• Annual Conference of KIPS
• /
• 2024.10a
• /
• pp.32-33
• /
• 2024

온디바이스 환경에서 딥러닝 모델 최적화는 필수적이지만, 제한된 자원으로 고성능 모델을 직접 적용하는 데에는 한계가 있다. 본 논문에서는 이를 극복하기 위한 주요 기법인 가지치기, 양자화, 지식 증류, 신경망 아키텍처 탐색 및 이들의 결합 기법을 소개하고 분석한다. 각 기법의 정의와 특징, 적용 사례를 통해 성능 향상과 자원 효율성을 극대화하는 방법을 제시하며, 이를 바탕으로 최근 연구 동향을 소개한다.

• Annual Conference of KIPS
• /
• 2024.10a
• /
• pp.34-35
• /
• 2024

본 논문에서는 데이터 집약적 연산의 효율성 향상을 위한 새로운 프로세서 아키텍처 접근법인 Near-Data Processing(NDP) 기술의 최근 동향을 살펴본다. AI 모델 학습과 빅데이터 분석 등에서 NDP 기술의 적용 사례를 분석하고, Processing-In-Memory(PIM)와 In-Storage Processing(ISP)를 중심으로 한 최신 연구를 소개한다. 또한, 실제 하드웨어 구현 사례로 Xilinx FPGA 개발보드의 HBM 통합 사례를 다룬다.

• Annual Conference of KIPS
• /
• 2024.10a
• /
• pp.863-864
• /
• 2024

유동 인구가 많고 다양한 서비스를 필요로 하는 병원에서는 여러 기능을 통합적으로 수행할 수 있는 토탈 서비스 로봇이 필요하다. 자율주행, 얼굴인식, 음성 처리기술 등을 적용한 AI 자율주행 토탈 서비스 로봇을 도입함으로써 의료진의 업무 부담을 줄이고, 병원의 운영 효율성을 높이며, 전반적인 서비스의 질을 향상할 수 있다.

• Annual Conference of KIPS
• /
• 2024.10a
• /
• pp.143-144
• /
• 2024

네트워크 트래픽이 최근 급격히 증가하면서, 전통적인 CPU 기반의 패킷 처리 방식은 성능의 한계에 직면하고 있다. 이러한 문제를 해결하기 위해 SmartNIC은 네트워크 처리를 CPU로부터 오프로드하여 시스템 전체의 성능을 향상시키는 데 중요한 역할을 한다. 그리하여 네트워크 서비스를 SmartNIC로 오프로드 하여 CPU 점유율을 줄이고 시스템의 효율성을 높이는 데 큰 이점을 제공한다. 본 논문에서는 NVIDIA의 DOCA와 AMD의 SSDK를 소개하며, SmartNIC에서 구현된 다양한 보안 서비스들을 살펴보고자 한다.

• Journal of the Korea Institute of Building Construction
• /
• v.15 no.1
• /
• pp.9-16
• /
• 2015

The goal of this research was evaluating and suggesting the solution of preventing carbonation of concrete replaced high-volume of slag. The concrete mixtures were prepared with high-volume slag and recycled aggregate, and the concrete samples were evaluated the carbonation depth with various surface treatment methods. For various surface treatment methods and surface protecting sheets, bonding strength and carbonation depth were measured. Basically, from the results, the carbonation of concrete was completely prevented with any type of surface treatment method and surface protecting sheet as far as the surface treatment materials were remained. Therefore, in this research, it was known and suggested that the easiness of handling and sufficient bonding performance was much important than the quality of surface protecting sheets.

• Korean Chemical Engineering Research
• /
• v.54 no.6
• /
• pp.775-780
• /
• 2016

Recently, photovoltaic industry needs a new design of Czochralski (Cz) process for higher productivity with reasonable energy consumption as well as solar cell's efficiency. If the process uses the large size reactor for increasing productivity, it is possible to produce a 12-inch, rather than the 8-inch. Also the continuous czochralski process method can be maximized to increase productivity. In this study, it was designed to improve the yield value of ingot with optimal condition which reduce consumption of electrical power. It has increased the productivity of the 12-inch ingot process condition by using CFD simulation. I have found optimal growth rate, by comparing each growth rate the interface shape, Temperature gradient, power consumption. As a result, the optimal process parameters of the growth furnace has been derived to improve for the productivity and to reduce energy. This study will contribute to the improvement of the productivity in the solar cell industry.

