• ETRI Journal
• /
• 제39권1호
• /
• pp.21-29
• /
• 2017

This paper addresses the problem of unsupervised speech separation based on robust non-negative matrix factorization (RNMF) with ${\beta}$-divergence, when neither speech nor noise training data is available beforehand. We propose a robust version of non-negative matrix factorization, inspired by the recently developed sparse and low-rank decomposition, in which the data matrix is decomposed into the sum of a low-rank matrix and a sparse matrix. Efficient multiplicative update rules to minimize the ${\beta}$-divergence-based cost function are derived. A convolutional extension of the proposed algorithm is also proposed, which considers the time dependency of the non-negative noise bases. Experimental speech separation results show that the proposed convolutional RNMF successfully separates the repeating time-varying spectral structures from the magnitude spectrum of the mixture, and does so without any prior training.

• Communications for Statistical Applications and Methods
• /
• 제20권3호
• /
• pp.235-240
• /
• 2013

Generalized linear mixed models(GLMMs) are frequently used for the analysis of longitudinal categorical data when the subject-specific effects is of interest. In GLMMs, the structure of the random effects covariance matrix is important for the estimation of fixed effects and to explain subject and time variations. The estimation of the matrix is not simple because of the high dimension and the positive definiteness; subsequently, we practically use the simple structure of the covariance matrix such as AR(1). However, this strong assumption can result in biased estimates of the fixed effects. In this paper, we introduce Bayesian modeling approaches for the random effects covariance matrix using a modified Cholesky decomposition. The modified Cholesky decomposition approach has been used to explain a heterogenous random effects covariance matrix and the subsequent estimated covariance matrix will be positive definite. We analyze metabolic syndrome data from a Korean Genomic Epidemiology Study using these methods.

• ETRI Journal
• /
• 제24권6호
• /
• pp.469-472
• /
• 2002

Speaker adaptation techniques are generally used to reduce speaker differences in speech recognition. In this work, we focus on the features fitted to a linear regression-based speaker adaptation. These are obtained by feature transformation based on independent component analysis (ICA), and the feature transformation matrices are estimated from the training data and adaptation data. Since the adaptation data is not sufficient to reliably estimate the ICA-based feature transformation matrix, it is necessary to adjust the ICA-based feature transformation matrix estimated from a new speaker utterance. To cope with this problem, we propose a smoothing method through a linear interpolation between the speaker-independent (SI) feature transformation matrix and the speaker-dependent (SD) feature transformation matrix. From our experiments, we observed that the proposed method is more effective in the mismatched case. In the mismatched case, the adaptation performance is improved because the smoothed feature transformation matrix makes speaker adaptation using noisy speech more robust.

• Communications for Statistical Applications and Methods
• /
• 제27권2호
• /
• pp.201-210
• /
• 2020

Baseline-category logit random effects models have been used to analyze longitudinal nominal data. The models account for subject-specific variations using random effects. However, the random effects covariance matrix in the models needs to explain subject-specific variations as well as serial correlations for nominal outcomes. In order to satisfy them, the covariance matrix must be heterogeneous and high-dimensional. However, it is difficult to estimate the random effects covariance matrix due to its high dimensionality and positive-definiteness. In this paper, we exploit the modified Cholesky decomposition to estimate the high-dimensional heterogeneous random effects covariance matrix. Bayesian methodology is proposed to estimate parameters of interest. The proposed methods are illustrated with real data from the McKinney Homeless Research Project.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제13권11호
• /
• pp.5616-5630
• /
• 2019

Fully homomorphic encryption allows a third-party to perform arbitrary computation over encrypted data and is especially suitable for secure outsourced computation. This paper investigates secure outsourced computation of multiple matrix multiplication based on fully homomorphic encryption. Our work significantly improves the latest Mishra et al.'s work. We improve Mishra et al.'s matrix encoding method by introducing a column-order matrix encoding method which requires smaller parameter. This enables us to develop a binary multiplication method for multiple matrix multiplication, which multiplies pairwise two adjacent matrices in the tree structure instead of Mishra et al.'s sequential matrix multiplication from left to right. The binary multiplication method results in a logarithmic-depth circuit, thus is much more efficient than the sequential matrix multiplication method with linear-depth circuit. Experimental results show that for the product of ten 32×32 (64×64) square matrices our method takes only several thousand seconds while Mishra et al.'s method will take about tens of thousands of years which is astonishingly impractical. In addition, we further generalize our result from square matrix to non-square matrix. Experimental results show that the binary multiplication method and the classical dynamic programming method have a similar performance for ten non-square matrices multiplication.

