• Annual Conference on Human and Language Technology
• /
• 2021.10a
• /
• pp.274-277
• /
• 2021

의존 구문 분석은 문장 내 의존소와 지배소 사이의 관계를 예측하여 문장 구조를 분석하는 자연어처리 태스크이다. 최근의 딥러닝 기반 의존 구문 분석 연구는 주로 포인터 네트워크를 사용하는 방법으로 연구되고 있다. 포인터 네트워크는 내부적으로 사용하는 attention 기법에 따라 성능이 달라질 수 있다. 따라서 본 논문에서는 포인터 네트워크 모델에 적용되는 attention 기법들을 비교 분석하고, 한국어 의존 구문 분석 모델에 가장 효과적인 attention 기법을 선별한다. KLUE 데이터 셋을 사용한 실험 결과, UAS는 biaffine attention을 사용할 때 95.14%로 가장 높은 성능을 보였으며, LAS는 multi-head attention을 사용했을 때 92.85%로 가장 높은 성능을 보였다.

• Journal of Korea Multimedia Society
• /
• v.18 no.3
• /
• pp.359-367
• /
• 2015

In this paper, we propose intelligent fire learning and detection system using hybrid visual attention mechanism of human. Proposed fire learning system generates leaned data by learning process of fire and smoke images. The features used as learning feature are selected among many features which are extracted based on bottom-up visual attention mechanism of human, and these features are modified as learned data by calculating average and standard variation of them. Proposed fire detection system uses learned data which is generated in fire learning system and features of input image to detect fire.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2021.06a
• /
• pp.28-31
• /
• 2021

전세계적으로 우울증은 정신 건강 질환으로써 문제가 되고 있으며, 이를 해결하기 위해 일상생활에서의 우울증 탐지에 대한 연구가 진행되고 있다. 따라서 본 논문에서는 일상생활에 밀접하게 연관되어 있는 AI 스피커를 사용한 어텐션 메커니즘(Attention Mechanism) 기반 멀티모달 우울증 감지 시스템을 제안한다. 제안된 방법은 AI 스피커로부터 수집할 수 있는 음성 및 텍스트 데이터를 수집하고 CNN(Convolutional Neural Network)과 BiLSTM(Bidirectional Long Short-Term Memory Network)를 통해 각 데이터에서의 학습을 진행한다. 학습과정에서 Self-Attention 을 적용하여 특징 벡터에 추가적인 가중치를 부여하는 어텐션 메커니즘을 사용한다. 최종적으로 음성 및 텍스트 데이터에서 어텐션 가중치가 추가된 특징들을 합하여 SoftMax 를 통해 우울증 점수를 예측한다.

• The Journal of the Acoustical Society of Korea
• /
• v.39 no.2
• /
• pp.137-142
• /
• 2020

Many studies based on I-vector have been conducted in a variety of environments, from text-dependent short-utterance to text-independent long-utterance. In this paper, we propose a speaker verification system employing a combination of I-vector with Probabilistic Linear Discriminant Analysis (PLDA) and speaker embedding of Long Short Term Memory (LSTM) with attention mechanism in far-field and noisy environments. The LSTM model's Equal Error Rate (EER) is 15.52 % and the Attention-LSTM model is 8.46 %, improving by 7.06 %. We show that the proposed method solves the problem of the existing extraction process which defines embedding as a heuristic. The EER of the I-vector/PLDA without combining is 6.18 % that shows the best performance. And combined with attention-LSTM based embedding is 2.57 % that is 3.61 % less than the baseline system, and which improves performance by 58.41 %.

• International Journal of Contents
• /
• v.15 no.4
• /
• pp.8-15
• /
• 2019

This paper presents a solution of the 'Quick, Draw! Doodle Recognition Challenge' hosted by Google. Doodles are drawings comprised of concrete representational meaning or abstract lines creatively expressed by individuals. In this challenge, a doodle is presented as a sequence of sketches. From the view of at the sketch level, to learn the pattern of strokes representing a doodle, we propose a sequential model stacked with multiple convolution layers and Long Short-Term Memory (LSTM) cells following the attention mechanism [15]. From the view at the image level, we use multiple models pre-trained on ImageNet to recognize the doodle. Finally, an ensemble and a post-processing method using the minimum cost flow algorithm are introduced to combine multiple models in achieving better results. In this challenge, our solutions garnered 11th place among 1,316 teams. Our performance was 0.95037 MAP@3, only 0.4% lower than the winner. It demonstrates that our method is very competitive. The source code for this competition is published at: https://github.com/ngxbac/Kaggle-QuickDraw.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.16 no.1
• /
• pp.245-265
• /
• 2022

