• Journal of Computing Science and Engineering
• /
• v.4 no.2
• /
• pp.97-109
• /
• 2010

Nonnegative matrix factorization (NMF) is a popular method for multivariate analysis of nonnegative data, which is to decompose a data matrix into a product of two factor matrices with all entries restricted to be nonnegative. NMF was shown to be useful in a task of clustering (especially document clustering), but in some cases NMF produces the results inappropriate to the clustering problems. In this paper, we present an algorithm for orthogonal nonnegative matrix factorization, where an orthogonality constraint is imposed on the nonnegative decomposition of a term-document matrix. The result of orthogonal NMF can be clearly interpreted for the clustering problems, and also the performance of clustering is usually better than that of the NMF. We develop multiplicative updates directly from true gradient on Stiefel manifold, whereas existing algorithms consider additive orthogonality constraints. Experiments on several different document data sets show our orthogonal NMF algorithms perform better in a task of clustering, compared to the standard NMF and an existing orthogonal NMF.

• Journal of KIISE:Software and Applications
• /
• v.36 no.5
• /
• pp.347-352
• /
• 2009

Nonnegative matrix factorization (NMF) is a popular method for multivariate analysis of nonnegative data, the goal of which is decompose a data matrix into a product of two factor matrices with all entries in factor matrices restricted to be nonnegative. NMF was shown to be useful in a task of clustering (especially document clustering). In this paper we present an algorithm for orthogonal nonnegative matrix factorization, where an orthogonality constraint is imposed on the nonnegative decomposition of a term-document matrix. We develop multiplicative updates directly from true gradient on Stiefel manifold, whereas existing algorithms consider additive orthogonality constraints. Experiments on several different document data sets show our orthogonal NMF algorithms perform better in a task of clustering, compared to the standard NMF and an existing orthogonal NMF.

• Honam Mathematical Journal
• /
• v.24 no.1
• /
• pp.103-108
• /
• 2002

Applications of a column-reduced orthogonal rational matrix function to McMillan degrees and Wiener-Hopf factorizations are considered.

• Bulletin of the Korean Mathematical Society
• /
• v.39 no.3
• /
• pp.359-376
• /
• 2002

We give a new interpretation of Darboux transforms in the context of orthogonal polynomials and find conditions in or-der for any Darboux transform to yield a new set of orthogonal polynomials. We also discuss connections between Darboux trans-forms and factorization of linear differential operators which have orthogonal polynomial eigenfunctions.

• Journal of the Korean Mathematical Society
• /
• v.45 no.6
• /
• pp.1613-1622
• /
• 2008

Let G be a graph with vertex set V(G) and edge set E(G), and let g, f be two nonnegative integer-valued functions defined on V(G) such that $g(x)\;{\leq}\;f(x)$ for every vertex x of V(G). We use $d_G(x)$ to denote the degree of a vertex x of G. A (g, f)-factor of G is a spanning subgraph F of G such that $g(x)\;{\leq}\;d_F(x)\;{\leq}\;f(x)$ for every vertex x of V(F). In particular, G is called a (g, f)-graph if G itself is a (g, f)-factor. A (g, f)-factorization of G is a partition of E(G) into edge-disjoint (g, f)-factors. Let F = {$F_1$, $F_2$, ..., $F_m$} be a factorization of G and H be a subgraph of G with mr edges. If $F_i$, $1\;{\leq}\;i\;{\leq}\;m$, has exactly r edges in common with H, we say that F is r-orthogonal to H. If for any partition {$A_1$, $A_2$, ..., $A_m$} of E(H) with $|A_i|=r$ there is a (g, f)-factorization F = {$F_1$, $F_2$, ..., $F_m$} of G such that $A_i\;{\subseteq}E(F_i)$, $1\;{\leq}\;i\;{\leq}\;m$, then we say that G has (g, f)-factorizations randomly r-orthogonal to H. In this paper it is proved that every (0, mf - (m - 1)r)-graph has (0, f)-factorizations randomly r-orthogonal to any given subgraph with mr edges if $f(x)\;{\geq}\;3r\;-\;1$ for any $x\;{\in}\;V(G)$.

• Korean Management Science Review
• /
• v.21 no.2
• /
• pp.43-60
• /
• 2004

Every iteration of interior-point methods of large scale optimization requires computing at least one orthogonal projection. In the practice, symmetric variants of the Gaussian elimination such as Cholesky factorization are accepted as the most efficient and sufficiently stable method. In this paper several specific implementation issues of the symmetric factorization that can be applied for solving such equations are discussed. The code called McSML being the result of this work is shown to produce comparably sparse factors as another implementations in the $MATLAB^{***}$ environment. It has been used for computing projections in an efficient implementation of self-regular based interior-point methods, McIPM. Although primary aim of developing McSML was to embed it into an interior-point methods optimizer, the code may equally well be used to solve general large sparse systems arising in different applications.

• Journal of the Korean Mathematical Society
• /
• v.46 no.2
• /
• pp.281-294
• /
• 2009

Accuracy of the scaling function is very crucial in wavelet theory, or correspondingly, in the study of wavelet filter banks. We are mainly interested in vector-valued filter banks having matrix factorization and indicate how to choose block central symmetric matrices to construct multi-wavelets with suitable accuracy.

• Journal of Communications and Networks
• /
• v.5 no.4
• /
• pp.328-343
• /
• 2003

Orthogonal pulse-shape modulation using Hermite pulses for ultra-wideband communications is reviewed. Closedform expressions of cross-correlations among Hermite pulses and their corresponding transmit and receive waveforms are provided. These show that the pulses lose orthogonality at the receiver in the presence of differentiating antennas. Using these expressions, an algebraic model is established based on the projections of distorted receive waveforms onto the orthonormal basis given by the set of normalized orthogonal Hermite pulses. Using this new matrix model, a number of pulse-shape modulation schemes are analyzed and a novel orthogonal design is proposed. In the proposed orthogonal design, transmit waveforms are constructed as combinations of elementary Hermites with weighting coefficients derived by employing the Gram-Schmidt (QR) factorization of the differentiating distortion model’s matrix. The design ensures orthogonality of the vectors at the output of the receiver bank of correlators, without requiring compensation for the distortion introduced by the antennas. In addition, a new set of elementary Hermite Pulses is proposed which further enhances the performance of the new design while enabling a simplified hardware implementation.

• Journal of applied mathematics & informatics
• /
• v.29 no.3_4
• /
• pp.1037-1042
• /
• 2011

Multiwavelet coefficients can be constructed from the multi-scaling coefficients by using the factorization for paraunitary matrices. In this paper we present a procedure for parametrizing all possible multi-wavelet coefficients corresponding to the multiscaling coefficients of DGHM type.

• Journal of information and communication convergence engineering
• /
• v.6 no.3
• /
• pp.261-265
• /
• 2008

This work presents a simplified efficient blind detection algorithm for orthogonal space-time codes(OSTBC). First, the proposed decoder exploits a proper decomposition approach of the upper triangular matrix R, which resulted from Cholesky-factorization of the composition channel matrix, to form an easy-to-solve blind detection equation. Secondly, in order to avoid suffering from the high computational load, the proposed decoder applies a sub-optimal QR-based decoder. Computer simulation results verify that the proposed decoder allows to significantly reduce computational complexity while still satisfying the bit-error-rate(BER) performance.