• 대한수학회지
• /
• 제61권2호
• /
• pp.395-408
• /
• 2024

There have been numerous studies on the characteristics of the solutions of ordinary differential equations for optimization methods, including gradient descent methods and alternating direction methods of multipliers. To investigate computer simulation of ODE solutions, we need to trace pseudo-orbits by real orbits and it is called shadowing property in dynamics. In this paper, we demonstrate that the flow induced by the alternating direction methods of multipliers (ADMM) for a C² strongly convex objective function has the eventual shadowing property. For the converse, we partially answer that convexity with the eventual shadowing property guarantees a unique minimizer. In contrast, we show that the flow generated by a second-order ODE, which is related to the accelerated version of ADMM, does not have the eventual shadowing property.

• 대한전자공학회:학술대회논문집
• /
• 대한전자공학회 2002년도 ITC-CSCC -3
• /
• pp.1626-1629
• /
• 2002

This paper describes novel methods to construct fuzzy inference rules with gradient descent. The present methods have a constructive mechanism of the rule unit that is applicable in two parameters: the central value and the width of the membership function in the antecedent part. The first approach is to create the rule unit at the nearest position from the input space, for the central value of the membership function in the antecedent part. The second is to create the rule unit which has the minimum width, for the width of the membership function in the antecedent part. Experimental results are presented in order to show that the proposed methods are effective in difference on the inference error and the number of learning iterations.

• The Journal of the Acoustical Society of Korea
• /
• 제16권1E호
• /
• pp.64-74
• /
• 1997

In many decision making systems, rule-based approaches are used to solve complex problems in the areas of pattern analysis and computer vision. In this paper, we present methods for generating fuzzy IF-THEN rules automatically from training data for pattern classification and high-level computer vision. The rules are generated by construction minimal approximate fuzzy aggregation networks and then training the networks using gradient descent methods. The training data that represent features are treated as linguistic variables that appear in the antecedent clauses of the rules. Methods to generate the corresponding linguistic labels(values) and their membership functions are presented. In addition, an inference procedure is employed to deduce conclusions from information presented to our rule-base. Two experimental results involving synthetic and real are given.

• 전기전자학회논문지
• /
• 제23권4호
• /
• pp.1302-1308
• /
• 2019

본 논문은 CNN (Convolutional Neural Network)와 2D Lidar 센서에서 획득한 위치 데이터를 이용하여 이미지를 분류하는 방법을 제시한다. Lidar 센서는 데이터 정확도, 형상 왜곡 및 광 변화에 대한 강인성 측면에서의 이점으로 인해 무인 장치에 널리 사용되어 왔다. CNN 알고리즘은 하나 이상의 컨볼루션 및 풀링 레이어로 구성되며 이미지 분류에 만족스러운 성능을 보여 왔다. 본 논문에서는 학습 방법에 따라 다른 유형의 CNN 아키텍처들인 Gradient Descent (GD) 및 Levenberg-arquardt (LM)를 구현하였다. LM 방법에는 학습 파라메터를 업데이트하는 요소 중 하나인 Hessian 행렬 근사 빈도에 따라 두 가지 유형이 있다. LM 알고리즘의 시뮬레이션 결과는 GD 알고리즘보다 이미지 데이터의 분류 성능이 우수하였다. 또한 Hessian 행렬 근사가 더 빈번한 LM 알고리즘은 다른 유형의 LM 알고리즘보다 작은 오류를 보여주었다.

• 정보관리학회지
• /
• 제38권2호
• /
• pp.113-127
• /
• 2021

본 연구는 대학 도서관 사용 증진을 위하여 개인별 맞춤 도서 추천시스템을 구축하는 것을 목적으로 한다. 특히 사용자의 아이템에 대한 선호도가 존재하는 다수의 추천시스템과는 달리, 선호도가 존재하지 않을 때에 도서 추천이 가능하도록 하는 방안인 도서관 이용자의 도서 대출 목록과 성향을 활용하여 평가지표를 생성하는 방법을 제안하고자 한다. 이용자가 아직 읽지 않은 책에 대한 예상 선호도를 산출하는 방식으로 도서를 추천하는 행렬 분해 방법인 Singular Value Decomposition(SVD)과 Stochastic Gradient Descent(SGD) 알고리즘을 활용한 모델을 구축했다. 더불어 유사도가 높은 이용자 그룹 내의 도서 대출 목록을 참조하여 추천하는 사용자 기반 협업 필터링 알고리즘을 활용해 모델을 구현했다. 최종적으로 평가지표를 활용한 세 가지 모델에 대하여 사용자 평가를 진행했다. 각각의 모델이 제시한 개인별 맞춤 도서 다섯 권의 목록을 해당 대출자에게 제공하고, 추천 도서에 대한 만족/불만족 여부를 이진화 점수화하여 모델에 대한 평가를 진행했다.

• 제어로봇시스템학회논문지
• /
• 제2권1호
• /
• pp.21-25
• /
• 1996

In this paper we suggest a general purpose RNN training algorithm which is derived on the optimal control concepts and variational methods. First, learning is regared as an optimal control problem, then using the variational methods we obtain optimal weights which are given by a two-point boundary-value problem. Finally, the modified gradient descent algorithm is applied to RNN for on-line training. This algorithm is intended to be used on learning complex dynamic mappings between time varing I/O data. It is useful for nonlinear control, identification, and signal processing application of RNN because its storage requirement is not high and on-line learning is possible. Simulation results for a nonlinear plant identification are illustrated.

