• 방송공학회논문지
• /
• 제25권2호
• /
• pp.157-165
• /
• 2020

본 논문에서는 임베디드 시스템에서의 양자화 기계학습을 수행할 경우 발생하는 양자화 오차를 효과적으로 보상하기 위한 방법론을 제안한다. 경사 도함수(Gradient)를 사용하는 기계학습이나 비선형 신호처리 알고리즘에서 양자화 오차는 경사 도함수의 조기 소산(Early Vanishing Gradient)을 야기하여 전체적인 알고리즘의 성능 하락을 가져온다. 이를 보상하기 위하여 경사 도함수의 최대 성분에 대하여 직교하는 방향의 보상 탐색 벡터를 유도하여 양자화 오차로 인한 성능 하락을 보상하도록 한다. 또한, 기존의 고정 학습률 대신, 내부 순환(Inner Loop) 없는 비선형 최적화 알고리즘에 기반한 적응형 학습률 결정 알고리즘을 제안한다. 실험 결과 제안한 방식의 알고리즘을 로젠블록 함수를 통한 비선형 최적화 문제에 적용할 시 양자화 오차로 인한 성능 하락을 최소화시킬 수 있음을 확인하였다.

• International Journal of Control, Automation, and Systems
• /
• 제3권4호
• /
• pp.552-563
• /
• 2005

In this paper, we propose a wavelet based fuzzy neural network (WFNN) based direct adaptive control scheme for the solution of the tracking problem of mobile robots. To design a controller, we present a WFNN structure that merges the advantages of the neural network, fuzzy model and wavelet transform. The basic idea of our WFNN structure is to realize the process of fuzzy reasoning of the wavelet fuzzy system by the structure of a neural network and to make the parameters of fuzzy reasoning be expressed by the connection weights of a neural network. In our control system, the control signals are directly obtained to minimize the difference between the reference track and the pose of a mobile robot via the gradient descent (GD) method. In addition, an approach that uses adaptive learning rates for training of the WFNN controller is driven via a Lyapunov stability analysis to guarantee fast convergence, that is, learning rates are adaptively determined to rapidly minimize the state errors of a mobile robot. Finally, to evaluate the performance of the proposed direct adaptive control system using the WFNN controller, we compare the control results of the WFNN controller with those of the FNN, the WNN and the WFM controllers.

• 한국정보통신학회논문지
• /
• 제10권1호
• /
• pp.88-95
• /
• 2006

PID 제어기는 구조가 간단하고 적용이 용이하다는 장점으로 인하여 널리 사용되고 있는 제어방식이다. 이러한 선형 PID 제어기는 시스템의 파라메터가 변화가 있거나 부하 특성이 비 선형적으로 변화할 때에 적절한 이득과 성능을 얻기 어려워 고성능 제어 특성을 기대하기 어렵다. 본 연구에서는 세포성 면역 반응과 경사감소학습에 기초하여 비선형 PID 제어기를 설계하고, 설계된 제어기의이득과 비선형 함수의 파라메터들을 실시간 적응적으로 학습할 수 있는 학습 알고리즘을 개발하고, 이를 제어시스템에 적용하였다. 제안된 비선형 PID 제어기는 비선형 직류 모터 시스템의 파라메터들이 변화하거나 주파수가 다른 추종 명령에 대하여, 적응적으로 이득을 변화 시키며 추종함을 보였다.

• 제어로봇시스템학회:학술대회논문집
• /
• 제어로봇시스템학회 2004년도 ICCAS
• /
• pp.640-645
• /
• 2004

This paper proposes a direct adaptive control method using self-recurrent wavelet neural network (SRWNN) for stable path tracking of mobile robots. The architecture of the SRWNN is a modified model of the wavelet neural network (WNN). Unlike the WNN, since a mother wavelet layer of the SRWNN is composed of self-feedback neurons, the SRWNN has the ability to store the past information of the wavelet. For this ability of the SRWNN, the SRWNN is used as a controller with simpler structure than the WNN in our on-line control process. The gradient-descent method with adaptive learning rates (ALR) is applied to train the parameters of the SRWNN. The ALR are derived from discrete Lyapunov stability theorem, which are used to guarantee the stable path tracking of mobile robots. Finally, through computer simulations, we demonstrate the effectiveness and stability of the proposed controller.

• Structural Engineering and Mechanics
• /
• 제45권1호
• /
• pp.53-67
• /
• 2013

This study aimed to develop a model to accurately predict the acceleration of structural systems during an earthquake. The acceleration and applied force of a structure were measured at current time step and the velocity and displacement were estimated through linear integration. These data were used as input to predict the structural acceleration at next time step. The computation tool used was the Volterra/Wiener neural network (VWNN) which contained the mathematical model to predict the acceleration. For alleviating problems of relatively large-dimensional and nonlinear systems, the VWNN model was utilized as the signal processing tool, including the Taylor series components in the input nodes of the neural network. The number of the intermediate layer nodes in the neural network model, containing the training and simulation stage, was evaluated and optimized. Discussions on the influences of the gradient descent with adaptive learning rate algorithm and the Levenberg-Marquardt algorithm, both for determining the network weights, on prediction errors were provided. During the simulation stage, different earthquake excitations were tested with the optimized settings acquired from the training stage to find out which of the algorithms would result in the smallest error, to determine a proper simulation model.

