• 전자통신동향분석
• /
• 제38권5호
• /
• pp.12-22
• /
• 2023

The development of neural networks is evolving towards the adoption of transformer structures with attention modules. Hence, active research focused on extending the concept of lightweight neural network algorithms and hardware acceleration is being conducted for the transition from conventional convolutional neural networks to transformer-based networks. We present a survey of state-of-the-art research on lightweight neural network algorithms and hardware architectures to reduce memory usage and accelerate both inference and training. To describe the corresponding trends, we review recent studies on token pruning, quantization, and architecture tuning for the vision transformer. In addition, we present a hardware architecture that incorporates lightweight algorithms into artificial intelligence processors to accelerate processing.

• 한국표면공학회:학술대회논문집
• /
• 한국표면공학회 2017년도 춘계학술대회 논문집
• /
• pp.126-126
• /
• 2017

Using first principles calculations we unveil fundamental mechanism of hydrolysis reactions of two hazardous chemicals $PCl_3$ and $POCl_3$ with molecular water clusters nearby. It is found that the water molecules play a key role as a catalyst significantly lowing the activation barriers by transferring its protons to the reaction intermediates. Interestingly, torsional angles of molecular complexes at transition states are identified as a vital descriptor on the reaction rate. Analysis of charge distribution over the complexes further reinforces the finding with atomic level correlation between the torsional angle and variation of the orbital hybridization state of P in the complex. Electronic charge separation (or polarization) enhances thermodynamic stability of the activated complex at transition state and reduces the activation energy through hydrogen bonding network with water molecules nearby. Calculated potential energy surfaces (PES) for the hydrolysis reactions of $PCl_3$ and $POCl_3$ depict their two contrastingly different profiles of double- and triple-deep wells, respectively. It is ascribed to the unique double-bonding O=P in the $POCl_3$. Our results on the activation free energy show well agreements with previous experimental data within $7kcalmol^{-1}$ deviation.

• 한국정보과학회논문지:시스템및이론
• /
• 제33권9호
• /
• pp.642-649
• /
• 2006

지능형 서비스를 제공하는 환경에서 상황을 인식하기 위하여 임베디드 시스템이 갖는 제한적인 리소스와 컴퓨팅 파워 환경에 적합한 상태 전이 기반 상황 표현 모델을 제안하고, 이를 인식하는 상황 인식기와 제어 신호를 발생시키는 동작 발생기를 결합한 임베디드 시스템에 탑재하기에 적합한 상황-동작 변환기(SAC)를 설계하였다. 또한 ARM 프로세서 기반의 임베디드 보드에 설계된 상황-동작 변환기를 응용한 상황 관리기를 구현하여 이를 스마트 스케줄러 서비스에 활용하였다.

• 정보보호학회논문지
• /
• 제20권2호
• /
• pp.23-32
• /
• 2010

IEEE 802.16e 표준을 기반으로 하는 와이브로(WiBro) 서비스는 순수 국내 기술로 개발된 초고속 무선 휴대인터넷 기술이다. 본 논문에서는 와이브로 초기인증 단계에서 생길 수 있는 보안취약점을 분석하고 그러한 보안위협을 대상으로 한 공격을 탐지할 수 있는 명세기반의 침입탐지시스템을 제안한다. 제안된 침입탐지시스템은 PKMv2 EAP-AKA 기반의 정상적인 초기 인증 동작방식을 명세화하여 상태전이머신으로 모델링한 후 명세에 기반한 침입탐지를 실시한다. 본 논문에서는 초기인증 과정에서 발생 가능한 다섯 가지의 공격들을 시나리오로 모델링하고, 시나리오 기반의 실험을 실시하여 제안된 침입탐지시스템의 탐지성능을 검증한다.

• 한국세라믹학회지
• /
• 제31권9호
• /
• pp.1044-1052
• /
• 1994

Tape casting technique was successfully applied to produce porous ceramics and multi-layered ceramics containing porous layers, where spherical hollow polymer particles were introduced as pore precursors. In the presence of extreme differences in density and size between Al2O3 and pore precursor particles, hindered settling was effective in preventing segregation of component particles and packing behavior of mixed powders was improved through bimodal packing. There were two transitions in packing behavior of mixed powders. The first transition took place at 40~50 vol% pore precursor addition, where majority of pores changed from close to open pore state. The other transition occured at 60~70 vol% pore precursor addition, where pore precursor particles formed a continuous network structure.

• International Journal of Control, Automation, and Systems
• /
• 제6권1호
• /
• pp.109-118
• /
• 2008

We present two estimators for discrete non-Gaussian and nonstationary probability density estimation based on a dynamic Bayesian network (DBN). The first estimator is for off line computation and consists of a DBN whose transition distribution is represented in terms of kernel functions. The estimator parameters are the weights and shifts of the kernel functions. The parameters are determined through a recursive learning algorithm using maximum likelihood (ML) estimation. The second estimator is a DBN whose parameters form the transition probabilities. We use an asymptotically convergent, recursive, on-line algorithm to update the parameters using observation data. The DBN calculates the state probabilities using the estimated parameters. We provide examples that demonstrate the usefulness and simplicity of the two proposed estimators.

