• 방송공학회논문지
• /
• 제27권3호
• /
• pp.341-350
• /
• 2022

본 논문은 온디바이스 환경에서 다중 시점 영상을 입력 받아 객체를 증강하고, 현실 공간에 의한 가려짐을 구현하는 딥러닝 기반의 증강현실 시스템을 제안한다. 이는 세부적으로 카메라 자세 추정, 깊이 추정, 객체 증강 구현의 세 기술적 단계로 나눠지며 각 기법은 온디바이스 환경에서의 최적화를 위해 다양한 모바일 프레임워크를 사용한다. 카메라 자세 추정 단계에서는 많은 계산량을 필요로 하는 특징 추출 알고리즘을 GPU 병렬처리 프레임워크인 OpenCL을 통해 가속하여 사용하며, 깊이 영상 추론 단계에서는 모바일 심층신경망 프레임워크 TensorFlow Lite를 사용하여 가속화된 단안, 다중 영상 기반의 깊이 영상 추론을 수행한다. 마지막으로 모바일 그래픽스 프레임워크 OpenGL ES를 활용해 객체 증강 및 가려짐을 구현한다. 제시하는 증강현실 시스템은 안드로이드 환경에서 GUI를 갖춘 애플리케이션으로 구현되며 모바일과 PC 환경에서의 동작 정확도 및 처리 시간을 평가한다.

• Journal of Radiation Protection and Research
• /
• 제47권3호
• /
• pp.111-133
• /
• 2022

The exciting advancement related to the "modeling of digital human" in terms of a computational phantom for radiation dose calculations has to do with the latest hype related to deep learning. The advent of deep learning or artificial intelligence (AI) technology involving convolutional neural networks has brought an unprecedented level of innovation to the field of organ segmentation. In addition, graphics processing units (GPUs) are utilized as boosters for both real-time Monte Carlo simulations and AI-based image segmentation applications. These advancements provide the feasibility of creating three-dimensional (3D) geometric details of the human anatomy from tomographic imaging and performing Monte Carlo radiation transport simulations using increasingly fast and inexpensive computers. This review first introduces the history of three types of computational human phantoms: stylized medical internal radiation dosimetry (MIRD) phantoms, voxelized tomographic phantoms, and boundary representation (BREP) deformable phantoms. Then, the development of a person-specific phantom is demonstrated by introducing AI-based organ autosegmentation technology. Next, a new development in GPU-based Monte Carlo radiation dose calculations is introduced. Examples of applying computational phantoms and a new Monte Carlo code named ARCHER (Accelerated Radiation-transport Computations in Heterogeneous EnviRonments) to problems in radiation protection, imaging, and radiotherapy are presented from research projects performed by students at the Rensselaer Polytechnic Institute (RPI) and University of Science and Technology of China (USTC). Finally, this review discusses challenges and future research opportunities. We found that, owing to the latest computer hardware and AI technology, computational human body models are moving closer to real human anatomy structures for accurate radiation dose calculations.

