• Asia-pacific Journal of Multimedia Services Convergent with Art, Humanities, and Sociology
• /
• v.8 no.2
• /
• pp.351-358
• /
• 2018

Reinforcement learning is a machine learning method which aims to determine a policy to get optimal actions in dynamic and stochastic environments. But reinforcement learning has high computational complexity and needs a lot of time to get solution, so it is not easily applicable to uncertain and continuous environments. To tackle the complexity problem, AC (actor-critic) method is used and it separates an action-value function into a value function and an action decision policy. Also, in transfer learning method, the knowledge constructed in one environment is adapted to another environment, so it reduces the time to learn in a reinforcement learning method. In this paper, we present AC method and transfer learning method to solve the problem of a reinforcement learning method. Finally, we analyze the case study which a transfer learning method is used to solve BS(base station) switching problem in wireless access networks.

• Journal of the Korea Organic Resources Recycling Association
• /
• v.2 no.1
• /
• pp.51-64
• /
• 1994

To compost the food garbage with the dry bean curd and sawdust as the bulking agents, the method of high-speed fermentation by the characteristic microorganisms group was applied. The results of experiments are summarized as follows ; 1. Korean food garbage, which is high in water content, is difficult to compost only by microorganism fermentation without the addition of bulking agents such as dry bean curd cake and sawdust. 2. Weight reduction rates are ranging from 35.6% to 64.5% and varying with the composition of food garbage. The less weight reduction rate is, the longer continuous-fermentation is. And the color of compost is changing sequentially as yellow -> brown -> black. 3. Comparing with the controlled microorganism group, the weight reduction rate and $H_2CO_3$ production rate in the characteristic microorganism group fermentation reactors are higher. And the fermentation rate is satisfactory when the characteristic microorganism group is added. 4. The value of fermented composting as fertilizer diminishes, and the contents of Total Nitrogen, $P_2O_5$, $K_2O$ increase on the condition that the fermentation continues. However, the organic contents and C/N ratio diminish as the fermentation continues. 5. The high-speed fermentation technology demonstractes the possibility of recycling as well as the reduction of composting time provided that it is applied as a pretreatment process for composting.

• The Transactions of the Korean Institute of Electrical Engineers
• /
• v.32 no.11
• /
• pp.394-403
• /
• 1983

This paper discusses a model reference adaptive control for a multi-input multi-output continuos system in matrix fraction description. The controller is of Monopoli-Narendra type with a time-varying gain matrix in the parameter adaptation law. The transfer matrix of the given plant with an adjustable controller is made to approach to that of the reference model asymptotically. It is shown that, under some plausible assumptions such as on the knowlidge of an interactor matrix, the algorithm for a single-input single-output system can be appropriately extended to a multi-input multi-output system. The convergence of an adaptation law is estavlished with some stability theory and stability of the overall system is asserted by an analytical investigation.

• The KIPS Transactions:PartB
• /
• v.14B no.7
• /
• pp.541-548
• /
• 2007

Due to the mathematical convergence property, Simulated Annealing (SA) has been one of the most popular optimization algorithms. However, because of its problem of slow convergence in the practical use, many variations of SA like Fast SA (FSA) have been developed for faster convergence. In this paper, we propose and prove that Greedy SA (GSA) also finds the global optimum in probability in the continuous space optimization problems. Because the greedy selection does not allow the cost to become worse, GSA is expected to have faster convergence than the conventional FSA that uses Metropolis selection. In the computer simulation, the proposed method is shown to have as good performance as FSA with Metropolis selection in the viewpoints of the convergence speed and the quality of the found solution. Furthermore, the greedy selection does not concern the cost value itself but uses only dominance of the costs of solutions, which makes GSA invariant to the problem scaling.

• Journal of Institute of Control, Robotics and Systems
• /
• v.7 no.9
• /
• pp.723-732
• /
• 2001

In this paper, the neurofuzzy controller of optically guided AGV is proposed to improve the path-tracking performance A differential steered AGV has front-side and rear-side optical sensors, which can identify the guiding path. Due to the discontinuity of measured data in optical sensors, optically guided AGVs break away easily from the guiding path and path-tracking performance is being degraded. Whenever the On/Off signals in the optical sensors are generated discontinuously, the motion errors can be measured and updated. After sensing, the variation of motion errors can be estimated continuously by the dead reckoning method according to left/right wheel angular velocity. We define the estimated contour error as the sum of the measured contour in the sensing error and the estimated variation of contour error after sensing. The neurofuzzy system consists of incorporating fuzzy controller and neural network. The center and width of fuzzy membership functions are adaptively adjusted by back-propagation learning to minimize th estimated contour error. The proposed control system can be compared with the traditional fuzzy control and decision system in their network structure and learning ability. The proposed control strategy is experience through simulated model to check the performance.

