• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.12
• /
• pp.1769-1776
• /
• 2022

Automatically summarizing long sentences is an important technique. The BART model is one of the widely used models in the summarization task. In general, in order to generate a summarization model of a specific domain, fine-tuning is performed by re-training a language model trained on a large dataset to fit the domain. The fine-tuning is usually done by changing the number of nodes in the last fully connected layer. However, in this paper, we propose a fine-tuning method by adding an attention layer, which has been recently applied to various models and shows good performance. In order to evaluate the performance of the proposed method, various experiments were conducted, such as accumulating layers deeper, fine-tuning without skip connections during the fine tuning process, and so on. As a result, the BART model using two attention layers with skip connection shows the best score.

• Journal of Auto-vehicle Safety Association
• /
• v.14 no.3
• /
• pp.65-70
• /
• 2022

Lane change in urban environments is a challenge for both human-driving and automated driving due to their complexity and non-linearity. With the recent development of deep-learning, the use of the RNN network, which uses time series data, has become the mainstream in this field. Many researches using RNN show high accuracy in highway environments, but still do not for urban environments where the surrounding situation is complex and rapidly changing. Therefore, this paper proposes a lane change possibility decision network by adopting Attention layer, which is an SOTA in the field of seq2seq. By weighting each time step within a given time horizon, the context of the road situation is more human-like. A total 7D vectors of x, y distances and longitudinal relative speed of side front and rear vehicles, and longitudinal speed of ego vehicle were used as input. A total 5,614 expert data of 4,098 yield cases and 1,516 non-yield cases were used for training, and the performance of this network was tested through 1,817 data. Our network achieves 99.641% of test accuracy, which is about 4% higher than a network using only LSTM in an urban environment. Furthermore, it shows robust behavior to false-positive or true-negative objects.

• Annual Conference on Human and Language Technology
• /
• 2019.10a
• /
• pp.120-124
• /
• 2019

딥러닝은 자연어처리, 이미지 처리, 음성인식 등에서 우수한 성능을 보이고 있다. 그렇지만 복잡한 인공신경망 내부에서 어떠한 동작이 일어나는지 검증하지 못하고 있다. 본 논문에서는 비디오 캡셔닝 분야에서 인공신경망 내부에서 어떠한 동작이 이루어지는지 검사한다. 이를 위해서 우리는 각 단계에서 출력층을 추가하였다. 출력된 결과를 검토하여 인공 신경망의 정상동작 여부를 검증한다. 우리는 한국어 MSR-VTT에 적용하여 우리의 방법을 평가하였다. 이러한 방법을 통해 인공 신경망의 동작을 이해하는데 도움을 줄 수 있을 것으로 기대된다.

• Annual Conference on Human and Language Technology
• /
• 2017.10a
• /
• pp.35-40
• /
• 2017

기계 독해(Machine reading comprehension)는 주어진 문맥을 이해하고, 질문에 적합한 답을 문맥 내에서 찾는 문제이다. Simple Recurrent Unit (SRU)은 Gated Recurrent Unit (GRU)등과 같이 neural gate를 이용하여 Recurrent Neural Network (RNN)에서 발생하는 vanishing gradient problem을 해결하고, gate 입력에서 이전 hidden state를 제거하여 GRU보다 속도를 향상시킨 모델이며, Self-matching Network는 R-Net 모델에서 사용된 것으로, 자기 자신의 RNN sequence에 대하여 어텐션 가중치 (attention weight)를 계산하여 비슷한 의미 문맥 정보를 볼 수 있기 때문에 상호참조해결과 유사한 효과를 볼 수 있다. 본 논문에서는 한국어 기계 독해 데이터 셋을 구축하고, 여러 층의 SRU를 이용한 Encoder에 Self-matching layer를 추가한 $S^2$-Net 모델을 제안한다. 실험 결과, 본 논문에서 제안한 $S^2$-Net 모델이 한국어 기계 독해 데이터 셋에서 EM 65.84%, F1 78.98%의 성능을 보였다.

• 한국어정보학회:학술대회논문집
• /
• 2017.10a
• /
• pp.35-40
• /
• 2017

기계 독해(Machine reading comprehension)는 주어진 문맥을 이해하고, 질문에 적합한 답을 문맥 내에서 찾는 문제이다. Simple Recurrent Unit (SRU)은 Gated Recurrent Unit (GRU)등과 같이 neural gate를 이용하여 Recurrent Neural Network (RNN)에서 발생하는 vanishing gradient problem을 해결하고, gate 입력에서 이전 hidden state를 제거하여 GRU보다 속도를 향상시킨 모델이며, Self-matching Network는 R-Net 모델에서 사용된 것으로, 자기 자신의 RNN sequence에 대하여 어텐션 가중치 (attention weight)를 계산하여 비슷한 의미 문맥 정보를 볼 수 있기 때문에 상호참조해결과 유사한 효과를 볼 수 있다. 본 논문에서는 한국어 기계 독해 데이터 셋을 구축하고, 여러 층의 SRU를 이용한 Encoder에 Self-matching layer를 추가한 $S^2$-Net 모델을 제안한다. 실험 결과, 본 논문에서 제안한 $S^2$-Net 모델이 한국어 기계 독해 데이터 셋에서 EM 65.84%, F1 78.98%의 성능을 보였다.

• Smart Media Journal
• /
• v.11 no.10
• /
• pp.65-75
• /
• 2022

Deep learning is used as a creative tool that could overcome the limitations of existing analysis models and generate various types of results such as text, image, and music. In this paper, we propose a method necessary to preprocess audio data using the Niko's MIDI Pack sound source file as a data set and to generate music using Bi-LSTM. Based on the generated root note, the hidden layers are composed of multi-layers to create a new note suitable for the musical composition, and an attention mechanism is applied to the output gate of the decoder to apply the weight of the factors that affect the data input from the encoder. Setting variables such as loss function and optimization method are applied as parameters for improving the LSTM model. The proposed model is a multi-channel Bi-LSTM with attention that applies notes pitch generated from separating treble clef and bass clef, length of notes, rests, length of rests, and chords to improve the efficiency and prediction of MIDI deep learning process. The results of the learning generate a sound that matches the development of music scale distinct from noise, and we are aiming to contribute to generating a harmonistic stable music.

• Phonetics and Speech Sciences
• /
• v.12 no.2
• /
• pp.29-37
• /
• 2020

Recently, the neural network-based deep learning algorithm has dramatically improved performance compared to the classical Gaussian mixture model based hidden Markov model (GMM-HMM) automatic speech recognition (ASR) system. In addition, researches on end-to-end (E2E) speech recognition systems integrating language modeling and decoding processes have been actively conducted to better utilize the advantages of deep learning techniques. In general, E2E ASR systems consist of multiple layers of encoder-decoder structure with attention. Therefore, E2E ASR systems require data with a large amount of speech-text paired data in order to achieve good performance. Obtaining speech-text paired data requires a lot of human labor and time, and is a high barrier to building E2E ASR system. Therefore, there are previous studies that improve the performance of E2E ASR system using relatively small amount of speech-text paired data, but most studies have been conducted by using only speech-only data or text-only data. In this study, we proposed a semi-supervised training method that enables E2E ASR system to perform well in corpus in different domains by using both speech or text only data. The proposed method works effectively by adapting to different domains, showing good performance in the target domain and not degrading much in the source domain.