• International Journal of Contents
• /
• 제10권1호
• /
• pp.18-22
• /
• 2014

Hidden nodes have a key role in the information processing of feed-forward neural networks in which inputs are processed through a series of weighted sums and nonlinear activation functions. In order to understand the role of hidden nodes, we must analyze the effect of the nonlinear activation functions on the weighted sums to hidden nodes. In this paper, we focus on the effect of nonlinear functions in a viewpoint of information theory. Under the assumption that the nonlinear activation function can be approximated piece-wise linearly, we prove that the entropy of weighted sums to hidden nodes decreases after piece-wise linear functions. Therefore, we argue that the nonlinear activation function decreases the uncertainty among hidden nodes. Furthermore, the more the hidden nodes are saturated, the more the entropy of hidden nodes decreases. Based on this result, we can say that, after successful training of feed-forward neural networks, hidden nodes tend not to be in linear regions but to be in saturated regions of activation function with the effect of uncertainty reduction.

• 한국통신학회논문지
• /
• 제23권9A호
• /
• pp.2333-2344
• /
• 1998

We present a two step approach for estimating the motionand sturcture parameters from region orrespondences in two frames. Given four or more region corresondences on the same planar surface, the motion and planar orientation parameters are first linearly estimated based on second-order approximation of the displacement field of the image plane. Then, using this linear estimate as an initial guess, a nonlinear estimate is obtained by iteratively minimizing an objective function using the exact experession of the displacement field. The objective function involves the centroids of corresponding regions and relationships among low-order moments. Through simulations, we show that the two-step region-based approach gives robust estimates. The performance of nonlinear region-based estimation is compared with that of linear region-based and point-based methods. Experimental results for two image pairs, on esynthetic and one real, ar epresented to show the practical applicability of our approach.

• 제어로봇시스템학회논문지
• /
• 제1권1호
• /
• pp.13-19
• /
• 1995

In this paper, generalized singular value is defined. Using the generalized singular value, robust stability conditions and robust pole placement conditions of structured uncertain systems with star shaped uncertainties are derived. Especially, norm bounded and polytopic uncertainty regions are considered as star shaped uncertainty regions. Linear matrix inequality problems are proposed in order to compute the upper bound of the generalized singular value. The proposed linear matrix inequality problems can be solved by using the convex optimization method.

• Bulletin of the Korean Chemical Society
• /
• 제9권4호
• /
• pp.223-231
• /
• 1988

Some pattern recognition methods have been used to characterize Korean ancient porcelainsherds using their elemental composition as analyzed by instrumental neutron activation analysis. A combination of analytical data by means of statistical linear discriminant analysis(SLDA) has resulted in removal of redundant variables, optimal linear combination of meaningful variables and formulation of classification rules. The plot in the first-to-second discriminant scores has shown that the three distinct territorial regions exist among porcelainsherds of Kyungki, Chunbuk-Chungnam, and Chunnam, with respective efficiencies of 20/30, 22/27 and 14/15. Similar regions have been found to exist among punchong porcelain and ceradonsherds of Kyungki, Chungnam and Chunbuk, with respective efficiencies of 7/9, 15/16 and 6/6. Classification has been further attempted by statistical isolinear multiple component analysis(SIMCA), using the sample set selected appropriately through SLDA as training set. For this purpose, all analytical data have been used. An agreement has generally been found between two methods, i.e., SLDA and SIMCA.

• 한국지능시스템학회논문지
• /
• 제23권1호
• /
• pp.18-23
• /
• 2013

In this paper we propose an image enhancement technique based on histogram specification method over local overlapping regions referred as Local Histogram Specification. First, both reference and original images are splitted into local regions that each overlaps half of its adjacent regions and general histogram specification method is used between corresponding local regions of reference and original image. However it produces noticeable boundary effects. Linear weighted image blending method is used to reduce this effect in order to make seamless image and we also proposed new technique dealing with over-enhanced contrast areas. We satisfied with our experimental results that showed better enhancement accuracy and less noise amplifications compared to other well-known image enhancement methods. We conclude that the proposed method is well suited for motion detection systems as a responsible part to overcome sudden illumination changes.