• 유체기계공업학회:학술대회논문집
• /
• 유체기계공업학회 1999년도 유체기계 연구개발 발표회 논문집
• /
• pp.221-226
• /
• 1999

Three-dimensional flow analysis and numerical optimization methods are presented for the design of an axial-flow fan. Steady, Incompressible, three-dimensional Reynolds-averaged Wavier-Stokes equations are used as governing equations, and standard k-$\epsilon$ turbulence model is chosen as a turbulence model. Governing equations are discretized using finite volume method. Steepest descent method, conjugate gradient method and BFGS method are compared to determine the searching directions. Golden section method and quadratic fit-sectioning method are tested for one dimensional search. Objective function is defined as a ratio of generation rate of the turbulent kinetic energy to pressure head. Sweep angle distributions are used as design variables.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제11권6호
• /
• pp.3029-3045
• /
• 2017

The main processes of a cognitive radio system include spectrum sensing, spectrum decision, spectrum sharing, and spectrum conversion. Experimental results show that these stages introduce a time delay that affects the spectrum sensing accuracy, reducing its efficiency. To reduce the time delay, the frequency spectrum prediction was proposed to alleviate the burden on the spectrum sensing. In this paper, the deep recurrent neural network (DRNN) was proposed to predict the spectrum of multiple time slots, since the existing methods only predict the spectrum of one time slot. The continuous state of a channel is divided into a many time slots, forming a time series of the channel state. Since there are more hidden layers in the DRNN than in the RNN, the DRNN has fading memory in its bottom layer as well as in the past input. In addition, the extended Kalman filter was used to train the DRNN, which overcomes the problem of slow convergence and the vanishing gradient of the gradient descent method. The spectrum prediction based on the DRNN was verified with a WiFi signal, and the error of the prediction was analyzed. The simulation results proved that the multiple slot spectrum prediction improved the spectrum efficiency and reduced the energy consumption of spectrum sensing.

• 스마트미디어저널
• /
• 제3권2호
• /
• pp.9-13
• /
• 2014

본 논문은 형광 X선 분석 시스템에서 관찰되는 스펙트럼에서 원하는 원소의 피크값을 검출하는데 쓰이는 마쿼트 알고리즘을 모바일 환경에서 더욱 효과적으로 사용하는 데에 있다. 이러한 마쿼트 알고리즘은 본래 잡음이 섞이기 전의 순수한 데이터가 무엇인지 알아가기 위한 유추해 가는 과정의 방법이다. 이러한 마쿼트 알고리즘에서 매우 중요한 역할을 하는 매개변수가 있는데 이 매개변수에 따라서 구하고자 하는 변수 값을 더욱 빠르게 구할 수도 있고 아닐 수도 있다. 기존의 방법에서 불필요한 계산량을 줄이기 위하여 매우 중요한 역할을 하는 매개변수인 ${\mu}$ 자리에 이 매개변수 대신 다른 매개변수를 도입한다. 또한 하드웨어적 측면을 고려시, 여러개의 정규분포의 모양으로 되어있는 함수를 여러개의 정규분포로 나누어서 생각하면 원하는 값을 구하기 더욱 간단해지지만 신뢰도 문제가 발생할 수 있다. 이러한 문제를 해결할 새로운 시스템을 제시한다.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제18권2호
• /
• pp.456-477
• /
• 2024

With information technology's rapid development, the Internet faces serious security problems. Studies have shown that malware has become a primary means of attacking the Internet. Therefore, adversarial samples have become a vital breakthrough point for studying malware. By studying adversarial samples, we can gain insights into the behavior and characteristics of malware, evaluate the performance of existing detectors in the face of deceptive samples, and help to discover vulnerabilities and improve detection methods for better performance. However, existing adversarial sample generation methods still need help regarding escape effectiveness and mobility. For instance, researchers have attempted to incorporate perturbation methods like Fast Gradient Sign Method (FGSM), Projected Gradient Descent (PGD), and others into adversarial samples to obfuscate detectors. However, these methods are only effective in specific environments and yield limited evasion effectiveness. To solve the above problems, this paper proposes a malware adversarial sample generation method (PixGAN) based on the pixel attention mechanism, which aims to improve adversarial samples' escape effect and mobility. The method transforms malware into grey-scale images and introduces the pixel attention mechanism in the Deep Convolution Generative Adversarial Networks (DCGAN) model to weigh the critical pixels in the grey-scale map, which improves the modeling ability of the generator and discriminator, thus enhancing the escape effect and mobility of the adversarial samples. The escape rate (ASR) is used as an evaluation index of the quality of the adversarial samples. The experimental results show that the adversarial samples generated by PixGAN achieve escape rates of 97%, 94%, 35%, 39%, and 43% on the Random Forest (RF), Support Vector Machine (SVM), Convolutional Neural Network (CNN), Convolutional Neural Network and Recurrent Neural Network (CNN_RNN), and Convolutional Neural Network and Long Short Term Memory (CNN_LSTM) algorithmic detectors, respectively.