• International conference on construction engineering and project management
• /
• 2022.06a
• /
• pp.847-854
• /
• 2022

The aim of this study is to develop a Named Entity Recognition (NER) model to automatically identify construction-related organizations from news articles. This study collected news articles using web crawling technique and construction-related organizations were labeled within a total of 1,000 news articles. The Bidirectional Encoder Representations from Transformers (BERT) model was used to recognize clients, constructors, consultants, engineers, and others. As a pilot experiment of this study, the best average F1 score of NER was 0.692. The result of this study is expected to contribute to the establishment of international business strategies by collecting timely information and analyzing it automatically.

• Journal of the Korean Society for Precision Engineering
• /
• v.13 no.9
• /
• pp.138-148
• /
• 1996

In moving structures such as robots and feeders of production lines, vibrations may not be ignored. Recently it becomes a big problem to control the vibration in a motion because moving structures are in higher speed, larger size and lighter weight. In this study a nonlinear system was model- led and identified by using neural networks and the vibration in motions was controlled actively by using a neural network controller. To investigate vilidity of this method, an experimental apparatus was made and tested. The model was composed of a DC servomotor, a carrier and a flexible plate. Its motion was measured by a gap sensor and an encoder. Trapezoidal, cycloid and trapecloid type trajectories were used in this exper- riment. Computer simulations and experiments weredone for each trajectory.

• Annual Conference on Human and Language Technology
• /
• 2019.10a
• /
• pp.3-7
• /
• 2019

자연어 표상은 자연어가 가진 정보를 컴퓨터에게 전달하기 위해 표현하는 방법이다. 현재 자연어 표상은 학습을 통해 고정된 벡터로 표현하는 것이 아닌 문맥적 정보에 의해 벡터가 변화한다. 그 중 BERT의 경우 Transformer 모델의 encoder를 사용하여 자연어를 표상하는 기술이다. 하지만 BERT의 경우 학습시간이 많이 걸리며, 대용량의 데이터를 필요로 한다. 본 논문에서는 빠른 자연어 표상 학습을 위해 의미 정보와 BERT를 결합한 개념 언어 모델을 제안한다. 의미 정보로 단어의 품사 정보와, 명사의 의미 계층 정보를 추상적으로 표현했다. 실험을 위해 ETRI에서 공개한 한국어 BERT 모델을 비교 대상으로 하며, 개체명 인식을 학습하여 비교했다. 두 모델의 개체명 인식 결과가 비슷하게 나타났다. 의미 정보가 자연어 표상을 하는데 중요한 정보가 될 수 있음을 확인했다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.05a
• /
• pp.639-641
• /
• 2023

이미지 기반 폐기물 처리시스템에서 품목별 상이한 수집 난이도로 인해 발생하는 데이터 불균형으로 분류 모델 학습에 어려움이 따른다. 따라서 본 논문에서는 폐기물 분류 모델의 성능 비교를 통해 적합한 이미지 생성 모델을 탐색한다. 데이터의 불균형을 해결할 수 있도록 VAE(Variational Auto-Encoder), GAN(Generative Adversarial Networks) 및 Diffusion Model을 이용하여 이미지를 생성한다. 이후 각각의 생성 방법에 따라 학습데이터와 병합하여 객체 분류를 진행하였다. 정확도는 VAE가 84.41%로 3.3%의 성능 향상을, F1-점수는 Diffusion Model이 91.94%로 6.14%의 성능 향상을 이루었다. 이를 통해, 데이터 수집에서 나타나는 데이터 불균형을 해결하여 실 사용환경에 알맞은 시스템을 구축이 가능함을 확인하였다.

• Nuclear Engineering and Technology
• /
• v.55 no.10
• /
• pp.3716-3724
• /
• 2023

Loose parts monitoring and detecting alarm type in real Nuclear Power Plant have challenges such as background noise, insufficient alarm data, and difficulty of distinction between alarm data that occur during start and stop. Although many signal processing methods and alarm determination algorithms have been developed, it is not easy to determine valid alarm and extract the meaning data from alarm signal including background noise. To address these issues, this paper proposes a denoising autoencoder-based majority vote classification. Training and test data are prepared by acquiring alarm data from real NPP and simulation facility for data augmentation, and noisy data is reproduced by adding Gaussian noise. Using DAEs with 3, 5, 7, and 9 layers, features are extracted for each model and classified into neural networks. Finally, the results obtained from each DAE are classified by majority voting. Also, through comparison with other methods, the accuracy and the false alarm rate are compared, and the excellence of the proposed method is confirmed.