• 스마트미디어저널
• /
• 제12권2호
• /
• pp.66-75
• /
• 2023

최근 기업 및 공공기관 정보시스템의 클라우드 전환이 가속화되면서 클라우드 환경에서 운영되는 웹 애플리케이션이 증가하고 있다. 클라우드 웹 애플리케이션에 대한 전통적인 네트워크 공격은 대량의 패킷으로 네트워크 자원을 고갈시키는 DoS(Denial of Service) 공격이 대표적이지만, 최근에는 애플리케이션 자원을 고갈시키는 HTTP DoS 공격도 증가하고 있어 이에 대응하기 위한 보안기술 마련이 필요하다. 특히, HTTP DoS 공격 중, 저대역폭으로 수행되는 공격은 네트워크 자원을 고갈시키지 않기 때문에 네트워크 메트릭을 모니터링 하는 전통적인 보안 솔루션으로 탐지하는 것이 어렵다. 본 논문에서는 클라우드 웹 애플리케이션에 HTTP DoS 공격을 주입하면서 웹 서버의 애플리케이션 메트릭을 수집하고, 이를 머신러닝 기반으로 학습하여 공격을 탐지하는 새로운 탐지 모델을 제안한다. 애플리케이션 메트릭으로는 아파치 웹 서버의 18종을 수집하였고, 5종의 머신러닝 모델과 2종의 딥러닝 모델을 사용하여 수집한 데이터를 학습하였다. 또한, 6종의 네트워크 메트릭을 추가로 수집 및 학습하고, 제안된 애플리케이션 메트릭 기반 모델과 성능을 비교함으로써 애플리케이션 메트릭 기반 머신러닝 모델의 우수성을 검증한다. HTTP DoS 공격 중, 저대역폭으로 수행되는 RUDY 공격과 고대역폭으로 수행되는 HULK 공격을 제안된 모델로 탐지한 결과, 두 공격 탐지에 있어서 애플리케이션 메트릭 기반 머신러닝 모델의 F1-Score가 네트워크 메트릭 기반의 모델보다 각각 약 0.3, 0.1 높은 것을 확인하였다.

• 한국시뮬레이션학회논문지
• /
• 제23권2호
• /
• pp.73-84
• /
• 2014

최근 IT 기술을 전통 산업 등 다양한 분야에 적용하려는 연구가 많이 진행되고 있다. 교육 분야에서는 기술을 이용하여 학습을 개선시키는 방법에 대한 관심이 높아지고 있으며, 교육에 IT 기술을 접목하는 자기주도적 학습에 대한 연구가 활발하게 진행되고 있다. 자동차 정비 훈련 분야에서도 이-러닝 형태의 교육이 이루어지고 있지만, 자기주도형 학습을 하기에는 어려움이 많다. 기존에 만들어진 자동차 정비 훈련 프로그램은 개발자에 의해 정해진 훈련 시나리오에 따라 미리 정해진 동작만을 차례로 수행하게 되어있다. 그러나 훈련자가 자기주도적으로 정비 훈련을 수행하기 위해서는 훈련자 스스로 여러가지 정비 동작을 수행하고 다양한 상황을 경험할 수 있어야 한다. 그러한 기능을 제공하기 위해서는 훈련자의 다양한 동작에 대해서 실제와 같은 결과가 나오도록 프로그램이 개발되어야 하지만, 그러려면 자동차의 여러가지 복잡한 전기적, 기계적 동작에 관한 매우 복잡한 계산이 수행되어야 한다. 본 논문에서는 자동차 정비 훈련에서 복잡한 물리적 시뮬레이션 없이 훈련자의 다양한 동작에 따라 자동차의 동작을 시뮬레이션 함으로써 자기주도적인 학습을 할 수 있는 JESS 추론 엔진을 이용한 논리 시뮬레이션 에이전트를 구현한다.

• 한국철도학회:학술대회논문집
• /
• 한국철도학회 2011년도 정기총회 및 추계학술대회 논문집
• /
• pp.239-244
• /
• 2011

Technical advancement in the field of world-class railway has brought the development of passenger safety, efficient operation and fast passenger service. The study of the railway speed-up has been progressed for a long time. As for the speed-up of trunkline railways, its design work has already promoted in some railway tacks and some track sections has already been under construction. In such a perspective we should review if the domestic signaling system is the optimal system and figure out the trend of signaling system for the speed-up that is recently applied over the world. In this way learning about justification and futurition, we should secure the national competitiveness in the long term and review the system available to international exports. Further, it is important to apply the result to the research project currently being pursued. In addition, predicting the recent international trends, we should show the direction of the future-oriented and economic signaling system.

