• Journal of the Korean Institute of Intelligent Systems
• /
• v.21 no.1
• /
• pp.12-18
• /
• 2011

In this paper, we proposed Interval Type-2 polynomial Radial Basis Function Neural Networks. In the receptive filed of hidden layer, Interval Type-2 fuzzy set is used. The characteristic of Interval Type-2 fuzzy set has Footprint Of Uncertainly(FOU), which denotes a certain level of robustness in the presence of un-known information when compared with the type-1 fuzzy set. In order to improve the performance of proposed model, we used the linear polynomial function as connection weight of network. The parameters such as center values of receptive field, constant deviation, and connection weight between hidden layer and output layer are optimized by Conjugate Gradient Method(CGM) and Space Search Evolutionary Algorithm(SSEA). The proposed model is applied to gas furnace dataset and its result are compared with those reported in the previous studies.

• The KIPS Transactions:PartD
• /
• v.14D no.2
• /
• pp.169-180
• /
• 2007

This paper discusses a new unsupervised XML document clustering technique based on the function transform and FFT(Fast Fourier Transform). An XML document is transformed into a discrete function based on the hierarchical nesting structure of the elements. The discrete function is, then, transformed into vectors using FFT. The vectors of two documents are compared using a weighted Euclidean distance metric. If the comparison is lower than the pre specified threshold, the two documents are considered similar in the structure and are grouped into the same cluster. XML clustering can be useful for the storage and searching of XML documents. The experiments were conducted with 800 synthetic documents and also with 520 real documents. The experiments showed that the function transform and FFT are effective for the incremental and unsupervised clustering of XML documents similar in structure.

• Proceedings of the KIEE Conference
• /
• 2011.07a
• /
• pp.1956-1957
• /
• 2011

본 연구는 White LED를 이용하여 주변 밝기 변화에 빠르게 적응하는 퍼지 뉴로 Dimming Control System을 설계한다. 본 논문에서는 방사형기저함수 신경회로망(Radial Basis Function Neural Network: RBFNN)을 설계하여 실제 White LED Dimming Control System에 적용시켜 모델의 근사화 및 일반화 성능을 평가한다. 제안한 모델에서의 은닉층은 방사형기저함수를 사용하여 적합도를 구현하였고, 후반부의 연결가중치는 경사하강법을 사용한다. 이때 멤버쉽 함수의 중심점은 HCM 클러스터링 (Hard C-Means Clustering)을 적용하여 결정한다. 연결가중치는 4가지 형태의 다항식을 대입하여 출력을 평가하였다. 최종 출력의 최적화를 위하여 PSO(Particle Swarm Optimization)을 이용하여 은닉층 노드수 및 다항식 형태를 결정한다. 본 논문에서 제안한 LED Dimming Control System은 Atmega8535를 사용하여 PWM 제어 방식을 사용하고, 조도계(Cds)를 이용하여 LED의 밝기에 따른 주변의 밝기를 감지하여 조명에 적응시키는 방법을 적용하였다.

• Proceedings of the KIEE Conference
• /
• 2008.10b
• /
• pp.247-248
• /
• 2008

본 논문에서는 TYPE-2 퍼지 추론 기반의 RBF 뉴럴 네트워크(TYPE-2 Radial Basis Function Neural Network, T2RBFNN)를 설계하고 PSO(Particle Swarm Optimization) 알고리즘을 이용하여 모델의 파라미터를 동정한다. 제안된 모델의 은닉층은 TYPE-2 가우시안 활성 함수로 구성되며, 출력층은 Interval set 형태의 연결가중치를 갖는다. 여기에서 규칙 전반부 활성함수의 중심 선택은 C-means 클러스터링 알고리즘을 이용하고, 규칙 후반부 Interval set 형태의 연결가중치 결정에는 경사 하강법(Gradient descent method)을 이용한 오류 역전파 알고리즘을 사용하여 학습한다. 또한, 최적의 모델을 설계하기 위한 학습율 및 활성함수의 활성화 영역 결정에는 입자 군집 최적화(PSO; Particle Swarm Optimization) 알고리즘으로 동조한다. 마지막으로, 제안된 모델의 평가를 위하여 모의 데이터 집합(Synthetic dadaset)을 적용하고 근사화 및 일반화 능력에 대하여 토의한다.

