• Proceedings of the Korean Society for Language and Information Conference
• /
• 2007.11a
• /
• pp.430-439
• /
• 2007

In this paper, we use non-negative matrix factorization (NMF) to refine the document clustering results. NMF is a dimensional reduction method and effective for document clustering, because a term-document matrix is high-dimensional and sparse. The initial matrix of the NMF algorithm is regarded as a clustering result, therefore we can use NMF as a refinement method. First we perform min-max cut (Mcut), which is a powerful spectral clustering method, and then refine the result via NMF. Finally we should obtain an accurate clustering result. However, NMF often fails to improve the given clustering result. To overcome this problem, we use the Mcut object function to stop the iteration of NMF.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2008.05a
• /
• pp.427-430
• /
• 2008

본 논문은 비음수 행렬 인수분해(NMF, non-negative matrix factorization)와 NMF 군집방법을 이용하여 다중문서를 요약하는 새로운 방법을 제안하였다. 본 논문에서 NMF에 의해 계산된 의미 특징(semantic feature)은 문서의 고유 구조(inherent structure)를 반영하여 문장을 추출함으로써 요약의 질을 높일 수 있고, 의미 변수(semantic variable)를 이용한 문장의 군집은 문장 간의 유사성과 다양성 고려하여서 쉽게 과잉정보를 제거하여 문장을 요약할 수 있는 장점을 갖는다.

• Genomics & Informatics
• /
• v.4 no.1
• /
• pp.33-39
• /
• 2006

Alternative splicing (AS) is an important mechanism of producing transcriptome diversity and microarray techniques are being used increasingly to monitor the splice variants. There exist three types of microarrays interrogating AS events-junction, exon, and tiling arrays. Junction probes have the advantage of monitoring the splice site directly. Johnson et al., performed a genome-wide survey of human alternative pre-mRNA splicing with exon junction microarrays (Science 302:2141-2144, 2003), which monitored splicing at every known exon-exon junctions for more than 10,000 multi-exon human genes in 52 tissues and cell lines. Here, we describe an algorithm to deduce the relative concentration of isoforms from the junction array data. Non-negative Matrix Factorization (NMF) is applied to obtain the transcript structure inferred from the expression data. Then we choose the transcript models consistent with the ECgene model of alternative splicing which is based on mRNA and EST alignment. The probe-transcript matrix is constructed using the NMF-consistent ECgene transcripts, and the isoform abundance is deduced from the non-negative least squares (NNLS) fitting of experimental data. Our method can be easily extended to other types of microarrays with exon or junction probes.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.41 no.2
• /
• pp.65-76
• /
• 2004

Today, most documents are electronically produced and their paleography is digitalized by imaging, resulting in a tremendous number of electronic documents in the shape of images. Therefore, to process these document images, many methods of document structure analysis and recognition have already been proposed, including font classification. Accordingly, the current paper proposes a font classification method for document images that uses non-negative matrix factorization (NMF), which is able to learn part-based representations of objects. In the proposed method, spatially total features of font images are automatically extracted using NMF, then the appropriateness of the features specifying each font is investigated. The proposed method is expected to improve the performance of optical character recognition (OCR), document indexing, and retrieval systems, when such systems adopt a font classifier as a preprocessor.

• Korean Journal of Cognitive Science
• /
• v.22 no.4
• /
• pp.347-364
• /
• 2011

In this paper, we propose a automatic image bases extraction method for visual image reconstruction from brain activity using Non-negative Matrix Factorization (NMF). Image bases are basic elements to construct and present a visual image. Previous method used brain activity that evoked by predefined 361 image bases of four different sizes: $1{\times}1$, $2{\times}1$, $1{\times}2$, $2{\times}2$, and $2{\times}2$. Then the visual stimuli were reconstructed by linear combination of all the results from these image bases. While the previous method used 361 predefined image bases, the proposed method automatically extracts image bases which represent the image data efficiently. From the experiments, we found that the proposed method reconstructs the visual stimuli better than the previous method.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.50 no.8
• /
• pp.196-202
• /
• 2013