• Journal of information and communication convergence engineering
• /
• 제13권2호
• /
• pp.123-131
• /
• 2015

Training neural networks is a complex task with great importance in the field of supervised learning. In the training process, a set of input-output patterns is repeated to an artificial neural network (ANN). From those patterns weights of all the interconnections between neurons are adjusted until the specified input yields the desired output. In this paper, a new hybrid algorithm is proposed for global optimization of connection weights in an ANN. Dynamic swarms are shown to converge rapidly during the initial stages of a global search, but around the global optimum, the search process becomes very slow. In contrast, the gradient descent method can achieve faster convergence speed around the global optimum, and at the same time, the convergence accuracy can be relatively high. Therefore, the proposed hybrid algorithm combines the dynamic particle swarm optimization (DPSO) algorithm with the backpropagation (BP) algorithm, also referred to as the DPSO-BP algorithm, to train the weights of an ANN. In this paper, we intend to show the superiority (time performance and quality of solution) of the proposed hybrid algorithm (DPSO-BP) over other more standard algorithms in neural network training. The algorithms are compared using two different datasets, and the results are simulated.

• 한국전자파학회논문지
• /
• 제29권1호
• /
• pp.68-76
• /
• 2018

표적 우선순위 할당은 다수의 표적이 존재하는 전술 환경에서 다기능 레이다(Multifunction Radar: MFR)가 중요한 표적을 추적하고 레이다 자원을 효율적으로 관리하기 위해 필요한 기능이다. 본 논문에서는 레이다에서 수집한 정보로부터 표적에 대한 우선순위를 산출하는 인공 신경망(Artificial Neural Network: ANN) 모델을 구현한다. 더 나아가, 기존의 경사 하강법(gradient descent) 기반 역전파(backpropagation) 알고리즘을 발전시켜 표적 우선순위 할당에 더욱 적합한 최급 강하법(steepest descent) 기반 신경망 학습 알고리즘을 제안한다. 시뮬레이션에서는 훈련 데이터와 신경망의 결과값 사이의 오차와 특정 테스트 시나리오에서 할당된 우선순위의 합리성을 분석하여 제안된 방법의 성능을 확인한다.

• 한국인터넷방송통신학회논문지
• /
• 제21권1호
• /
• pp.87-92
• /
• 2021

빅데이터 기반의 추천시스템 모델링에서 바이어스, 분산, 오류 및 학습은 성능에 중요한 요소이다. 이러한 시스템에서는 추천 모델이 설명도를 유지하면서 복잡도를 줄여야 한다. 또한 데이터의 희소성과 시스템의 예측은 서로 반비례의 속성을 가지기 마련이다. 따라서 희소성의 데이터를 인수분해 방법을 활용하여 상품간의 유사성을 학습을 통한 상품추천모델이 제안되어 왔다. 본 논문에서는 이 모델의 손실함수에 대한 최적화 방안으로 max-norm 규제를 적용하여 모델의 일반화 능력을 향상시키고자 한다. 해결방안은 기울기를 투영하는 확률적 투영 기울기 강하법을 적용하는 것이다. 많은 실험을 통하여 데이터가 희박해질수록 기존의 방법에 비해 제안된 규제 방법이 상대적으로 효과가 있다는 것을 확인하였다.

• International Journal of Advanced Culture Technology
• /
• 제5권2호
• /
• pp.82-89
• /
• 2017

The range of problems that can be handled by the activation of big data and the development of hardware has been rapidly expanded and machine learning such as deep learning has become a very versatile technology. In this paper, mnist data set is used as experimental data, and the Cross Entropy function is used as a loss model for evaluating the efficiency of machine learning, and the value of the loss function in the steepest descent method is We applied the GradientDescentOptimize algorithm to minimize and updated weight and bias via backpropagation. In this way we analyze optimal reliability value corresponding to the number of exercises and optimal reliability value without overfitting. And comparing the overfitting time according to the number of data changes based on the number of training times, when the training frequency was 1110 times, we obtained the result of 92%, which is the optimal reliability value without overfitting.

• 한국인공지능학회지
• /
• 제5권1호
• /
• pp.29-37
• /
• 2017

The range of problems that can be handled by the activation of big data and the development of hardware has been rapidly expanded and machine learning such as deep learning has become a very versatile technology. In this paper, mnist data set is used as experimental data, and the Cross Entropy function is used as a loss model for evaluating the efficiency of machine learning, and the value of the loss function in the steepest descent method is We applied the Gradient Descent Optimize algorithm to minimize and updated weight and bias via backpropagation. In this way we analyze optimal reliability value corresponding to the number of exercises and optimal reliability value without overfitting. And comparing the overfitting time according to the number of data changes based on the number of training times, when the training frequency was 1110 times, we obtained the result of 92%, which is the optimal reliability value without overfitting.