• Proceedings of the Korea Contents Association Conference
• /
• 2005.11a
• /
• pp.364-370
• /
• 2005

LBL(Layer-By-Layer) algorithms have been proposed to accelerate the training speed of MLPs(Multilayer Perceptrons). In this LBL algorithms, each layer needs a error function for optimization. Especially, error function for hidden layer has a great effect to achieve good performance. In this sense, this paper proposes a new hidden layer error function for improving the performance of LBL algorithm for MLPs. The hidden layer error function is derived from the mean squared error of output layer. Effectiveness of the proposed error function was demonstrated for a handwritten digit recognition and an isolated-word recognition tasks and very fast learning convergence was obtained.

• Journal of IKEEE
• /
• v.20 no.2
• /
• pp.136-142
• /
• 2016

This paper presents a non-local means (NLM) denoising algorithm based on a new weighting function using a mixed norm. The fidelity of the difference between an anchor patch and the reference patch in the NLM denoising depends on noise level and local activity. This paper introduces a new weighting function based on a mixed norm type of which the order is determined by noise level and local activity of an anchor patch, so that the performance of the NLM denoising can be enhanced. Experimental results demonstrate the objective and subjective capability of the proposed algorithm. In addition, it was verified that the proposed algorithm can be used to improve the performance of the other $l_2$ norm based non-local means denoising algorithms

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.46 no.1
• /
• pp.71-78
• /
• 2009

Sound source localization systems in service robot applications estimate the direction of a human voice. Time delay information obtained from a few separate microphones is widely used for the estimation of the sound direction. Correlation is computed in order to calculate the time delay between two signals. In addition, PHAT weighting function can be applied to significantly improve the accuracy of the estimation. However, FFT and IFFT operations in the PHAT weighting function occupy more than half of the area of the sound source localization system. Thus efficient FFT and IFFT designs are essential for the IP implementation of sound source localization system. In this paper, we propose an efficient FFT/IFFT design method based on the characteristics of human voice.

• Journal of the Institute of Convergence Signal Processing
• /
• v.10 no.1
• /
• pp.49-59
• /
• 2009

In this paper a watermarking algorithm is proposed to optimize the capacity of the digital watermark insertion in an experimental threshold using the characteristics of human visual system(HVS), adaptive scale factors, and weight functions based on discrete wavelet transform. After the original image is decomposed by a 3-level discrete wavelet transform, the watermarks for capacity optimization are inserted into all subbands except the baseband, by applying the important coefficients from the experimental threshold in the wavelet region. The adaptive scale factors and weight functions based on HVS are considered for the capacity optimization of the digital watermark insertion in order to enhance the robustness and invisibility. The watermarks are consisted of gaussian random sequences and detected by correlation. The experimental results showed that this algorithm can preserve a fine image quality against various attacks such as the JPEG lossy compression, noise addition, cropping, blurring, sharpening, linear and non-linear filtering, etc.

• KIPS Transactions on Software and Data Engineering
• /
• v.8 no.3
• /
• pp.123-128
• /
• 2019

This paper proposes the stock price prediction based on the artificial intelligence, where the model with recurrent convolution neural network (RCNN) layers is adopted. In the motivation of this prediction, long short-term memory model (LSTM)-based neural network can make the output of the time series prediction. On the other hand, the convolution neural network provides the data filtering, averaging, and augmentation. By combining the advantages mentioned above, the proposed technique predicts the estimated stock price of next day. In addition, in order to emphasize the recent time series, a custom weighted loss function is adopted. Moreover, stock data related to the stock price index are adopted to consider the market trends. In the experiments, the proposed stock price prediction reduces the test error by 3.19%, which is over other techniques by about 19%.

• The Journal of the Acoustical Society of Korea
• /
• v.25 no.8
• /
• pp.370-374
• /
• 2006

Mel-frequency cepstral coefficients are widely used as the feature for speech recognition. In FMCC extraction process. the spectrum. obtained by Fourier transform of input speech signal is divided by met-frequency bands, and each band energy is extracted for the each frequency band. The coefficients are extracted by the discrete cosine transform of the obtained band energy. In this Paper. we calculate the output energy for each bandpass filter by taking the weighting function when applying met-frequency scaled bandpass filter. The weighting function is Gaussian distributed function whose center is at the formant frequency In the experiments, we can see the comparative performance with the standard MFCC in clean condition. and the better Performance in worse condition by the method proposed here.