In response to problems such as insufficient extraction information, low detection accuracy, and frequent misdetection in the field of Thangka image defects, this paper proposes a YOLOv5 prediction algorithm fused with the attention mechanism. Firstly, the Backbone network is used for feature extraction, and the attention mechanism is fused to represent different features, so that the network can fully extract the texture and semantic features of the defect area. The extracted features are then weighted and fused, so as to reduce the loss of information. Next, the weighted fused features are transferred to the Neck network, the semantic features and texture features of different layers are fused by FPN, and the defect target is located more accurately by PAN. In the detection network, the CIOU loss function is used to replace the GIOU loss function to locate the image defect area quickly and accurately, generate the bounding box, and predict the defect category. The results show that compared with the original network, YOLOv5-SE and YOLOv5-CBAM achieve an improvement of 8.95% and 12.87% in detection accuracy respectively. The improved networks can identify the location and category of defects more accurately, and greatly improve the accuracy of defect detection of Thangka images.

• Communications for Statistical Applications and Methods
• /
• v.28 no.1
• /
• pp.39-57
• /
• 2021

Time series prediction is an area of great interest to many people. Algorithms for time series prediction are widely used in many fields such as stock price, temperature, energy and weather forecast; in addtion, classical models as well as recurrent neural networks (RNNs) have been actively developed. After introducing the attention mechanism to neural network models, many new models with improved performance have been developed; in addition, models using attention twice have also recently been proposed, resulting in further performance improvements. In this paper, we consider time series prediction by introducing attention twice to an RNN model. The proposed model is a method that introduces H-attention and T-attention for output value and time step information to select useful information. We conduct experiments on stock price, temperature and energy data and confirm that the proposed model outperforms existing models.

• Annual Conference on Human and Language Technology
• /
• 2018.10a
• /
• pp.167-171
• /
• 2018

개체명 인식은 문서에서 인명, 지명, 기관명 등의 고유한 의미를 나타내는 단위인 개체명을 추출하고, 추출된 개체명의 범주를 결정하는 작업이다. 최근 개체명 인식과 관련된 연구는 입력 데이터의 앞, 뒤를 고려하기 위한 Bi-RNNs와 출력 데이터 간의 전이 확률을 이용한 CRFs를 결합한 방식을 기반으로 다양한 변형의 심층학습 방법론이 제안되고 있다. 그러나 대부분의 연구는 입력 단위를 단어나 형태소로 사용하고 있으며, 성능 향상을 위해 띄어쓰기 정보, 개체명 사전 자질, 품사 분포 정보 등 다양한 정보를 필요로 한다는 어려움이 있다. 본 논문은 기본적인 학습 말뭉치에서 얻을 수 있는 문자 기반의 입력 정보와 Multi-Head Attention을 추가한 Bi-GRU/CRFs을 이용한 다국어 개체명 인식 방법을 제안한다. 한국어, 일본어, 중국어, 영어에 제안 모델을 적용한 결과 한국어와 일본어에서는 우수한 성능(한국어 $F_1$ 84.84%, 일본어 $F_1$ 89.56%)을 보였다. 영어에서는 $F_1$ 80.83%의 성능을 보였으며, 중국어는 $F_1$ 21.05%로 가장 낮은 성능을 보였다.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.23 no.6
• /
• pp.89-95
• /
• 2023

In recent decades, human action recognition (HAR) has demonstrated potential applications in sports analysis, human-robot interaction, and large-scale signage content. In this paper, spatial temporal attention graph convolutional network (STAGCN)-based HAR system is proposed. Spatioal-temmporal features of skeleton sequences are assigned different weights by STAGCN, enabling the consideration of key joints and viewpoints. From simulation results, it has been shown that the performance of the proposed model can be improved in terms of classification accuracy in the NTU RGB+D dataset.

• Annual Conference on Human and Language Technology
• /
• 2018.10a
• /
• pp.608-610
• /
• 2018

한국어 의존 구문 분석(Dependency Parsing)은 문장 어절의 중심어(head)와 수식어(modifier)의 의존관계를 표현하는 자연어 분석 방법이다. 최근에는 이러한 의존 관계를 표현하기 위해 주의 집중 메커니즘(Attention Mechanism)과 LSTM(Long Short Term Memory)을 결합한 모델들이 높은 성능을 보이고 있다. 본 논문에서는 개선된 Biaffine Attention 의존 구문 분석 모델을 제안한다. 제안된 모델은 기존의 Biaffine Attention에서 의존성과 의존 관계를 결정하는 방법을 개선하였고, 한국어 의존 구문 분석을 위한 입력 열의 형태소 표상을 확장함으로써 기존의 모델보다 UAS(Unlabeled Attachment Score)가 0.15%p 더 높은 성능을 보였다.