• Information Systems Review
• /
• v.24 no.4
• /
• pp.41-54
• /
• 2022

Despite the recent rapid advancement of science and technology, we have been experiencing the decline in productivity since the 2000s. This study aims to investigate the decline at both industry and firm levels, by looking at the emergence and growth of large firms such as Amazon, Alphabet, and Apple and M&A trends. Following the results of previous studies, our results show that productivity at industry level has decreased since the 2000s. Particularly, in the period after 2011, the deterioration of allocative efficiency due to the large firms and the decline in the growth rate of surviving firms in the industry with low ratio of large firms contributed to the productivity decline. On the other hand, our analysis at firm level demonstrates that the productivity of firms that acquired IT firms improved over the entire period. While M&As have a positive impact on productivity, M&As with a demand-side motive such as market penetration and expansion of channels have a relatively larger impact than the ones for production or operation efficiency. Our results also suggest that the higher the proportion of large firms in a specific industry, the lower the productivity of individual firms in the same industry. Overall, given that the industry's structural changes for digital transformation tends to strengthen the growth of large firms, our findings have significant implications by empirically identifying the relationships of the emergence of large firms, the acquisition of IT/Non-IT firms, and motivations for M&As to firm/industry productivity.

• The KIPS Transactions:PartB
• /
• v.17B no.6
• /
• pp.459-466
• /
• 2010

Natural language processing has several ambiguity problems, and English-Korean machine translation especially includes those problems to be solved in each translation step. This paper focuses on resolving part-of-speech ambiguity of English words in order to improve the efficiency of English analysis, which is in part of efforts for developing practical English-Korean machine translation system. In order to improve the efficiency of the English analysis, the part-of-speech determination must be fast and accurate for being integrated with machine translation system. This paper proposes the probabilistic models for part-of-speech determination. We use Penn Treebank corpus in building the probabilistic models. In experiment, we present the performance of the part-of-speech determination models and the efficiency improvement of the machine translation system by the proposed part-of-speech determination method.

• Journal of Korean Society of Environmental Engineers
• /
• v.29 no.1
• /
• pp.98-105
• /
• 2007

The objectives of this study were to demonstrate that enhanced anaerobic digestion could be achieved by adopting ultrasonic cavitation pretreatment on the basis of the biogas production and to compare the simulation results of ADM1 (Anaerobic Digestion Model No. 1) with results of the experimentally operated digester the ultrasonic pretreatment of sewage sludge showed the hindered effect on the dewaterbility and the increase of SCOD production. In this study, four sets of lab-scale anaerobic digester were operated with untreated(control), 30 min, 60 min and 90 min ultrasonic pretreated sludge. TCOD removal efficiencies in digesters of control, 30, 60, 90 min sonicated sludge were 31.9%, 37.9, 38.5% and42.2%, and 75 removal were 15.9%, 20.8%, 21.5%, 24.1% respectively. Also more biogas was produced gradually with the increased sludge loading and the pretreatment time. Overall the simulation results had a correspondence tendency with the experimental efficiencies.

• KOREAN JOURNAL OF CROP SCIENCE
• /
• v.41 no.3
• /
• pp.354-361
• /
• 1996

Near Infrared Reflectance Spectroscopy(NIRS) is accepted as today's most versatile method for rapid chemical analysis. The technique offers rapid multicomponent analysis. This study was conducted to improve the efficiency of quality analysis in covered barley grain, and to search for the effects of chemical components variation in covered barley grain on NIRS. Among the three groups with different range in the contents, each equation for starch contents increased standard error of prediction(SEP) and increased correlation coefficient from 0.872 to 0.883. According as, $\beta$-glucan and protein contents decreased SEP and increased correlation coefficient by expanded chemical components variation. Effective equation for ash contents analysis was obtained from group 3. Among the covered barley chemical components, starch and ash contents were required to conduct futher studies in term of accuracy and variation of contents. It was concluded that NIRS method would be applicable for the rapid determination of $\beta$-glucan and protein contents in covered barley grains.