• ETRI Journal
• /
• v.41 no.2
• /
• pp.235-241
• /
• 2019

In this paper, we propose a rank-weighted reconstruction feature to improve the robustness of a feed-forward deep neural network (FFDNN)-based acoustic model. In the FFDNN-based acoustic model, an input feature is constructed by vectorizing a submatrix that is created by slicing the feature vectors of frames within a context window. In this type of feature construction, the appropriate context window size is important because it determines the amount of trivial or discriminative information, such as redundancy, or temporal context of the input features. However, we ascertained whether a single parameter is sufficiently able to control the quantity of information. Therefore, we investigated the input feature construction from the perspectives of rank and nullity, and proposed a rank-weighted reconstruction feature herein, that allows for the retention of speech information components and the reduction in trivial components. The proposed method was evaluated in the TIMIT phone recognition and Wall Street Journal (WSJ) domains. The proposed method reduced the phone error rate of the TIMIT domain from 18.4% to 18.0%, and the word error rate of the WSJ domain from 4.70% to 4.43%.

• Proceedings of the Acoustical Society of Korea Conference
• /
• 1994.06a
• /
• pp.1009-1014
• /
• 1994

This paper improves the previously proposed spectral mapping method for supervised speaker adaptation in which a mapped spectrum is interpolated from speaker difference vectors at typical spectra based on a minimized distortion criterion. In estimating these difference vectors, it is important to find an appropriate number of typical points. The previous method empirically adjusts the number of typical points, while the present method optimizes the effective number by rank reduction of normal equation. This algorithm was applied to a supervised speaker adaptation for Korean word recognition using the templates form a prototype male speaker. The result showed that the rank reduction technique not only can automatically determine an optimal number of code vectors, but also slightly improves the recognition scores compared with those obtained by the previous method.

• The Korean Journal of Pain
• /
• v.30 no.2
• /
• pp.86-92
• /
• 2017

Osteoblasts, originating from mesenchymal cells, make the receptor activator of the nuclear factor kappa B ligand (RANKL) and osteoprotegerin (OPG) in order to control differentiation of activated osteoclasts, originating from hematopoietic stem cells. When the RANKL binds to the RANK of the pre-osteoclasts or mature osteoclasts, bone resorption increases. On the contrary, when OPG binds to the RANK, bone resorption decreases. Denosumab (AMG 162), like OPG (a decoy receptor), binds to the RANKL, and reduces binding between the RANK and the RANKL resulting in inhibition of osteoclastogenesis and reduction of bone resorption. Bisphosphonates (BPs), which bind to the bone mineral and occupy the site of resorption performed by activated osteoclasts, are still the drugs of choice to prevent and treat osteoporosis. The merits of denosumab are reversibility targeting the RANKL, lack of adverse gastrointestinal events, improved adherence due to convenient biannual subcutaneous administration, and potential use with impaired renal function. The known adverse reactions are musculoskeletal pain, increased infections with adverse dermatologic reactions, osteonecrosis of the jaw, hypersensitivity reaction, and hypocalcemia. Treatment with 60 mg of denosumab reduces the bone resorption marker, serum type 1 C-telopeptide, by 3 days, with maximum reduction occurring by 1 month. The mean time to maximum denosumab concentration is 10 days with a mean half-life of 25.4 days. In conclusion, the convenient biannual subcutaneous administration of 60 mg of denosumab can be considered as a first-line treatment for osteoporosis in cases of low compliance with BPs due to gastrointestinal trouble and impaired renal function.

• Investigative Magnetic Resonance Imaging
• /
• v.20 no.4
• /
• pp.215-223
• /
• 2016

Purpose: The management of metal-induced field inhomogeneities is one of the major concerns of distortion-free magnetic resonance images near metallic implants. The recently proposed method called "Slice Encoding for Metal Artifact Correction (SEMAC)" is an effective spin echo pulse sequence of magnetic resonance imaging (MRI) near metallic implants. However, as SEMAC uses the noisy resolved data elements, SEMAC images can have a major problem for improving the signal-to-noise ratio (SNR) without compromising the correction of metal artifacts. To address that issue, this paper presents a novel reconstruction technique for providing an improvement of the SNR in SEMAC images without sacrificing the correction of metal artifacts. Materials and Methods: Low-rank approximation in each coil image is first performed to suppress the noise in the slice direction, because the signal is highly correlated between SEMAC-encoded slices. Secondly, SEMAC images are reconstructed by the best linear unbiased estimator (BLUE), also known as Gauss-Markov or weighted least squares. Noise levels and correlation in the receiver channels are considered for the sake of SNR optimization. To this end, since distorted excitation profiles are sparse, $l_1$ minimization performs well in recovering the sparse distorted excitation profiles and the sparse modeling of our approach offers excellent correction of metal-induced distortions. Results: Three images reconstructed using SEMAC, SEMAC with the conventional two-step noise reduction, and the proposed image denoising for metal MRI exploiting sparsity and low rank approximation algorithm were compared. The proposed algorithm outperformed two methods and produced 119% SNR better than SEMAC and 89% SNR better than SEMAC with the conventional two-step noise reduction. Conclusion: We successfully demonstrated that the proposed, novel algorithm for SEMAC, if compared with conventional de-noising methods, substantially improves SNR and reduces artifacts.