Basal cell carcinoma (BCC) is the most common skin cancer and its incidence is increasing rapidly. In this paper, we propose a diagnosis method of basal cell carcinoma by Raman spectra of skin tissue using the NMF(non-negative matrix factorization) algorithm. After preprocessing steps, measured Raman spectra is used classification experiments. The weight and the basis can be obtained in a simple matrix operation and a column vector of the matrix decompsed by the NMF. Linear combination of bases and weights, it is possible to approximate the average of Raman spectra. The classification method is to select the class which to minimize the root mean square of the difference of the linear combination and the objective spectrum. According to the experimental results, the proposed method shows the promising results to diagnosis BCC. In addition, it confirmed that the proposed method compared with the previous research result could be effectively applied in the analysis of the Raman spectra.

• Journal of the Korea Society of Computer and Information
• /
• v.13 no.4
• /
• pp.11-18
• /
• 2008

Document clustering is an important method for document analysis and is used in many different information retrieval applications. This paper proposes a new document clustering model using the re-weighted term based NMF(non-negative matrix factorization) to cluster documents relevant to a user's requirement. The proposed model uses the re-weighted term by using user feedback to reduce the gap between the user's requirement for document classification and the document clusters by means of machine. The Proposed method can improve the quality of document clustering because the re-weighted terms. the semantic feature matrix and the semantic variable matrix, which is used in document clustering, can represent an inherent structure of document set more well. The experimental results demonstrate appling the proposed method to document clustering methods achieves better performance than documents clustering methods.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.23 no.6
• /
• pp.682-689
• /
• 2019

In part-based image representation, the partial shapes of an object are represented as basis vectors, and an image is decomposed as a linear combination of basis vectors where the coefficients of those basis vectors represent the partial (or local) feature of an object. In this paper, a face recognition for occluded faces is proposed in which face images are represented using non-negative matrix factorization(NMF), one of part-based representation techniques, and recognized using an artificial neural network technique. Standard NMF, projected gradient NMF and orthogonal NMF were used in part-based representation of face images, and their performances were compared. Learning vector quantizer were used in the recognizer where Euclidean distance was used as the distance measure. Experimental results show that proposed recognition is more robust than the conventional face recognition for the occluded faces.

• The Journal of the Acoustical Society of Korea
• /
• v.42 no.1
• /
• pp.32-41
• /
• 2023

Recently, studies on reverberation suppression using Non-negative Matrix Factorization (NMF) have been actively conducted. The NMF method uses a cost function based on the Kullback-Leibler divergence for optimization. And some constraints are added such as temporal continuity, pulse length, and energy ratio between reverberation and target. The tendency of constraints are controlled by hyperparameters. Therefore, in order to effectively suppress reverberation, hyperparameters need to be optimized. However, related studies are insufficient so far. In this paper, the reverberation suppression performance according to the three hyperparameters of the NMF was analyzed by using sea experimental data. As a result of analysis, when the value of hyperparameters for time continuity and pulse length were high, the energy ratio between the reverberation and the target showed better performance at less than 0.4, but it was confirmed that there was variability depending on the ocean environment. It is expected that the analysis results in this paper will be utilized as a useful guideline for planning precise experiments for optimizing hyperparameters of NMF in the future.

• The Journal of the Acoustical Society of Korea
• /
• v.41 no.3
• /
• pp.326-334
• /
• 2022

In this paper, we propose speech enhancement algorithm as a pre-processing for robust speech recognition in noisy environments. Auxiliary-function-based Independent Vector Analysis (AuxIVA) is performed with weighted covariance matrix using time-varying variances with scaling factor from target masks representing time-frequency contributions of target speech. The mask estimates can be obtained using Neural Network (NN) pre-trained for speech extraction or diffuseness using Coherence-to-Diffuse power Ratio (CDR) to find the direct sounds component of a target speech. In addition, outputs for omni-directional noise are closely chained by sharing the time-varying variances similarly to independent subspace analysis or IVA. The speech extraction method based on AuxIVA is also performed in Independent Low-Rank Matrix Analysis (ILRMA) framework by extending the Non-negative Matrix Factorization (NMF) for noise outputs to Non-negative Tensor Factorization (NTF) to maintain the inter-channel dependency in noise output channels. Experimental results on the CHiME-4 datasets demonstrate the effectiveness of the presented algorithms.