• 한국어정보학회:학술대회논문집
• /
• 2016.10a
• /
• pp.165-170
• /
• 2016

본 논문에서 제안하는 문맥의존 철자오류 교정은 통계 정보를 이용한 방법으로 통계적 언어처리에서 가장 널리 쓰이는 샤논(Shannon)이 발표한 노이지 채널 모형(noisy channel model)을 기반으로 한다. 선행연구에서 부족하였던 부분의 성능 향상을 위해 교정대상단어의 오류생성 및 통계 데이터의 저장 방식을 개선하여 Default 연산을 적용한 모델을 제안한다. 선행 연구의 모델은 교정대상단어의 오류생성 시 편집거리의 제약을 1로 하여 교정 실험을 하지만 제안한 모델은 같은 환경에서 더욱 높은 검출과 정확도를 보였으며, 오류단어의 편집거리(edit distance) 제약을 넓게 적용하더라도 신뢰도가 있는 검출과 교정을 보였다.

• Annual Conference on Human and Language Technology
• /
• 2016.10a
• /
• pp.165-170
• /
• 2016

본 논문에서 제안하는 문맥의존 철자오류 교정은 통계 정보를 이용한 방법으로 통계적 언어처리에서 가장 널리 쓰이는 샤논(Shannon)이 발표한 노이지 채널 모형(noisy channel model)을 기반으로 한다. 선행연구에서 부족하였던 부분의 성능 향상을 위해 교정대상단어의 오류생성 및 통계 데이터의 저장 방식을 개선하여 Default 연산을 적용한 모델을 제안한다. 선행 연구의 모델은 교정대상단어의 오류생성 시 편집거리의 제약을 1로 하여 교정 실험을 하지만 제안한 모델은 같은 환경에서 더욱 높은 검출과 정확도를 보였으며, 오류단어의 편집거리(edit distance) 제약을 넓게 적용하더라도 신뢰도가 있는 검출과 교정을 보였다.

• IEMEK Journal of Embedded Systems and Applications
• /
• v.16 no.5
• /
• pp.201-207
• /
• 2021

Edge Computing enables image-based inference in close proximity to end users and real-world objects. However, since edge servers have limited computational and accelerator resources, efficient resource management is essential. In this paper, we present HeteroAccel system that performs optimal scheduling in Kubernetes platform based on available node and accelerator information for various inference requests. Our experiments showed 25.3% improvement in overall inference performance over the default scheduling scheme in edge computing environment in which four types of inference services are requested.

• Journal of KIISE:Computing Practices and Letters
• /
• v.16 no.4
• /
• pp.507-511
• /
• 2010

In this paper, we propose to enhance the performance of software transcoder by using GPGPU for scaling filters. Video transcoding is a technique that translates a video file to another video file that has a different coding algorithm and/or a different frame size. Its demand increases as more multimedia devices with different specification coexist in our daily life. Since transcoding is computationally intensive, a software transcoder that runs on a CPU takes long processing time. In this paper, we achieve significant speed-up by parallelizing the scaling filter using a GPGPU that can provide significantly large computation power. Through extensive experiments with various video scripts of different size and with various scaling filter options, it is verified that the enhanced transcoder could achieve 36% performance improvement in the default option, and up to 101% in a certain option.

• Journal of Broadcast Engineering
• /
• v.17 no.4
• /
• pp.625-633
• /
• 2012

Extended Depth of Focus (EDF) algorithms for extracting three-dimensional (3D) information from a set of optical image slices are studied by many researches recently. Due to the limited depth of focus of the microscope, only a small portion of the image slices are in focus. Most of the EDF algorithms try to find the in-focus area to generate a single focused image and a 3D depth image. Inherent to most image processing algorithms, the EDF algorithms need parameters to be properly initialized to perform successfully. In this paper, we select three popular transform-based EDF algorithms which are each based on pyramid, wavelet transform, and complex wavelet transform, and study the performance of the algorithms according to the initialization of its parameters. The parameters we considered consist of the number of levels used in the transform, the selection of the lowest level image, the window size used in high frequency filter, the noise reduction method, etc. Through extended simulation, we find a good relationship between the initialization of the parameters and the properties of both the texture and 3D ground truth images. Typically, we find that a proper initialization of the parameters improve the algorithm performance 3dB ~ 19dB over a default initialization in recovering the 3D information.