• Korean Journal of Remote Sensing
• /
• v.25 no.1
• /
• pp.21-30
• /
• 2009

The purposes of this paper are to present a selection method of neighboring pixels whose local statistics are similar to the center pixel and combine the selection result with mean curvature diffusion filter to reduce noises in remote sensed imagery. The order of selection of neighboring pixels is critical, especially for finding a pixel belonging to the homogeneous region, since the statistics of the homogeneous region vary according to the selection order. An effective strategy for selecting neighboring pixels, which uses rank-order differences vector obtained by computing the intensity differences between the center pixel and neighboring pixels and arranging them in ascending order, is proposed in this paper. By using region growing method, we divide the elements of the rank-ordered differences vector into two groups, homogeneous rank-ordered differences vector and outlier rank-ordered differences vector. The mean curvature diffusion filter is combined with a line process, which chooses selectively diffusion coefficient of the neighboring pixels belonging into homogeneous rank-ordered differences vector. Experimental results using an aerial image and a TerraSAR-X satellite image showed that the proposed method reduced more efficiently noises than some conventional adaptive filters using all neighboring pixels in updating the center pixel.

• The Transactions of the Korean Institute of Electrical Engineers D
• /
• v.50 no.2
• /
• pp.93-99
• /
• 2001

A new type geometric and mathematical network reduction model is introduced. Large-scale network is analyzed with analytic approach. The graph has many nodes, branches and loops. Circuit equation are obtained from these elements and connection rule. In this paper, the analytic relation between voltage source has a mutual different graphic property. Node-reduction procedure is achieved with this circuit property. Consequently voltage source value is included into the adjacent node-analyzing equation. A resultant model equations are reduced as much as voltage source number. Matrix rank is (n-1-k), where n, k is node and voltage source number. The reduction procedure is described and verified with geometric principle and circuit theory. Matrix type circuit equation can be composed with this technique. The last results shall be calculated by using computer.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.36 no.2B
• /
• pp.159-167
• /
• 2011

LTE-Advanced system requires uplink multi-antenna transmission in order to achieve the peak spectral efficiency of 15bps/Hz. In this paper, the uplink MIMO system model for the LTE-Advanced is proposed and an efficient link adaptation shceme using precoding is considered providing error rate reduction and system capacity enhancement. In particular, the proposed scheme determines a transmission rank by selecting the optimal wideband precoding matrix, which is based on the derived signal-to-interference and noise ratio (SINR) for the minimum mean squared error (MMSE) receivers of $2{\times}4$ multiple input multiple output (MIMO). The proposed scheme is verified by simulation with a practical MIMO channel model. The simulation results of average block-error-rate(BLER) reflect that the gain due to the proposed rank adapted transmission over full-rank transmission is evident particularly in the case of lower modulation and coding scheme (MCS) and high mobility, which means the severe channel fading environment.

• Journal of the Korea Society of Computer and Information
• /
• v.18 no.11
• /
• pp.31-38
• /
• 2013

Feature normalization as a pre-processing step has been widely used in classification problems to reduce the effect of different scale in each feature dimension and error as a result. Most of the existing methods, however, assume some distribution function on feature distribution. Even worse, existing methods do not use the labels of data points and, as a result, do not guarantee the optimality of the normalization results in classification. In this paper, proposed is a supervised rank normalization which combines rank normalization and a supervised learning technique. The proposed method does not assume any feature distribution like rank normalization and uses class labels of nearest neighbors in classification to reduce error. SVM, in particular, tries to draw a decision boundary in the middle of class overlapping zone, the reduction of data density in that area helps SVM to find a decision boundary reducing generalized error. All the things mentioned above can be verified through experimental results.

• Journal of Korean Physical Therapy Science
• /
• v.9 no.1
• /
• pp.1-8
• /
• 2002

This study set out to investigate what kind of effects the consistent visual stimuli and verbal and non verbal auditory stimuli have on pain alleviation, as well as to see the influence of joint application of visual and auditory stimuli at the same time on pain alleviation, according to lightness of 50lux and 200lux, ultimately providing basic data in setting up an environment in case of treating pain. The subject were comprised of 30 male and female adults with pain in the neck and back area. The subject were treated in their pain area with Transcutaneous Electrical Nerve Stimulator(TENS) 100HZ for 20 minutes in the research set where each visual, auditory, and joint visual and auditory stimuli was given. For analysis methods, Visual Analogue Scale(VAS) and McGill Pain Questionnaire were adopted to see the changes before and after treatment, and the electrocardiogram, systolic and diastolic pressure, number of heart rate and breathing frequence and endorphin were compared and analyzed using the Wilcoxon singed-rank test. And The Kreskal-walllis test was used to compare the two subgroups from each group. Wilcoxon singed-rank test and the Kreskal-walllis test was used to compare the two subgroups from each group. The results were as follows: 1. The group of 50lux and 200lux were compared given varying degrees of visual stimuli. The group of 200lux showed more reduction in pain points, average systolic and diastolic pressure and average endorphin. 2. The group of verbal and non verbal were compared given varying degrees of auditory stimuli. The group of non-verbal showed more reduction in average systolic and diastolic pressure. 3. The group of 200lux+verbal and 200lux+non verbal were compared given varying degrees of joint visual and auditory stimuli. There was found a statistical significance(p<0.05) in endorphin between the two groups, with more endorphin reduction for 200lux+non verbal group. And there was a statistically significant reduction in VAS and McGill before and after the treatment between the two groups.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.52 no.4
• /
• pp.145-155
• /
• 2015

Noise reduction has been actively studied in the digital image processing and recently, block-based denoising algorithms are widely used. In particular, a low rank approximation employing WNNM(Weighted Nuclear Norm Minimization) and block-based approaches demonstrated the potential for effective noise reduction. However, the algorithm based on low rank a approximation generates the artifacts in the image restoration step. In this paper, we analyzes the image content using the STFT(Short Time Fourier Transform) and proposes an effective method of minimizing the artifacts generated from the conventional algorithm. To evaluate the performance of the proposed scheme, we use the test images containing a wide range of noise levels and compare the results with the state-of-art algorithms.