• 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 the Korea Society of Computer and Information
• /
• v.22 no.8
• /
• pp.15-23
• /
• 2017

In this paper, the block reweighting and global offset initialization methods are proposed to complement the improved IRLMS algorithm which is the effective algorithm in registration based SBNUC algorithm. Proposed block weighting method reweights the error map whose abnormal data are excluded. The global offset initialization method compensates the global nonuniformity initially. The ordinary registration based SBNUC algorithm is hard to compensate global nonuniformity because of low scene motion. We employ the proposed methods to improved IRLMS algorithm, and apply it to real-world infrared raw image stream. The result shows that new implementation provides 3.5~4.0dB higher PSNR and convergence speed 1.5 faster then the improved IRLMS algorithm.

• Journal of the Korean Institute of Illuminating and Electrical Installation Engineers
• /
• v.21 no.4
• /
• pp.11-18
• /
• 2007

A moving object detection algorithm for surveillance video is here proposed which employs background initialization based on two-stage median filtering and a background updating method based on state transition diagram. In the background initialization, the spatiotemporal similarity is measured in the subinterval. From the accumulated difference between the base frame and the other frames in a subinterval, the regions affected by moving objects are located. The median is applied over the subsequence in the subinterval in which regions share similarity. The outputs from each subinterval are filtered by a two-stage median filter. The background of every frame is updated by the suggested state transition diagram The object is detected by the difference between the current frame and the updated background. The proposed method showed good results even for busy, crowded sequences which included moving objects from the first frame.

• Journal of Korea Multimedia Society
• /
• v.21 no.4
• /
• pp.441-449
• /
• 2018

Deep learning has been used for various processing centered on image recognition. One core algorithms of the deep learning, convolutional neural network is an deep neural network that specialized in image recognition. In this paper, we use a convolutional neural network to classify forest insects and propose an optimization method. Experiments were carried out by combining two weight initialization and six parameter update methods. As a result, the Xavier-SGD method showed the highest performance with an accuracy of 82.53% in the 12 different combinations of experiments. Through this, the latest learning algorithms, which complement the disadvantages of the previous parameter update method, we conclude that it can not lead to higher performance than existing methods in all application environments.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.33 no.1C
• /
• pp.110-116
• /
• 2008

Maximum likelihood(ML) diretion-of-arrival(DOA) estimation is essentially optimization of multivariable nonlinear cost function. Since the final estimate is highly dependent on the initial estimate, an initialization is critical in nonlinear optimization. We propose a multi-dimensional(M-D) search scheme of uniform exhaustive search and improved exhaustive search. Improved exhaustive search is superior to uniform exhaustive search in terms of the computational complexity and the accuracy of the estimates.

• Communications for Statistical Applications and Methods
• /
• v.12 no.2
• /
• pp.497-508
• /
• 2005

This paper aims at studying on K-means Clustering focusing on initialization which affect the clustering results in K-means cluster analysis. The four different methods(the MA method, the KA method, the Max-Min method and the Space Partition method) were compared and the clustering result shows that there were some differences among these methods, especially that the MA method sometimes leads to incorrect clustering due to the inappropriate initialization depending on the types of data and the Max-Min method is shown to be more effective than other methods especially when the data size is large.

• Proceedings of the Korean Nuclear Society Conference
• /
• 1996.05a
• /
• pp.88-93
• /
• 1996

The initial condition should be consistent with real reactor core state for the simulation of the core transient. The initial xenon distribution, which cad not be measured in the core, has a significant effect on the transient with xenon dynamics of PWR. In the simulation of the transient stating from non-equilibrium xenon state, the accurate initialization of the non-equilibrium xenon distribution is essential to predict the core transient behavior. In this study, the xenon initialization method to predict the core transient more accurately was developed through the first-order perturbation theory of the relationship between simulated power and measured power distribution and verified by the application of the simulation for a startup test of Yonggwang Unit 3.

• The Korean Journal of Applied Statistics
• /
• v.32 no.5
• /
• pp.693-702
• /
• 2019

Deep learning has gained popularity for the classification and prediction task. Neural network layers become deeper as more data becomes available. Saturation is the phenomenon that the gradient of an activation function gets closer to 0 and can happen when the value of weight is too big. Increased importance has been placed on the issue of saturation which limits the ability of weight to learn. To resolve this problem, Glorot and Bengio (Proceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics, 249-256, 2010) claimed that efficient neural network training is possible when data flows variously between layers. They argued that variance over the output of each layer and variance over input of each layer are equal. They proposed a method of initialization that the variance of the output of each layer and the variance of the input should be the same. In this paper, we propose a new method of establishing initialization by adopting truncated normal distribution and truncated cauchy distribution. We decide where to truncate the distribution while adapting the initialization method by Glorot and Bengio (2010). Variances are made over output and input equal that are then accomplished by setting variances equal to the variance of truncated distribution. It manipulates the distribution so that the initial values of weights would not grow so large and with values that simultaneously get close to zero. To compare the performance of our proposed method with existing methods, we conducted experiments on MNIST and CIFAR-10 data using DNN and CNN. Our proposed method outperformed existing methods in terms of accuracy.

• Proceedings of the Korean Society of Surveying, Geodesy, Photogrammetry, and Cartography Conference
• /
• 2004.04a
• /
• pp.57-57
• /
• 2004

This study analyzed and compared the results of baseline processing using Epoch-By-Epoch algorithm which is not required initialization compared to conventional kinematic surveying which is required initialization There are rarely differences between 24 hours data of 30 seconds interval and 90 seconds of 30 seconds when it is processed with 26km baseline. This helps with economic surveying using data of GPS CORS

• Proceedings of the IEEK Conference
• /
• 1998.06a
• /
• pp.887-890
• /
• 1998

We present a method for initialization of asynchronous circuits using binary decision space representation. From state transition graph(STG) which is given as a specification a circuit, the BDD is generated to solve the state space explosion problem which is caused by concurrecy of STG. We first step, we construct the necessary informaton as a form of K-map from BDD, then find an initial state on the K-map by assignment of don't care assignment.