• 한국안전학회지
• /
• 제11권3호
• /
• pp.104-111
• /
• 1996

A system was developed that could predict hazardous states for safe operation of chemical plants. This system aimed to prevent hazards of chemical processes from misoperation before real operation. In this study, a data base was organized which consisted of all hazards in the chemical processes. The structure of process was represented by signed direct graph(SDG) of NODEs and ARCs. Each NODE and ARC have property variables ; connected structure and state of processes etc. The hazards that could be occurred in processes were divided into two classes ; one is by operation of unit and the other is by hazardous materials. Using Hazardous States Transition Network, we could recognize transition progress of process states.

• The Journal of the Acoustical Society of Korea
• /
• 제13권1E호
• /
• pp.55-63
• /
• 1994

다중센서 시스템의 문제점, 특히 임의의 방향에 대한 목표물의 유무를 판단하는 문제에 접근하기 위하여 결절이론을 소개한다. 이 문제에 대한 해는 대상 방향에 대한 간단한 수 즉, 0 또는 1로 구성된 수의 집합으로 표현된다. 이렇게 코드로 표현된 수는 Hopfield 신경회로망에 의한 최적화 기법으로 변환되며, 이러한 변환은 음원의 방향탐지에 대한 이해를 쉽게 한다. Hopfield 신경회로망 모델의 기존 상태 개선 방법들을 사용할 때 직명하게 되는 난점들을 소개하며, 그것과 관련된 문제점들을 제기한다. 이를 해결하기 위하여 새로운 착상, 즉 0에서 1의 상태로 변할 때 보다 큰 에너지 차이를 보이는 뉴런이 상태 개선의 우위를 점할 수 있다는 접을 제안한다. 이로부터 이전 연구들과는 다른 새로운 상태 개선 방법 및 새로운 관점을 도출하게 된다. 본 논문에서는 컴퓨터 모의 실험을 통해 제안된 방법의 전망과 음원 방향 탐지의 효율성을 보인다.

• 한국시뮬레이션학회논문지
• /
• 제3권1호
• /
• pp.151-165
• /
• 1994

Broadband network based on the Asynchronous Transfer Mode(ATM) concept are becoming the target technology for the emerging Broadband Integrated Services Digital Network(B-ISDN). Since B-ISDN is very complex and requites a great amount of investment, optimum design and performance analysis of such systems are very important. Simulation can be widely used to analyze and examine the broadband network behavior. However, for the complicated system like broadband networks it is extremely difficult and time-consuming to develop a complete model for simulation. In this paper, an object-oriented modeling approach for the broadband network simulation is presented for the effective and efficient modeling. Object-oriented approaches can provide a good structuring capability for complicated simulation models and facilitate the development of reusable and extensible simulation models. We have developed an object-oriented model which consists of object model and behavior model. In the object mode., the components of the broadband network and both constant bit rate(CBR) and variable bit rate(VBR) traffic types of call level, burst level, and cell level are modeled as object classes. In the behavior model, the dynamic features for each object class are represented using the state transition diagram. It has been shown by illustration that objectoriented modeling is an effective tool for modeling the complicated B-ISDN.

• 한국정보과학회논문지:시스템및이론
• /
• 제26권12호
• /
• pp.1468-1475
• /
• 1999

컴퓨터망 보안을 위해서는 취약점의 빠른 발견과 그에 따르는 신속한 조치가 요구된다. 이를 위해 컴퓨터망 여러 곳에서 사용자가 원하는 보안 정보를 수집, 분석하고 수집된 보안 정보들을 효과적으로 보여주는 통합적인 보안 관리를 제공하는 모니터링 방식이 필요하다. 본 논문에서는 보안 관리를 위한 동적 모니터링 형식 모델을 제시한다. 동적 모니터링은 상태 전이에 의한 시스템 상태, 사용자와의 상호작용, 시각 및 모니터링 객체의 동적 활성화를 이용하여 컴퓨터망 통합 보안 관리를 제공한다. 계속적으로 변하는 보안 정보, 컴퓨터망 관리자의 관심점과 모니터링 레벨의 변화를 즉각적으로 반영함으로써 관리자에게 컴퓨터망 관리의 정확성, 효율성 그리고 편의성을 제공한다. 모니터링 시스템을 형식적으로 모델링하여 여러 모니터링 응용프로그램의 개발이 가능한 모니터링 플랫폼으로 발전시킬 수 있다. Abstract It is highly required to quickly detect the vulnerability of a computer network system and an appropriate action toward it should be followed as soon as possible for its security. It leads us to the need of a monitoring schema that can provide an integrated security management with carefully selected and analysed data through the computer network for its users. This paper presents a formal model of dynamic monitoring for security management. It provides the comprehensive security management using continuously changing security information, user interactions and dynamic activation of visual and monitoring objects. It enables system security officers to manage computer systems accurately, efficiently and conveniently by reflecting the state transition and the transformation of concerns and a monitoring level of system security officers immediately. This model can be used as the basis of a monitoring platform.