v.15 no.5
• /
• pp.64-74
• /
• 2019

Although a non-rigid registration has high demands in clinical practice, it has a high computational complexity and it is very difficult for ensuring the accuracy and robustness of registration. This study proposes a method of applying a non-rigid registration to 3D magnetic resonance images of brain in an unsupervised learning environment by using a deep-learning network. A feature vector between two images is produced through the network by receiving both images from two different patients as inputs and it transforms the target image to match the source image by creating a displacement vector field. The network is designed based on a U-Net shape so that feature vectors that consider all global and local differences between two images can be constructed when performing the registration. As a regularization term is added to a loss function, a transformation result similar to that of a real brain movement can be obtained after the application of trilinear interpolation. This method enables a non-rigid registration with a single-pass deformation by only receiving two arbitrary images as inputs through an unsupervised learning. Therefore, it can perform faster than other non-learning-based registration methods that require iterative optimization processes. Our experiment was performed with 3D magnetic resonance images of 50 human brains, and the measurement result of the dice similarity coefficient confirmed an approximately 16% similarity improvement by using our method after the registration. It also showed a similar performance compared with the non-learning-based method, with about 10,000 times speed increase. The proposed method can be used for non-rigid registration of various kinds of medical image data.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2020.11a
• /
• pp.109-112
• /
• 2020

사람 재인식을 수행하기 위해서 많은 연구들이 진행되어 좋은 결과들을 보였다 그러나 이 결과들은 라벨이 있는 도메인에서의 지도 학습으로 얻은 결과들이었다. 라벨이 없는 도메인에서의 사람 재인식의 성능은 아직 많이 부족한 상태이다. 사람 재인식을 수행하고자 하는 목표 도메인에 반해 주어진 소스 도메인에서는 라벨이 풍부하다. 지금까지의 논문에서는 소스 도메인에서의 사람 이미지를 목표 도메인의 이미지처럼 만들어서 소스 도메인에서 높은 성능을 보이는 사람 재인식기를 목표 도메인에서도 잘 동작하도록 학습하는 방법들이 주를 이루었다. 하지만 이 방법에서는 소스 도메인의 사람 이미지를 목표 도메인의 이미지와 비슷하게 만들기만하고 사람의 신원에 대한 일관성을 유지시키지는 못하였다. 본 논문에서는 비지도 도메인 적응 사람 재인식을 수행하기 위해 클러스터 일관성(cluster consistency)을 유지하는 기법을 제안한다. 제안한 방법은 사람의 신원에 대한 일관성을 유지시켜서 사람 재인식의 성능을 높인다.

• KIPS Transactions on Software and Data Engineering
• /
• v.12 no.8
• /
• pp.355-364
• /
• 2023

As advanced cyber threats continue to increase in recent years, it is difficult to detect new types of cyber attacks with existing pattern or signature-based intrusion detection method. Therefore, research on anomaly detection methods using data learning-based artificial intelligence technology is increasing. In addition, supervised learning-based anomaly detection methods are difficult to use in real environments because they require sufficient labeled data for learning. Research on an unsupervised learning-based method that learns from normal data and detects an anomaly by finding a pattern in the data itself has been actively conducted. Therefore, this study aims to extract a latent vector that preserves useful sequence information from sequence log data and develop an anomaly detection learning model using the extracted latent vector. Word2Vec was used to create a dense vector representation corresponding to the characteristics of each sequence, and an unsupervised autoencoder was developed to extract latent vectors from sequence data expressed as dense vectors. The developed autoencoder model is a recurrent neural network GRU (Gated Recurrent Unit) based denoising autoencoder suitable for sequence data, a one-dimensional convolutional neural network-based autoencoder to solve the limited short-term memory problem that GRU can have, and an autoencoder combining GRU and one-dimensional convolution was used. The data used in the experiment is time-series-based NGIDS (Next Generation IDS Dataset) data, and as a result of the experiment, an autoencoder that combines GRU and one-dimensional convolution is better than a model using a GRU-based autoencoder or a one-dimensional convolution-based autoencoder. It was efficient in terms of learning time for extracting useful latent patterns from training data, and showed stable performance with smaller fluctuations in anomaly detection performance.