• Journal of Internet Computing and Services
• /
• v.21 no.3
• /
• pp.133-144
• /
• 2020

The number of elderly people with dementia is increasing as fast as the proportion of older people due to aging, which creates a social and economic burden. In particular, dementia care costs, including indirect costs such as increased care costs due to lost caregiver hours and caregivers, have grown exponentially over the years. In order to reduce these costs, it is urgent to introduce a management system to care for dementia patients. Therefore, this study proposes a sensor-based abnormal behavior detection system to manage dementia patients who live alone or in an environment where they cannot always take care of dementia patients. Existing studies were merely evaluating behavior or evaluating normal behavior, and there were studies that perceived behavior by processing images, not data from sensors. In this study, we recognized the limitation of real data collection and used both the auto-encoder, the unsupervised learning model, and the LSTM, the supervised learning model. Autoencoder, an unsupervised learning model, trained normal behavioral data to learn patterns for normal behavior, and LSTM further refined classification by learning behaviors that could be perceived by sensors. The test results show that each model has about 96% and 98% accuracy and is designed to pass the LSTM model when the autoencoder outlier has more than 3%. The system is expected to effectively manage the elderly and dementia patients who live alone and reduce the cost of caring.

• Journal of the Korean Association of Geographic Information Studies
• /
• v.27 no.1
• /
• pp.115-127
• /
• 2024

This research aimed to construct models with various structures based on the Transformer module and to perform land cover classification, thereby examining the applicability of the Transformer module. For the classification of land cover, the Unet model, which has a CNN structure, was selected as the base model, and a total of four deep learning models were constructed by combining both the encoder and decoder parts with the Transformer module. During the training process of the deep learning models, the training was repeated 10 times under the same conditions to evaluate the generalization performance. The evaluation of the classification accuracy of the deep learning models showed that the Model D, which utilized the Transformer module in both the encoder and decoder structures, achieved the highest overall accuracy with an average of approximately 89.4% and a Kappa coefficient average of about 73.2%. In terms of training time, models based on CNN were the most efficient. however, the use of Transformer-based models resulted in an average improvement of 0.5% in classification accuracy based on the Kappa coefficient. It is considered necessary to refine the model by considering various variables such as adjusting hyperparameters and image patch sizes during the integration process with CNN models. A common issue identified in all models during the land cover classification process was the difficulty in detecting small-scale objects. To improve this misclassification phenomenon, it is deemed necessary to explore the use of high-resolution input data and integrate multidimensional data that includes terrain and texture information.

• ETRI Journal
• /
• v.42 no.2
• /
• pp.258-271
• /
• 2020

A correlation exists between luminance samples and chrominance samples of a color image. It is beneficial to exploit such interchannel redundancy for color image compression. We propose an algorithm that predicts chrominance components Cb and Cr from the luminance component Y. The prediction model is trained by supervised learning with Laplacian-regularized least squares to minimize the total prediction error. Kernel principal component analysis mapping, which reduces computational complexity, is implemented on the same point set at both the encoder and decoder to ensure that predictions are identical at both the ends without signaling extra location information. In addition, chrominance subsampling and entropy coding for model parameters are adopted to further reduce the bit rate. Finally, luminance information and model parameters are stored for image reconstruction. Experimental results show the performance superiority of the proposed algorithm over its predecessor and JPEG, and even over JPEG-XR. The compensation version with the chrominance difference of the proposed algorithm performs close to and even better than JPEG2000 in some cases.

• ETRI Journal
• /
• v.38 no.1
• /
• pp.185-194
• /
• 2016

In this paper, a novel rate control scheme for low delay video coding of High Efficiency Video Coding (HEVC) is proposed. The proposed scheme is developed by considering a new temporal prediction structure of HEVC. In the proposed scheme, the relationship between bit rate and quantization step is exploited firstly to formulate an accurate quadratic rate-quantization (R-Q) model. Secondly, a method of determining the quantization parameters (QPs) for the first frames within a group of pictures is proposed. Thirdly, an accurate frame-level bit allocation method is proposed for HEVC. Finally, based on the proposed R-Q model and the target bit allocated for the frame, the QPs are predicted for coding tree units by using rate-distortion (R-D) optimization. We compare our scheme against that of three other state-of-the-art rate control schemes. Experimental results show that the proposed rate control scheme can increase the Bjøntegaard delta peak signal-to-noise ratio by 0.65 dB and 0.09 dB on average compared with the JCTVC-I0094 and JCTVC-M0036 schemes, respectively, both of which have been implemented in an HEVC test model encoder; furthermore, the proposed scheme achieves a similar R-D performance to Wang's scheme, as well as obtaining the smallest bit rate mismatch error of all the schemes.

• Journal of Intelligence and Information Systems
• /
• v.28 no.3
• /
• pp.45-55
• /
• 2022

Recently, large-scale language models (LPLM) have been shown state-of-the-art performances in various tasks of natural language processing including intent classification. However, fine-tuning LPLM requires much computational cost for training and inference which is not appropriate for dialog system. In this paper, we propose compressed intent classification model for multi-agent in low-resource like CPU. Our method consists of two stages. First, we trained sentence encoder from LPLM then compressed it through knowledge distillation. Second, we trained agent-specific adapter for intent classification. The results of three intent classification datasets show that our method achieved 98% of the accuracy of LPLM with only 21% size of it.