• Proceedings of the Korean Information Science Society Conference
• /
• 1998.10c
• /
• pp.333-335
• /
• 1998

에러 역전파 신경망에서 학습속도와 수렴률은 초기 가중의 분포에 따라 크게 영향을 받는다. 본 연구에서는 이를 위하여 비교사 학습 신경망(Hebbian learning rule)을 이용한 새로운 초기 가중치 결정 방법을 제안한다. 또는 비교사 학습 신경망이 에러 역전파 신경망 학습에 적당하도록 은닉층의 각 뉴런과 연결된 가중치의 norm을 이용하여 학습하였다. 시뮬레이션을 통하여 기존 에러 역전파 신경망 학습과 그 성능을 비교한 결과 제안한 초기 가중치 표현이 학습속도와 수렴능력에서 우수함을 나타낸다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2005.05a
• /
• pp.477-480
• /
• 2005

본 논문에서는 신경망과 유전자 알고리즘의 장점을 결합하고, 개선된 유전자 알고리즘 기반의 역전파 신경망 알고리즘을 이용한 신경망 학습 효과를 살펴 본다. 유전자 알고리즘을 이용한 신경망 학습은 비선형 함수를 이용하여 발생시킨 모의 자료를 통하여 수행하고 학습 수렴의 정도와 학습 속도 등을 비교할 수 있는 모의실험 결과를 일반 신경망 학습 결과와 함께 제시한다. 모의실험의 결과로서 유전자 알고리즘을 적용한 신경망 제어기가 일반 신경망 학습 결과보다 수렴 정확도 및 학습 속도에서 더 좋은 결과를 나타내 주고 있다.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.19 no.10
• /
• pp.2001-2013
• /
• 1994

In this paper, the method of processing a blurred noisy image has been researched. The conventional method of processing signal has faluts which are slow convergence speed and long time-consuming process at the singular point and or in the ill condition. There is the process, the Gauss Seidel's method to remove these faults, but it takes too much time because it processed singnal repeatedly. For overcoming the faults, this paper shows a image restoration method which takes shorter than the Gauss-Seidel's by comparing the Gauss Seidel's with proposed alogorithm and accelerating convergence speed at the singular point and/or in the ill condition. In this paper, the conventional process method(Gauss-Seidel) and proposed optimal algorithm were used to get a standard image($256{\times}56{\times}bits$). and then the results are simulated and compared each other in order to examine the variance of MSE(Mean Square Error) by the acceleration parameter in the proposed image restoration. The result of the signal process and the process time was measured at all change of acceleration parameter in order to verify the effectveness of the proposed algorithm.

• Journal of KIISE:Software and Applications
• /
• v.33 no.12
• /
• pp.999-1007
• /
• 2006

The Backpropagation learning algorithm is known to have slow and false convergence aroused from plateau and local minima. Many substitutes for backpropagation announced so far appear to pay some trade-off for convergence speed and stability of convergence according to parameters. Here, a new algorithm is proposed, which avoids some of those problems associated with the conventional backpropagation problems, especially with local minima, and gives relatively stable and fast convergence with low storage requirement. This is the separate learning algorithm in which the upper connections, hidden-to-output, and the lower connections, input-to-hidden, separately trained. This algorithm requires less computational work than the conventional backpropagation and other improved algorithms. It is shown in various classification problems to be relatively reliable on the overall performance.

• 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 the Korean Society of Propulsion Engineers
• /
• v.15 no.2
• /
• pp.8-14
• /
• 2011

In this paper, the combustion characteristics of hybrid rocket fuel with tapered grain port were investigated experimentally. The charging efficiency of convergent and divergent port shape fuel with $1^{\circ}$ taper angle was 6.8% higher than that of cylindrical port shape fuel. The regression rate was increased about 17.5% by using the convergent port shape fuel. On the other hand, in case of divergent port shape fuel, no notable difference of regression rate was observed when compared to that of the cylindrical port shape fuel. In the case of convergent port shape fuel, characteristic velocity and its efficiency were notably increased with respect to cylindrical port fuel. It was found that convergent port shape of hybrid rocket fuel can lead to a better option compared to the conventional cylindrical port in terms of combustion efficiency and performance improvement.