• 융합보안논문지
• /
• 제8권4호
• /
• pp.161-166
• /
• 2008

본 논문은 3차원 모델 표면의 특징 곡률(Feature Curvature) 정보를 이용하여 3차원 거리정보 데이터(Range Image)를 자동으로 정합하는 효율적인 방법을 제안하고 그 성능을 분석하였다. 제안한 알고리즘은 3차원 데이터에 대한 거리정보의 물리적 특성인 가우스 곡률(Gaussian Curvature)을 이용하여 모델의 특징점을 검출하고, 공분산 행렬(Covariance Matrix)을 이용하여 각 데이터의 지역좌표계(Local Coordinate System) 사이의 변위를 계산한다. 3차원 형상 취득장치의 카메라 위치는 3차원 데이터와 투영된 2차원 영상과의 사영행렬(Projection Matrix) 관계식으로 계산한다. 결론부분에서는 실험결과를 기존 연구방법과 비교하여 제안된 방법이 더 빠르고 정확하게 정합하는 결과를 보임으로써 3차원 물체인식이나 모델링에 응용성을 제시하였다.

• 대한임베디드공학회논문지
• /
• 제12권6호
• /
• pp.377-384
• /
• 2017

The scope of the matrix application is so broad that it can not be limited. A typical matrix application area in computer science is image processing. Particularly, radar scanning equipment implemented on a small embedded system requires real-time matrix transposition for image processing, and since its memory size is small, a general matrix transposition algorithm can not be applied. In this case, matrix transposition must be done in disk space, such as flash disk, using a limited memory buffer. In this paper, we analyze and improve a recently published flash disk-based matrix transposition algorithm named as asymmetric sub-matrix transposition algorithm. The performance analysis shows that the asymmetric sub-matrix transposition algorithm has lower performance than the conventional sub-matrix transposition algorithm, but the improved asymmetric sub-matrix transposition algorithm is superior to the sub-matrix transposition algorithm in 13 of the 16 experimental data.

• 대한전자공학회:학술대회논문집
• /
• 대한전자공학회 2002년도 하계종합학술대회 논문집(4)
• /
• pp.227-230
• /
• 2002

In this paper, we propose an algorithm to decode Data Matrix two dimensional barcode symbology. We employ hough transform and bilinear image warping to extract the barcode region from the image scanned using a CMOS digital camera. The location of barcode can be found by applying Hough transform. However, barcode image should be warped due to the nonlinearity of lens and the viewing angle of camera. In this paper, bilinear warping transform is adopted to wa게 and align the barcode region of the scanned image. Codeword can be detected from the aligned barcode region.

• 신재생에너지
• /
• 제16권1호
• /
• pp.15-24
• /
• 2020

For mechanical load testing of wind turbines, capture matrix is constructed for various range of wind speeds according to the international standard IEC 61400-13. The conventional method wastes considerable amount of data by its invalid data policy -segment data into 10 minutes then remove invalid ones. Previously, we have suggested an alternative way to save the total amount of data to build a capture matrix, but the efficient selection of data has been still under question. The paper introduces optimization algorithms to construct capture matrix with less data. Heuristic algorithm (simple stacking and lowest frequency first), population method (particle swarm optimization) and Q-Learning accompanied with epsilon-greedy exploration are compared. All algorithms show better performance than the conventional way, where the distribution of enhancement was quite diverse. Among the algorithms, the best performance was achieved by heuristic method (lowest frequency first), and similarly by particle swarm optimization: Approximately 28% of data reduction in average and more than 40% in maximum. On the other hand, unexpectedly, the worst performance was achieved by Q-Learning, which was a promising candidate at the beginning. This study is helpful for not only wind turbine evaluation particularly the viewpoint of cost, but also understanding nature of wind speed data.

• 대한전기학회:학술대회논문집
• /
• 대한전기학회 2007년도 심포지엄 논문집 정보 및 제어부문
• /
• pp.365-366
• /
• 2007

본 워터마크를 삽입 할 수 있는 임베디드 시스템을 구현 하였다. 워터마크 삽입을 위해 DWT와 Data Matrix가 사용되었다. DWT(Discrete Wavelet Transform)는 주파수 공간에서 워터마크를 삽입하기 위해 사용되었고, Data Matrix는 워터마크로 사용되었다. 데이터 매트릭스는 미국의 Data Matrix사가 만든 이차원 바코드로 오류검출 및 복원 알고리즘을 가지고 있어 작은 에러는 복원이 가능하다. 시스템으로는 PDA를 사용하였고, 틀로는 EVC를 사용하였다. 삽입 알고리즘은 다음과 같다. DWT를 한 경우 4개의 서브밴드로 나누어지며, 그 중 cV(horizontal detail)와 cH(vertical detail)를 선택하여 4*4블록 단위로 나눈다. 나누어진 블록과 대응하는 워터마크의 픽셀 값에 의해 계수에 일정 값(가중치)을 더하거나 때주어 워터마크를 삽입한다. 추출 알고리즘은 역으로 이루어진다. 성능평가는 PDA에서 워터마크 삽입 알고리즘을 통하여 워터마크를 삽입, 추출된 영상을 가지고 Matlap을 이용하여 평가하였다.