최근에 선박을 안전하게 설계 및 운항하기 위해 인공지능으로 운동성능을 예측하는 연구가 늘고 있다. 하지만 일반적인 선박 에 비해 소형 어선에 대한 연구는 부족한 실정이다. 본 논문에서는 소형 어선의 운동성능 계산에 필수적인 운동응답을 심층신경망으로 추정하는 모델을 제안한다. 15척의 소형 어선에 대하여 유체동역학 해석을 수행하였으며 이를 통해 데이터베이스를 구축하였다. 환경 조 건과 주요 제원을 입력 데이터로, 단위 파고에 대한 운동응답(Response Amplitude Operator)을 출력 데이터로 설정하였다. 훈련된 심층신경 망 모델을 통해 예측된 운동응답은 유체동역학 해석 결과와 유사한 경향을 보이며 고주파 성분을 가진 운동응답 함수를 낮은 오차로 근 사하는 결과를 보여준다. 본 연구의 결과를 바탕으로 어선의 선형 특성 고려한 심층신경망 모델로 확장하여 연구 결과의 활용도를 넓히 고자 한다.

Yong-hun Lee, JeeHee Yu, and Tae-Jin Yoon. 2017. Predicting the Occurrence of the English Modals Can and May Using Deep Neural Networks. Studies in Modern Grammar 96, 167-189. This paper tries to provide a computational modeling of language processing using deep neural networks. For this purpose, the corpus data in the ICE-USA was used. After all the sentences with can and may were encoded with eighteen linguistic factors, the annotated data were fed into the deep neural networks (DNN). The DNN was constructed with three layers, and each layer contained seventeen nodes. After the DNN was constructed, the learning process was performed with a training set. Then, the performance was measured with a test set. The processes were repeated one hundred times, and it was observed that the DNN had the classification accuracy of 91.5%. The results are promising in that reliable methods can be used in automatically classifying the frequently used modal auxiliary on the basis of the deep learning system.

One of the most frequently performed tasks in human-robot interaction (HRI), intelligent vehicles, and security systems is face related applications such as face recognition, facial expression recognition, driver state monitoring, and gaze estimation. In these applications, accurate head pose estimation is an important issue. However, conventional methods have been lacking in accuracy, robustness or processing speed in practical use. In this paper, we propose a novel method for estimating head pose with a monocular camera. The proposed algorithm is based on a deep neural network for multi-task learning using a small grayscale image. This network jointly detects multi-view faces and estimates head pose in hard environmental conditions such as illumination change and large pose change. The proposed framework quantitatively and qualitatively outperforms the state-of-the-art method with an average head pose mean error of less than 4.5° in real-time.