• Korean Journal of Remote Sensing
• /
• v.27 no.4
• /
• pp.403-410
• /
• 2011

Consistent observation with high temporal and spatial resolution is required for an efficient monitoring of coastal environments. Remote monitoring system installed on the ground is capable of simultaneous observation of wide coastal area and consistent observation with high frequency, which a small number of in-situ measurements cannot manage. This paper studies two typical ground based coastal monitoring system, marine radar and camera system. Marine radar can produce time series of frequency spectrum by integrating wave number spectrum calculated from spatial and temporal variation of waves in the radar image. The time averaged radar images of waves can analyze wave breaking zone, rip currents and location of littoral bars. Camera system can observe temporal variation of foam generation originated from coastal contamination as well as shoreline changes. By extracting the part of foams from rectified images, quantitative analysis of temporal foam variation can be done. By using the two above systems of different characteristics, synergetic benefit can be achieved.

• The Transactions of the Korea Information Processing Society
• /
• v.4 no.7
• /
• pp.1759-1769
• /
• 1997

The Error BackPropagation (EBP) algorithm for multi-layered neural networks is widely used in various areas such as associative memory, speech recognition, pattern recognition and robotics, etc. Nevertheless, many researchers have continuously published papers about improvements over the original EBP algorithm. The main reason for this research activity is that EBP is exceeding slow when the number of neurons and the size of training set is large. In this study, we developed new learning speed acceleration methods using variable learning rate, variable momentum rate and variable slope for the sigmoid function. During the learning process, these parameters should be adjusted continuously according to the total error of network, and it has been shown that these methods significantly reduced learning time over the original EBP. In order to show the efficiency of the proposed methods, first we have used binary data which are made by random number generator and showed the vast improvements in terms of epoch. Also, we have applied our methods to the binary-valued Monk's data, 4, 5, 6, 7-bit parity checker and real-valued Iris data which are famous benchmark training sets for machine learning.

• Journal of Preventive Medicine and Public Health
• /
• v.31 no.3 s.62
• /
• pp.490-502
• /
• 1998

This study purports to find the predictor of hospital bankruptcy in Korea and to examine the predictive power of the discriminant function model of hospital bankruptcy. Data on 17 financial and 4 non-financial indicators of 31 bankrupt and 31 profitable hospitals of 1, 2, and 3 years before bankruptcy were obtained from the hospital performance databank of Korea Institute of Health Services Management. Significant variables were identified through mean comparison of each indicator between bankrupt and profitable hospitals, and the discriminant function model of hospital bankruptcy was developed. The major findings are as follows 1. As for profitability indicators, net worth to total assets, operating profit to total capital, operating profit ratio to gross revenues, normal profit to total assets, normal profit to gross revenues, net profit to total assets were significantly different in mean comparison test in 1, 2, and 3 years before hospital bankruptcy. With regard to liquidity indicators, current ratio and quick ratio were significant in 1 year before bankruptcy. For activity indicators, patients receivable turnover was significant in 2 and 3 years before bankruptcy and added value per adjusted inpatient days was significant in 3 years before bankruptcy. 2. The discriminant function in 1, 2, and 3 years before bankruptcy were; $Z=-0.0166{\times}quick$ ratio-$0.1356{\times}normal$ profit to total assets-$1.545{\times}total$ assets turnrounds in 1 year before bankruptcy, $Z=-0.0119{\times}quick$ ratio-$0.1433{\times}operating$ profit to total assets-$0.0227{\times}value$ added to total assets in 2 years before bankruptcy, and $Z=-0.3533{\times}net$ profit to total assets-$0.1336{\times}patients$ receivables turn-rounds-$0.04301{\times}added$ value per adjusted $patient+0.00119{\times}average$ daily inpatient census in 3 years before bankruptcy. 3. The discriminant function's discriminant power in 1, 2, and 3 years before bankruptcy was 77.42, 79.03, 82.25% respectively.