• 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.

• The Journal of the Acoustical Society of Korea
• /
• 제22권3E호
• /
• pp.124-132
• /
• 2003

We propose a new method for speech/non-speech classifiers based on concepts of the adaptive boosting (AdaBoost) algorithm in order to detect speech for robust speech recognition. The method uses a combination of simple base classifiers through the AdaBoost algorithm and a set of optimized speech features combined with spectral subtraction. The key benefits of this method are the simple implementation, low computational complexity and the avoidance of the over-fitting problem. We checked the validity of the method by comparing its performance with the speech/non-speech classifier used in a standard voice activity detector. For speech recognition purpose, additional performance improvements were achieved by the adoption of new features including speech band energies and MFCC-based spectral distortion. For the same false alarm rate, the method reduced 20-50% of miss errors.

• 통합자연과학논문집
• /
• 제12권4호
• /
• pp.105-115
• /
• 2019

In this paper, a prediction system is proposed to control the brightness of smart street lamps by predicting the moving path through the reduction of consumption power and information of pedestrian's past moving direction while meeting the function of existing smart street lamps. The brightness of smart street lamps is adjusted by utilizing the walk tracking vector and soft hand-off characteristics obtained through the motion sensing sensor of smart street lamps. In addition, the motion vector is used to analyze and predict the pedestrian path, and the GPU is used for high-speed computation. Pedestrians were detected using adaptive Gaussian mixing, weighted difference imaging, and motion vectors, and motions of pedestrians were analyzed using the extracted motion vectors. The preprocessing process using linear interpolation is performed to improve the performance of the proposed prediction system. Fuzzy prediction system and neural network prediction system are designed in parallel to improve efficiency and rough set is used for error correction.

• 방송공학회논문지
• /
• 제26권5호
• /
• pp.652-655
• /
• 2021

Versatile Video Coding(VVC)의 압축 효율을 끌어올리기 위하여 다양한 화면 간 예측(inter prediction)기법 중 적응적 움직임 벡터 해상도(Adaptive motion vector resolution, 이하 AMVR)기술이 채택되어 왔다. 다만, AMVR을 적용하여 최적의 해상도를 결정하기 위해서는 매 부호화 유닛마다 다양한 테스트를 진행해야 하며, 이는 율-왜곡 비용의 계산 복잡도 증가를 야기한다. 따라서 VVC의 부호화 복잡도의 감소를 위해 효과적으로 최적의 AMVR 모드를 찾아야 한다. 본 논문에서는 보다 다양한 데이터셋 기반 하에 경량화된 신경망 기반의 AMVR 결정 알고리즘을 제안한다.

• 전자공학회논문지B
• /
• 제33B권1호
• /
• pp.161-169
• /
• 1996

본 논문에서는 동영상 압축의 핵심 기술인 움직임 벡터 추정에 있어서 신경 회로망을 이용한 벡터 양자화에 의해 탐색 영역을 예측하는 방법을 제안한다. 훈련영상을 입력으로 하여 전역 탐색법 등에 의하여 구해진 움직임 벡터를 이용하여 움직임 벡터 코드 북을 생성하고 이를 예측 탐색 점으로 이용한다. 움직임 벡터 코드 북을 생성하기 위해서 병렬 처리 특성과 다양한 학습 알고리즘을 갖는 신경 회로망을 이용하였다. 제안된 방법은 움직임 벡터들의 높은 공간적 상관성을 이용하게 되고 결과적으로 적은 탐색 점으로 움직임 벡터를 추정할 수 있으므로 계산량을 줄일 수 있을 뿐 아니라 움직임 벡터를 표현하기 위해 소요되는 비트 수도 크게 줄일 수 있다. 모의 실험을 통하여 제안된 방식이 기존의 고속 블록 매칭 알고리즘보다 우수함을 보였다.

• Advances in Computational Design
• /
• 제1권3호
• /
• pp.235-251
• /
• 2016

We have performed a design optimization of a stiffened panel with curvilinear stiffeners using an artificial neural network (ANN) residual kriging based surrogate modeling approach. The ANN residual kriging based surrogate modeling involves two steps. In the first step, we approximate the objective function using ANN. In the next step we use kriging to model the residue. We optimize the panel in an iterative way. Each iteration involves two steps-shape optimization and size optimization. For both shape and size optimization, we use ANN residual kriging based surrogate model. At each optimization step, we do an initial sampling and fit an ANN residual kriging model for the objective function. Then we keep updating this surrogate model using an adaptive sampling algorithm until the minimum value of the objective function converges. The comparison of the design obtained using our optimization scheme with that obtained using a traditional genetic algorithm (GA) based optimization scheme shows satisfactory agreement. However, with this surrogate model based approach we reach optimum design with less computation effort as compared to the GA based approach which does not use any surrogate model.