• Journal of the Korea Society of Computer and Information
• /
• v.27 no.5
• /
• pp.77-84
• /
• 2022

In this paper, we propose a novel unsupervised feature selection method. Conventional unsupervised feature selection method defines virtual label and uses a regression analysis that projects the given data to this label. However, since virtual labels are generated from data, they can be formed similarly in the space. Thus, in the conventional method, the features can be selected in only restricted space. To solve this problem, in this paper, features are selected using orthogonal projections and low-rank approximations. To solve this problem, in this paper, a virtual label is projected to orthogonal space and the given data set is also projected to this space. Through this process, effective features can be selected. In addition, projection matrix is restricted low-rank to allow more effective features to be selected in low-dimensional space. To achieve these objectives, a cost function is designed and an efficient optimization method is proposed. Experimental results for six data sets demonstrate that the proposed method outperforms existing conventional unsupervised feature selection methods in most cases.

• Annual Conference on Human and Language Technology
• /
• 2022.10a
• /
• pp.232-237
• /
• 2022

본 논문에서는 미세 조정(fine-tuning) 및 비지도 학습 기법을 사용하여 경제 분야 텍스트인 금융 리포트에 대해 속성 기반 감성 분석(aspect-based sentiment analysis) 데이터셋을 반자동적으로 구축할 수 있는 방법론에 대한 연구를 수행하였다. 구축 시에는 속성기반 감성분석 주석 요소 중 극성, 속성 카테고리 정보를 부착하였으며, 미세조정과 비지도 학습 기법인 BERTopic을 통해 주석 요소를 자동적으로 부착하는 한편 이를 수동으로 검수하여 데이터셋의 완성도를 높이고자 하였다. 데이터셋에 대한 실험 결과, 극성 반자동 주석의 경우 기존에 구축된 데이터셋과 비슷한 수준의 성능을 보였다. 한편 정성적 분석을 통해 자동 구축을 동일하게 수행하였더라도 기술의 원리와 발달 정도에 따라 결과가 상이하게 달라짐을 관찰함으로써 경제 도메인의 ABSA 데이터셋 구축에 여전히 발전 여지가 있음을 확인할 수 있었다.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.21 no.9
• /
• pp.345-351
• /
• 2020

Administrative agencies today are paying keen attention to big data analysis to improve their policy responsiveness. Of all the big data, news articles can be used to understand public opinion regarding policy and policy issues. The amount of news output has increased rapidly because of the emergence of new online media outlets, which calls for the use of automated bots or automatic document classification tools. There are, however, limits to the automatic collection of news articles related to specific agencies or departments based on the existing news article categories and keyword search queries. Thus, this paper proposes a method to process articles using classification glossaries that take into account each agency's different work features. To this end, classification glossaries were developed by extracting the work features of different departments using Word2Vec and topic modeling techniques from news articles related to different agencies. As a result, the automatic classification of newspaper articles for each department yielded approximately 71% accuracy. This study is meaningful in making academic and practical contributions because it presents a method of extracting the work features for each department, and it is an unsupervised learning-based automatic classification method for automatically classifying news articles relevant to each agency.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.05a
• /
• pp.618-620
• /
• 2023

최근 자연어처리 분야에서 의미론적 유사성을 반영하기 위한 대조 학습 (contrastive learning) 관련 연구가 활발히 이뤄지고 있다. 이러한 대조 학습의 핵심은 의미론적으로 가까워져야 하는 쌍과 멀어져야 하는 쌍을 잘 구축하는 것이지만, 기존의 손실 함수는 문장의 상대적인 유사성을 풍부하게 반영하는데 한계가 있다. 이를 해결하기 위해, 이전 연구에서는 삼중 항 손실 함수 (triplet loss)를 도입하였으며, 본 논문에서는 이러한 삼중 항을 구성하기 위해 대조 학습에서의 효과적인 토큰 컷오프(cutoff) 데이터 증강 기법을 제안한다. BERT, RoBERTa 등 널리 활용되는 언어 모델을 이용한 실험을 통해 제안하는 방법의 우수한 성능을 보인다.

• Journal of the Korea Computer Graphics Society
• /
• v.27 no.5
• /
• pp.45-54
• /
• 2021

Recently, deep learning-based denoising approaches have been actively studied. In particular, with the advances of blind denoising techniques, it become possible to train a deep learning-based denoising model only with noisy images in an image domain where it is impossible to obtain a clean image. We no longer require pairs of a clean image and a noisy image to obtain a restored clean image from the observation. However, it is difficult to recover the target using a deep learning-based denoising model trained by only noisy images if the distribution of the noisy image is far from the distribution of the clean image. To address this limitation, unpaired image denoising approaches have recently been studied that can learn the denoising model from unpaired data of the noisy image and the clean image. ISCL showed comparable performance close to that of supervised learning-based models based on pairs of clean and noisy images. In this study, we propose suitable normalization techniques for each purpose of architectures (e.g., generator, discriminator, and extractor) of ISCL. We demonstrate that the proposed method outperforms state-of-the-art unpaired image denoising approaches including ISCL.