• Korean Journal of Mathematics
• /
• v.6 no.1
• /
• pp.27-46
• /
• 1998

We are concerned with an optimal control problem governed by a Poisson equation in which body force acts like a control parameter. The cost functional to be optimized is taken to represent the error from the desired observation and the cost due to the control. We recast the problem into the mixed formulation to take advantage of the minimax principle for the duality method. The existence of a saddle point for the Lagrangian shall be shown and the optimality system will be derived therein. Finally, to attain an optimal control, we combine the optimality system with an operational technique. By achieving the gradient of the cost functional, a convergent algorithm based on the projected gradient method is established.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.50 no.7
• /
• pp.231-235
• /
• 2013

Lens distortions have a significant impact on captured or projected image geometry. This paper proposes a lens distortion correction with gradient components for wide-angle lenses. In most cases, distortion coefficients are estimated using a distortion model by point correspondences. Corrected images using only point correspondences can be compensated excessively, therefore, producing bended lines into the opposite direction near the corners. To curtail these phenomena, we propose to adopt the gradient components in addition to positions to obtain the distortion coefficients. We verified the improved accuracy and the straightness of the proposed method through experimentation.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.34 no.5
• /
• pp.1047-1057
• /
• 2024

Recently, deep neural network-based reinforcement learning models have been applied in various advanced industrial fields such as autonomous driving, smart factories, and home networks, but it has been shown to be vulnerable to malicious adversarial attack. In this paper, we applied deep reinforcement learning models, DQN and PPO, to the autonomous driving simulation environment HighwayEnv and conducted three adversarial attacks: FGSM(Fast Gradient Sign Method), BIM(Basic Iterative Method), PGD(Projected Gradient Descent) and CW(Carlini and Wagner). In order to respond to adversarial attack, we proposed a method for deep learning models based on reinforcement learning to operate normally by removing noise from adversarial images using a bilateral filter algorithm. Furthermore, we analyzed performance of adversarial attacks using two popular metrics such as average of episode duration and the average of the reward obtained by the agent. In our experiments on a model that removes noise of adversarial images using a bilateral filter, we confirmed that the performance is maintained as good as when no adversarial attack was performed.

• Annual Conference of KIPS
• /
• 2019.05a
• /
• pp.404-407
• /
• 2019

최근 많은 영역에 딥러닝이 활용되고 있다. 특히 CNN과 같은 아키텍처는 얼굴인식과 같은 이미지 분류 분야에서 활용된다. 이러한 딥러닝 기술을 완전한 기술로서 활용할 수 있는지에 대한 연구가 이뤄져왔다. 관련 연구로 PGD(Projected Gradient Descent) 공격이 존재한다. 해당 공격을 이용하여 원본 이미지에 노이즈를 더해주게 되면, 수정된 이미지는 전혀 다른 클래스로 분류되게 된다. 본 연구에서 기존의 FGSM(Fast gradient sign method) 공격기법에 Triplet loss를 활용한 Adversarial 공격 모델을 제안 및 구현하였다. 제안된 공격 모델은 간단한 시나리오를 기반으로 검증하였고 해당 결과를 분석하였다.

• The Korean Journal of Ecology
• /
• v.23 no.4
• /
• pp.293-298
• /
• 2000

To study the effect of global warming on the growth of plants and plant populations throughout their life cycle under a field-like condition, we constructed a Temperature Gradient Chamber (TGC) in Tsukuba, Japan. The chamber had slender shape : 30 m long. 3 m wide, and 2.5 m high. That satisfactory performance was confirmed by a test throughout all seasons in 1998: the projected global warming condition in the near future was simulated. That is, independent of a great daily or seasonal change in ambient meteorological conditions, air temperatures at the air outlet were warmed 5$^{\circ}C$ higher than those at the ambient (the annual mean was 14.3$^{\circ}C$) with precision of ${\pm}$0.2$^{\circ}C$ (the annual means were 19.2$^{\circ}C$) with a rising rate of approximately 1$^{\circ}C$ every 5 m. This chamber will enable us to study the effects of global warming on growth of plants and plant populations because their abilities to control air temperature are excellent. TGC is expected that it would be utilized for studying the effect of global warming on plant growth under natural weather conditions.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.21 no.1
• /
• pp.87-92
• /
• 2021

Bias, variance, error and learning are important factors for performance in modeling a big data based recommendation system. The recommendation model in this system must reduce complexity while maintaining the explanatory diagram. In addition, the sparsity of the dataset and the prediction of the system are more likely to be inversely proportional to each other. Therefore, a product recommendation model has been proposed through learning the similarity between products by using a factorization method of the sparsity of the dataset. In this paper, the generalization ability of the model is improved by applying the max-norm regularization as an optimization method for the loss function of this model. The solution is to apply a stochastic projection gradient descent method that projects a gradient. The sparser data became, it was confirmed that the propsed regularization method was relatively effective compared to the existing method through lots of experiment.