• The Journal of the Acoustical Society of Korea
• /
• v.19 no.8
• /
• pp.54-62
• /
• 2000

Adaptive filtering method is one of signal processing area which is frequently used in the case of statistical characteristic change in time-varing situation. The performance of adaptive filter is usually evaluated with complexity of its structure, convergence speed and misadjustment. The structure of adaptive filter must be simple and its speed of adaptation must be fast for real-time implementation. In this paper, we propose chirp discrete cosine transform (CDCT), which has the characteristics of CZT (chrip z-transform) and DCT (discrete cosine transform), and then CDCTLMS (chirp discrete cosine transform LMS) using the above mentioned algorithm for the improvement of its speed of adaptation. Using loaming curve, we prove that the proposed method is superior to the conventional US (normalized LMS) algorithm and DCTLMS (discrete cosine transform LMS) algorithm. Also, we show the real application for the ultrasonic signal processing.

• The Journal of the Acoustical Society of Korea
• /
• v.33 no.6
• /
• pp.420-426
• /
• 2014

In this paper, we propose a new affine projection sign algorithm (APSA) to improve the convergence speed of the conventional APSA which has been proposed to enable the affine projection algorithm (APA) to operate robustly in impulsive noise environment. The conventional APSA has two advantages; it operates robustly against impulsive noise and does not need calculation for the inverse matrix. The proposed algorithm also has the conventional algorithm's advantages and furthermore, better convergence speed than the conventional algorithm. In the conventional algorithm, each input signal is normalized by $l_2$-norm of all input signals, but the proposed algorithm uses input signals normalized by their corresponding $l_2$-norm. We carried out a performance comparison of the proposed algorithm with the conventional algorithm using a system identification model. It is shown that the proposed algorithm has the faster convergence speed than the conventional algorithm.

• The Journal of the Acoustical Society of Korea
• /
• v.34 no.2
• /
• pp.130-137
• /
• 2015

In this paper, we propose a new Modified Filtered-x Affine Projection Sign Algorithm(MFxAPSA) to improve the convergence speed of the conventional MFxAPSA which has been proposed for active control of impulsive noise. Under the impulsive noise environment, the adaptive algorithms based on the second order moment such as the Filtered-x Least Mean Square(FxLMS) show slow convergence speed or diverge because the noise source tends to have infinite variance. The MFxAPSA is the algorithm derived by applying the Affine Projection Sign Algorithm(APSA) to active noise control. The APSA has an advantage that it does not need the calculation for the inverse matrix, so it may be suitable for the active noise control that requires low computational burden. The proposed MFxAPSA also has APSA's advantage and furthermore, better performance than the conventional MFxAPSA. We carried out a performance comparison of the proposed MFxAPSA with the conventional MFxAPSA. It is shown that the proposed MFxAPSA has the faster convergence speed than the conventional MFxAPSA.

• The Journal of Korean Institute of Electromagnetic Engineering and Science
• /
• v.10 no.7
• /
• pp.1085-1094
• /
• 1999

In the global system for the mobile communication (GSM) system, the class of training sequences used in a TDMA frame is a preamble sequence with a period of 16 bits for the channel impulse response measurement and the start up equalization during the training period. If the transmitted preamble sequence and the binary sequence in the receiver properly satisfy a condition, this training sequences used for fastly adjusting the tap coefficients and impulse response can be measured by calculating the crosscorrelation function. In this paper, it is used that training sequences have zero values of the autocorrelation at all delays except zero and middle shifts. A comparison of convergence rate, a mathematical approach for fast start-up equalization and correctly channel impulse response measurement are proposed.