• 한국실내디자인학회논문집
• /
• 제18권4호
• /
• pp.105-115
• /
• 2009

As database become more variable and multimedia, the integration between conventional library based on printed matter and digital library on IT makes it accelerated that the changes in today's information circumstances. Due to more leisure time modern people who get intellectual and interested in variety so they need continuing education, intellectual satisfaction and cultural life. Consequently today's public library expands to become a source of every kind of information, an educational facility and community facility to fulfill the cultural needs. On the assumption above, this study aims at offering database for planning public library that meets the needs for multifunctional library through analyzing space, structure, and character of modern public libraries and chooses 8 cases in Japan provide both primary and additional functions. Quantitative data from blueprints, websites and publications has been collected to analyze the space arrangement corresponding with the functional characteristics.

• 공학교육연구
• /
• 제17권4호
• /
• pp.48-53
• /
• 2014

Recently, the tendency for young people losing interest in science has accelerated. This tendency is remarkable for not only Japan but also some countries of Asia. It is thought that this tendency was brought from the decrease of the chance to watch the actual manufacturing activities, and the decrease of the real experience which children makes something in their childhood. In order to bring up the capable engineers on such a social background, making a product practice which promotes the understanding of the application of theory for manufacturing products is important in addition to study by the text book. In this study, some practical work materials for the lower grades in the college were developed. Integration of the developed materials and the course of study becomes an effective teaching method of the of subjects on a special field. Arranging this practical work on the learning process makes a high effect of the integrated practical work on the curriculum.

• 로봇학회논문지
• /
• 제17권2호
• /
• pp.133-141
• /
• 2022

This paper introduces model compression algorithms which make a deep neural network smaller and faster for embedded systems. The model compression algorithms can be largely categorized into pruning, quantization and knowledge distillation. In this study, gradual pruning, quantization aware training, and knowledge distillation which learns the activation boundary in the hidden layer of the teacher neural network are integrated. As a large deep neural network is compressed and accelerated by these algorithms, embedded computing boards can run the deep neural network much faster with less memory usage while preserving the reasonable accuracy. To evaluate the performance of the compressed neural networks, we evaluate the size, latency and accuracy of the deep neural network, DenseNet201, for image classification with CIFAR-10 dataset on the NVIDIA Jetson Xavier.

• Journal of Information Processing Systems
• /
• 제18권4호
• /
• pp.500-509
• /
• 2022

With the rapid development of educational informatization, teaching methods become diversified characteristics, but a large number of information data restrict the evaluation on teaching subject and object in terms of the effect of English education. Therefore, this study adopts the concept of incremental learning and eigenvalue interval algorithm to improve the weighted decision tree, and builds an English education effect evaluation model based on association rules. According to the results, the average accuracy of information classification of the improved decision tree algorithm is 96.18%, the classification error rate can be as low as 0.02%, and the anti-fitting performance is good. The classification error rate between the improved decision tree algorithm and the original decision tree does not exceed 1%. The proposed educational evaluation method can effectively provide early warning of academic situation analysis, and improve the teachers' professional skills in an accelerated manner and perfect the education system.

• ETRI Journal
• /
• 제46권1호
• /
• pp.71-81
• /
• 2024

Recent advances in deep learning for speech and visual recognition have accelerated the development of multimodal speech recognition, yielding many innovative results. We introduce a Korean audiovisual speech recognition corpus. This dataset comprises approximately 150 h of manually transcribed and annotated audiovisual data supplemented with additional 2000 h of untranscribed videos collected from YouTube under the Creative Commons License. The dataset is intended to be freely accessible for unrestricted research purposes. Along with the corpus, we propose an open-source framework for automatic speech recognition (ASR) and audiovisual speech recognition (AVSR). We validate the effectiveness of the corpus with evaluations using state-of-the-art ASR and AVSR techniques, capitalizing on both pretrained models and fine-tuning processes. After fine-tuning, ASR and AVSR achieve character error rates of 11.1% and 18.9%, respectively. This error difference highlights the need for improvement in AVSR techniques. We expect that our corpus will be an instrumental resource to support improvements in AVSR.