• 정보처리학회논문지B
• /
• 제10B권4호
• /
• pp.397-402
• /
• 2003

본 연구에서는 뉴우턴법의 고정점 알고리즘에 적응 조정이 가능한 학습 파라미터를 이용한 효율적인 신경망 기반 독립성분분석기법을 제안하였다. 이는 엔트로피 최적화 함수의 1차 미분을 이용하는 뉴우턴법의 고정점 알고리즘에서 학습율과 모멘트를 역혼합행렬의 경신 상태에 따나 적응조정되도록 함으로써 분리속도와 분리성능을 개선시키기 위함이다 제안된 기법을 256$\times$256 픽셀의 8개 지문과 512$\times$512 픽셀의 10개 영상으로부터 임의의 혼합행렬에 따라 발생되는 지문과 영상의 분리에 적용한 결과, 기존의 고정점 알고리즘에 의한 결과보다 우수한 분리성능과 빠른 분리속도가 있음을 확인하였다. 특히 제안된 알고리즘은 문제의 규모가 클수록 분리성능과 분리속도의 개선 정도가 큼을 확인하였다.

• 한국통신학회논문지
• /
• 제28권10C호
• /
• pp.1007-1012
• /
• 2003

Multilayer perceptrons (MLPs)를 위한 일반적인 BP 알고리즘의 학습 속도를 개선하기 위하여 제약을 갖는 최적화 기술을 제안하고 이를 backpropagation (BP) 알고리즘에 적용한다. 먼저 잡음 제약을 갖는 LMS (noise constrained least mean square : NCLMS) 알고리즘과 영잡음 제약 LMS (ZNCLMS) 알고리즘을 BP 알고리즘에 적용한다. 이러한 알고리즘들은 다음과 같은 가정을 반드시 필요로 하여 알고리즘의 이용에 많은 제약을 갖는다. NCLMS 알고리즘을 이용한 NCBP 알고리즘은 정확한 잡음 전력을 알고 있다고 가정한다. 또한 ZNCLMS 알고리즘을 이용한 ZNCBP 알고리즘은 잡음의 전력을 0으로 가정, 즉 잡음을 무시하고 학습을 진행한다. 본 논문에서는 확장된(augmented) Lagrangian multiplier를 이용하여, 비용함수(cost function)를 변형한다. 이를 통하여 잡음에 대한 가정을 제거하고 ZNCBP와 NCBP 알고리즘을 확장, 일반화하여 적응 오류 제약 BP(adaptive error constrained BP : AECBP) 알고리즘을 유도, 제안한다. 제안한 알고리즘들의 수렴 속도는 일반적인 BP 알고리즘보다 약 30배정도 빠른 학습 속도를 나타내었으며, 일반적인 선형 필터와 거의 같은 수렴속도를 나타내었다.

• 한국통신학회논문지
• /
• 제21권11호
• /
• pp.2739-2749
• /
• 1996

This aper introduces the multiple Neocognitron module approach for the effective target reognition. The Neocognitron which is designed to classify a pattern by extracting the local features from it, seems to be an unique method that can perform a pattern recognition using the neural networks. But due to its rigid structure, the Neocognitron must be reconstructed whenever there exists a variation on the number of classes. This is a quite difficult problem for the target recognition application that needs huge amount of computation and numerous classes to be classified. In this paper, we construct several smaller Necognitrom modules and train each module to adapt each class. After construction of the mulules, we integrate them in parallel so as to adaopt input at the same time and to produce each score that shold be matched to be learned class. This approach can reduce the sizes of the networks and is adaptive to the increase of classes as well as the authentic distortion, shift, scale variation and slight rotation invariant properties of general Neocognitron. This paper show the effectiveness of the proposed approach through some experience and performs analysis of the inhibitory interconnections in the architecture of the multiple module structure.

• Journal of Positioning, Navigation, and Timing
• /
• 제11권3호
• /
• pp.163-172
• /
• 2022

Typical anti-jamming technologies based on array antennas, Space Time Adaptive Process (STAP) & Space Frequency Adaptive Process (SFAP), are very effective algorithms to perform nulling and beamforming. However, it does not perform equally well for all types of jamming signals. If the anti-jamming algorithm is not optimized for each signal type, anti-jamming performance deteriorates and the operation stability of the system become worse by unnecessary computation. Therefore, jamming classification technique is required to obtain optimal anti-jamming performance. Machine learning, which has recently been in the spotlight, can be considered to classify jamming signal. In general, performing supervised learning for classification requires a huge amount of data and new learning for unfamiliar signal. In the case of jamming signal classification, it is difficult to obtain large amount of data because outdoor jamming signal reception environment is difficult to configure and the signal type of attacker is unknown. Therefore, this paper proposes few-shot jamming signal classification technique using meta-learning and transfer-learning to train the model using a small amount of data. A training dataset is constructed by anti-jamming algorithm input data within the GNSS receiver when jamming signals are applied. For meta-learning, Model-Agnostic Meta-Learning (MAML) algorithm with a general Convolution Neural Networks (CNN) model is used, and the same CNN model is used for transfer-learning. They are trained through episodic training using training datasets on developed our Python-based simulator. The results show both algorithms can be trained with less data and immediately respond to new signal types. Also, the performances of two algorithms are compared to determine which algorithm is more suitable for classifying jamming signals.