• The Bulletin of The Korean Astronomical Society
• /
• v.45 no.1
• /
• pp.41.2-42
• /
• 2020

Understanding how the filamentary structure plays a role in the formation of the prestellar cores and stars is a key issue to challenge. We have observed two prestellar cores in surrounding filamentary environments in 13CO, C180 (3-2) and HCO+ (4-3) molecular lines with the Heterodyne Array Receiver Program (HARP) of the James Clerk Maxwell Telescope (JCMT), in order to search for the evidence related to the possible flow motions along the filament and/or the radial accretion (or infalling motions) of gas material toward the dense cores from their surrounding filamentary cloud. In L1544, the velocity gradient of 1.6 km s-1 pc-1 toward the core was measured in a small branch of filament lying on a radial direction of main filament while no velocity gradient along the main axis of filament in both 13CO and C18O lines. In L694-2, we found the velocity gradient of 0.6 km s-1 pc-1 along the filament in only 13CO lines. The projected accretion rate of ~6 M◉ Myr-1 was estimated in both cases. The infall (or radially contracting) velocity of gas material was measured ~0.16 km s-1 in both 13CO and HCO+ lines and in both L1544 and L694-2, which leads to estimate a mass infall rate of ~20 M◉ Myr-1. Our analysis suggests that our targets are at a stage where the gravitational contraction dominates the mass accretion through the surrounding filamentary cloud. This is consistent with the fact that our targets are highly evolved prestellar cores on a verge of star formation. More detailed results will be presented at the meeting.

• 한국전산유체공학회:학술대회논문집
• /
• 2009.11a
• /
• pp.49-55
• /
• 2009

This paper evaluates performances of a recently developed divergence-free finite element method based on Hermite interpolated stream functions. Velocity bases are derived from Hermite interpolated stream functions to form divergence-free basis functions. These velocity basis functions constitute a solenoidal function space, and the simple gradient of the Hermite functions constitute an irrotational function space. The incompressible Navier-Stokes equation is orthogonally decomposed into a solenoidal and an irrotational parts, and the decoupled Navier-Stokes equations are projected onto their corresponding spaces to form proper variational formulations. To access accuracy and convergence of the present algorithm, three test problems are selected. They are lid-driven cavity flow, flow over a backward-facing step and buoyancy-driven flow within a square enclosure. Hermite interpolation functions from cubic to quintic are chosen to run the test problems. Numerical results are shown. In all cases it has shown that the present method has performed well in accuracies and convergences. Moreover, the present method does not require an upwinding or a stabilized term.

• The KIPS Transactions:PartB
• /
• v.8B no.2
• /
• pp.215-222
• /
• 2001

본 논문에서는 새로운 형태 특징값으로서 형태 correlogram을 제안하고 이를 기반으로 한 효과적인 내용기반 영삼검색(content-based image retrieval) 방법을 제시한다. 기존읜 색상 correlogram은 색상 정보에 공간적인 정보를 부여함으로써 영상검색 성능을 향상시켰다. 그러나 이 특징값은 형태 정보를 포함하고 있지 않아서 색상이 다르면서 비슷한 윤곽선 형태를 갖는 물체의 검색에는 좋은 효과를 보이지 못한다.이 문제를 해결하기 위해 예지(edge)들의 correlogram인 형태(shape) correlogram을 제안한다. 색상 correlogram이 색상들의 거리에 따른 상관관계를 나타내는데 반해 형태 correlogram은 에지 각도들의 상관관게를 나타낸다. 형태 correlogram은 gradient 축과 각도 축을 가지는 2차원 특징 벡터(feature vector)로 표현된다. 각 축은 24개 빈(bin)으로 나뉘어져서 총 576개의 원소를 가지게 된다. 또한 본 논문에서는 형태 correlogram의 데이터 크기를 줄이고, 회전에 대해 불변인 특성을 가지게 하기 위해 투영(projected) 형태 correlogram을 제안한다. 실험결과를 통하여 본 논문에서 제안한 형태 correlogram과 투영 형태 correlogram을 사용한 영상검색 방법이 기존의 방법보다 성능면에서 우수함을 입증한다.

• 제어로봇시스템학회:학술대회논문집
• /
• 1992.10b
• /
• pp.1-6
• /
• 1992

A principle of "orthogonalization" is proposed as an extended notion of hybrid (force and position) control for robot manipulators under geometric endpoint constraints. The principle realizes the hybrid control in a strict sense by letting position and velocity feedback signals be orthogonal in joint space to the contact force vector whose components are exerted at corresponding joints. This orthogonalization is executed via a projection matrix computed in real-time from a gradient of the equation of the surface in joint coordinates and hence both projected position and velocity feedback signals become perpendicular to the force vector that is normal to the surface at the contact point in joint space. To show the important role of the principle in control of robot manipulators, three basic problems are analyzed, the first is a hybrid trajectory tracking problem by means of a "modified hybrid computed torque method", the second is a model-based adaptive control problem for robot manipulators under geometric endpoint constraints, and the third is an iterative learning control problem. It is shown that the passivity of residual error dynamics of robots follows from the orthogonalization principle and it plays a crucial role in convergence properties of both positional and force error signals.force error signals.