• 한국신재생에너지학회:학술대회논문집
• /
• 한국신재생에너지학회 2010년도 춘계학술대회 초록집
• /
• pp.72.1-72.1
• /
• 2010

Thin film a-Si solar cells deposited by PECVD have many advantages compared to the traditional crystalline Si solar cells. They do not require expensive Si wafer, the process temperature is relatively low, possibility of scaling up for mass production, etc. In order to produce thin film solar cells, understanding the relationship between the material characteristics and deposition conditions is important. It has been reported by many groups that the band gap of the a-Si material and the deposition rate has an linear relationship, when RF power is used to control both. However, when the process pressure is changed in order to control the deposition rate and the band gap, a diversion from the well known linear relationship occurs. Here, we explain this diversion by the deposition condition crossing different plasma regions in the Paschen curve with a simple model. This model will become a guide to which condition a-Si thin films must be fabricated in order to get a high quality film.

• 전자공학회논문지T
• /
• 제35T권1호
• /
• pp.14-21
• /
• 1998

GaAs MESFET는 동작점에 따라 선형영역, 포화영역으로 구분된 모델을 사용함에 따라, GaAs MESFET 회로 해석을 위한 컴퓨터 시뮬레이션 시 영역의 경계점에서의 1차 및 2차 미분 불연속으로 인한 해의 발산 문제가 발생하곤 하였다. 본 논문에서는 선형영역과 포화영역을 모두 포함한 통합된 채널길이변조식을 제안하였다. 새로이 제안된 채널길이변조식을 이용하여 전류-전압 모델과 커패시턴스-전압 모델을 제안하였다. 제안된 전류-전압 모델은 Shur의 모델과 비교하였으며 제안된 커패시턴스-전압 모델은 디바이스 시뮬레이션 결과와 비교하였다. 비교된 결과로부터 제안된 모델들은 기존의 모델과 유사한 결과를 얻었으며 연속성이 개선될 것으로 기대된다.

• 한국국방경영분석학회지
• /
• 제14권1호
• /
• pp.33-41
• /
• 1988

In Multiple Objective Linear Programming (MOLP), it is well known that efficient solution and weight are correspondent to each other. The purpose of this paper is to study relationships between efficient face and the region of weight in MOLP. It is shown that the regions of weights corresponding to two efficient extreme points are also neighbor if two efficient extreme points are neighbor each other, and that the set of the efficient solutions corresponding to the common part of weight regions is efficient face. Using the above, we present a method to find the efficient solutions corresponding to a given weight and vice versa.

• 한국국방경영분석학회지
• /
• 제13권2호
• /
• pp.33-41
• /
• 1987

In Multiple Objective Linear Programming (MOLP), it is well known that efficient solution and weight are correspondent to each other. The purpose of this paper is to study relationships between efficient face and the region of weight in MOLP. It is shown that the regions of weights corresponding to two efficient extreme points are also neighbor if two efficient extreme points are neighbor each other, and that the set of the efficient solutions corresponding to the common part of weight regions is efficient face. Using the above, we present a method to find the efficient solutions corresponding to a given weight and vice versa.

• 품질경영학회지
• /
• 제25권2호
• /
• pp.102-111
• /
• 1997

In this paper I discuss a method of constructing the confidence region for the logistic response surface model. The construction involves a, pp.ication of a general fitting procedure because the log odds is linear in its parameters. Estimation of parameters of the logistic response surface model can be accomplished by maximum likelihood, although this requires iterative computational method. Using the asymptotic results, asymptotic covariance of the estimators can be obtained. This can be used in the construction of confidence regions for the parameters and for the logistic response surface model.