• Proceedings of the IEEK Conference
• /
• 2001.06c
• /
• pp.13-16
• /
• 2001

In reinforcement teaming, Q-learning converges quite slowly to a good policy. Its because searching for the goal state takes very long time in a large stochastic domain. So I propose the speedup method using the Q-value initialization for model-free reinforcement learning. In the speedup method, it learns a naive model of a domain and makes boundaries around the goal state. By using these boundaries, it assigns the initial Q-values to the state-action pairs and does Q-learning with the initial Q-values. The initial Q-values guide the agent to the goal state in the early states of learning, so that Q-teaming updates Q-values efficiently. Therefore it saves exploration time to search for the goal state and has better performance than Q-learning. 1 present Speedup Q-learning algorithm to implement the speedup method. This algorithm is evaluated. in a grid-world domain and compared to Q-teaming.

• Journal of Korea Technology Innovation Society
• /
• v.5 no.2
• /
• pp.167-188
• /
• 2002

Summary: This paper analyzes Indian software industry in the perspective of innovation cluster. The research shows that the software industry has been following an upstream clustering process, where the major value activity is expanding from low value product/services to high value product/services. The growth of software industry could be successful because there was appropriate initial condition of Bangalore, such as the availability of high qualified human resources, excellent research institutes, small high-tech companies. The role of government was helpful for the late growth of software industry but not a critical factor for the initial development of the S/W cluster. It is suggested that government should consider the initial condition of a concerned location critically to implement a cluster-type innovation policy.

• Journal of the Korean Society for Industrial and Applied Mathematics
• /
• v.17 no.4
• /
• pp.221-237
• /
• 2013

In this paper an asymptotic numerical method named as Initial Value Method (IVM) is suggested to solve the singularly perturbed weakly coupled system of reaction-diffusion type second order ordinary differential equations with negative shift (delay) terms. In this method, the original problem of solving the second order system of equations is reduced to solving eight first order singularly perturbed differential equations without delay and one system of difference equations. These singularly perturbed problems are solved by the second order hybrid finite difference scheme. An error estimate for this method is derived by using supremum norm and it is of almost second order. Numerical results are provided to illustrate the theoretical results.

• Kyungpook Mathematical Journal
• /
• v.52 no.2
• /
• pp.167-177
• /
• 2012

In this article, we propose exponentially fitted error correction methods(EECM) which originate from the error correction methods recently developed by the authors (see [10, 11] for examples) for solving nonlinear stiff initial value problems. We reduce the computational cost of the error correction method by making a local approximation of exponential type. This exponential local approximation yields an EECM that is exponentially fitted, A-stable and L-stable, independent of the approximation scheme for the error correction. In particular, the classical explicit Runge-Kutta method for the error correction not only saves the computational cost that the error correction method requires but also gives the same convergence order as the error correction method does. Numerical evidence is provided to support the theoretical results.

• Journal of The Korean Astronomical Society
• /
• v.39 no.4
• /
• pp.147-150
• /
• 2006

In this paper, initial value problem for dynamical astronomy will be established using parabolic cylindrical coordinates. Computation algorithm is developed for the initial value problem of gravity perturbed trajectories. Applications of the algorithm for the problem of final state predication are illustrated by numerical examples of seven test orbits of different eccentricities. The numerical results are extremely accurate and efficient in predicating final state for gravity perturbed trajectories which is of extreme importance for scientific researches as well as for military purposes. Moreover, an additional efficiency of the algorithm is that, for each of the test orbits, the step size used for solving the differential equations of motion is larger than 70% of the step size used for obtaining its reference final state solution.

• Structural Engineering and Mechanics
• /
• v.22 no.2
• /
• pp.131-150
• /
• 2006

Exact analytical solutions for in-plane static problems of planar curved beams with variable curvatures and variable cross-sections are derived by using the initial value method. The governing equations include the axial extension and shear deformation effects. The fundamental matrix required by the initial value method is obtained analytically. Then, the displacements, slopes and stress resultants are found analytically along the beam axis by using the fundamental matrix. The results are given in analytical forms. In order to show the advantages of the method, some examples are solved and the results are compared with the existing results in the literature. One of the advantages of the proposed method is that the high degree of statically indeterminacy adds no extra difficulty to the solution. For some examples, the deformed shape along the beam axis is determined and plotted and also the slope and stress resultants are given in tables.

• Kyungpook Mathematical Journal
• /
• v.56 no.2
• /
• pp.311-327
• /
• 2016

In this paper, we propose an error embedded Runge-Kutta method to improve the traditional embedded Runge-Kutta method. The proposed scheme can be applied into most explicit embedded Runge-Kutta methods. At each integration step, the proposed method is comprised of two equations for the solution and the error, respectively. These solution and error are obtained by solving an initial value problem whose solution has the information of the error at each integration step. The constructed algorithm controls both the error and the time step size simultaneously and possesses a good performance in the computational cost compared to the original method. For the assessment of the effectiveness, the van der Pol equation and another one having a difficulty for the global error control are numerically solved. Finally, a two-body Kepler problem is also used to assess the efficiency of the proposed algorithm.

• Journal of the Korean Society for Industrial and Applied Mathematics
• /
• v.28 no.1
• /
• pp.33-58
• /
• 2024

This paper reviews basic concepts for Physics-Informed Neural Networks (PINN) applied to the initial value problems for ordinary differential equations. In particular, using only basic calculus, we derive the error estimates where the error functions (the differences between the true solution and the approximations expressed by neural networks) are dominated by training loss functions. Numerical experiments are conducted to validate our error estimates, visualizing the relationship between the error and the training loss for various first-order differential equations and a second-order linear equation.

• Proceedings of the KIEE Conference
• /
• 2006.10c
• /
• pp.446-448
• /
• 2006

DRAM with ECC is used widely and the size of DRAW increases. According to this, DRAM initial time, especially the time to make the whole area typical value, 0, increases. This paper introduces the method that without any additional hardware, using characteristic of DRAM and DRAM controller, minimize that memory initial time. Conservative reordering - it eliminates DRAM read time and makes write buffer used - reduces initial time to make the whole DRAM area 0, by 95.36% for DDR DRAM. 9341% for Rambus DRAM.

• Journal of Advanced Navigation Technology
• /
• v.23 no.2
• /
• pp.214-220
• /
• 2019

The calibration method of the initial value error obtained in the weight measurement through anchor bolt type strain gauge sensor is proposed. The strain gauge sensor is developed for preventing the overturning of forklift, which is the most frequent type of safety-accident in industry. It was confirmed that the initial value error is caused from the physical and mechanical error of anchor bolt, and the environmental problem. Since the elimination of these causes falls outside the realm of this research, we find out the calibrated values based on all the causes, and we adjust the initial values of analog-to-digital convertor (ADC) module consisted of strain gauge sensor block using the calibrated values. We use the linear interpolation method for our calibration. We confirm that four sensor modules have the different under 5% between the real weight and the measured value in the experiment applied with the calibration of initial values. The low correlation between the real weights and ADC values is also improved through the proposed calibration.