• 디지털산업정보학회논문지
• /
• 제17권4호
• /
• pp.53-62
• /
• 2021

Deep learning models such as convolutional neural networks and recurrent neual networks process a huge amounts of data, so they require a lot of storage and consume a lot of time and power due to memory access. Recently, research is being conducted to reduce memory usage and access by compressing data using the feature that many of deep learning data are highly sparse and localized. In this paper, we propose a compression-decompression method of storing only the non-zero data and the location information of the non-zero data excluding zero data. In order to make the location information of non-zero data, the matrix data is divided into sections uniformly. And whether there is non-zero data in the corresponding section is indicated. In this case, section division is not executed only once, but repeatedly executed, and location information is stored in each step. Therefore, it can be properly compressed according to the ratio and distribution of zero data. In addition, we propose a hardware structure that enables compression and decompression without complex operations. It was designed and verified with Verilog, and it was confirmed that it can be used in hardware deep learning accelerators.

• International Journal of Precision Engineering and Manufacturing
• /
• 제8권1호
• /
• pp.3-10
• /
• 2007

A new approach based on implicit surface interpolation combined with domain decomposition is proposed for filling complex-shaped holes in a large polygon model, A surface was constructed by creating a smooth implicit surface from an incomplete polygon model through which the actual surface would pass. The implicit surface was defined by a radial basis function, which is a continuous scalar-value function over the domain $R^{3}$. The generated surface consisted of the set of all points at which this scalar function is zero. It was created by placing zero-valued constraints at the vertices of the polygon model. The well-known domain decomposition method was used to treat the large polygon model. The global domain of interest was divided into smaller domains in which the problem could be solved locally. The LU decomposition method was used to solve the set of small local problems; the local solutions were then combined using weighting coefficients to obtain a global solution. The validity of this new approach was demonstrated by using it to fill various holes in large and complex polygon models with arbitrary topologies.

• Korean Journal of Mathematics
• /
• 제16권1호
• /
• pp.51-55
• /
• 2008

Let f be an analytic function with a simple zero in the reals or the complex numbers. An extraneous fixed-point of an iteration function is a fixed-point different from a zero of f. We prove that all extraneous fixed-points of Laguerre-like iteration functions and general Laguerre-like functions are repulsive.

• 제어로봇시스템학회:학술대회논문집
• /
• 제어로봇시스템학회 1987년도 한국자동제어학술회의논문집(한일합동학술편); 한국과학기술대학, 충남; 16-17 Oct. 1987
• /
• pp.812-816
• /
• 1987

A quadratic regulator problem for a class of nonlinear system, in which a small parameter multiplies the control cost, is considered. In the analysis of the problem, we utilize the method of multiple time-scale decomposition which has been devised for analyzing complex linear cheap control problems. In so doing, we extend the class of nonlinear systems, considerably, for which the minimum cost becomes zero as the small parameter goes to zero.

• 제어로봇시스템학회:학술대회논문집
• /
• 제어로봇시스템학회 2002년도 ICCAS
• /
• pp.44.4-44
• /
• 2002

This paper investigates some conditions such that zeros are added to the system with a monotone nondecreasing step response in order to hold the monotonicity as before. Two conditions are presented for the cases that a real zero and complex conjugate zeros are added to the system satisfying the monotonicity condition, respectively. To exemplify the proposed results, some simple examples via computer simulation are shown in this paper. Proposed conditions can be easily used in the control system design since they are only formulated in terms of pole-zero configurations.

• 대한수학회논문집
• /
• 제9권4호
• /
• pp.825-830
• /
• 1994

This paper is concerned with the zeros of successive derivatives of real entire functions. In order to state our results, we introduce the following notations : An entire function which assumes only real values on the real axis is said to be a real entire function. Thus, if a complex number is a zero of a real entire function, then its conjugate is also a zero of the same function.

• 대한수학회지
• /
• 제33권4호
• /
• pp.747-761
• /
• 1996

Thoughout this paper, R will be a commutative ring (with non-zero identity) and M will denote an R-module.

• International journal of advanced smart convergence
• /
• 제6권4호
• /
• pp.33-40
• /
• 2017

In this paper, complex quadruplet zeros of microwave filter systems are investigated. For the cascaded systems the chain matrices are most conveniently used to derive the voltage transfer function of Laplace transform with cascaded two-port subsystems. The convenient relations of transfer function and chain matrix are used in order to find the transmission zeros. Starting from a ladder network, we introduced a crossed-coupled lumped element, in order to show the improved response of bandpass filter. By solving the transmission zero characteristic equation derived from the cascaded subsystems, we found the zeros of filter system with externally cross-coupled lumped elements. With the cross-coupled elements of capacitors, the numerator polynomial of system transfer function is used to locate the quadruplet zeros in complex plane. When the two pairs of double are on the zeros -axis, with the perturbed values of element, we learned that the transition band of lowpass filter is improved. By solving the characteristic equation of cascaded transfer function, we can obtain the zeros of the cross-coupled filter system, as a result of perturbed values on lumped element.

• Bulletin of the Korean Chemical Society
• /
• 제24권12호
• /
• pp.1731-1736
• /
• 2003

The reaction of ${\alpha}$-Benzilmonoxime with palladium(II) produces a green complex in triton X-100 micellar media. Palladium has been determined using zero and first derivative spectrophotometric methods. The absorbances of Pd(II)- ${\alpha}$--benzilmonoxime complex at 441.8 and 677.0 nm in 0.10 M perchloric acid solution were monitored and linear working ranges of 0.3-12.0 and 0.7-20 ${\mu}$g mL$^{-1}$ with detection limits of 0.07 and 0.10 ${\mu}$g $mL^-1$ were obtained, respectively. Also, recoveries in the range of 92.8 to 100.1% and relative standard deviations in the range of 0.4 to 7.1% were obtained. First derivative spectrophotometry has also been applied for palladium determination under the optimum condition. The linear dynamic range of 0.2-24.0 ${\mu}$g $mL^{-1}$ palladium with relative standard deviations of 0.6-6.9% and recoveries in the range of 94.9-102.5% has been obtained by first derivative spectrophotometry. The method shows high selectivity because of the high concentration of acid used, which prevents formation of complexes of ${\alpha}$--benzilmonoxime with the other cations. The palladium complex formed was stable at least one day. The method was successfully applied to the determination of palladium in some synthetic palladium alloys and palladium-charcoal powder.

• 대한수학회지
• /
• 제47권5호
• /
• pp.1001-1015
• /
• 2010

Let M be a $C^{\infty}$ real hypersurface in $\mathbb{C}^{n+1}$, $n\;{\geq}\;1$, locally given as the zero locus of a $C^{\infty}$ real valued function r that is defined on a neighborhood of the reference point $P\;{\in}\;M$. For each k = 1,..., n we present a necessary and sufficient condition for there to exist a complex manifold of dimension k through P that is contained in M, assuming the Levi form has rank n - k at P. The problem is to find an integral manifold of the real 1-form $i{\partial}r$ on M whose tangent bundle is invariant under the complex structure tensor J. We present generalized versions of the Frobenius theorem and make use of them to prove the existence of complex submanifolds.