• 전자공학회논문지B
• /
• 제33B권9호
• /
• pp.49-61
• /
• 1996

Back-propagation(BP) algorithm needs a lot of time to train the artificial neural network (ANN) to get high accuracy level in classification tasks. So there have been extensive researches to process back-propagation algorithm on parallel processors. This paper prsents a linear systolic array which calculates forward-backward propagation of BP algorithm at the same time using effective space-time transformation and PE structure. First, we analyze data flow of forwared and backward propagations and then, represent the BP algorithm into data dapendency graph (DG) which shows parallelism inherent in the BP algorithm. Next, apply space-time transformation on the DG of ANN is turn with orthogonal direction projection. By doing so, we can get a snakelike systolic array. Also we calculate the interval of input for parallel processing, calculate the indices to make the right datas be used at the right PE when forward and bvackward propagations are processed in the same PE. And then verify the correctness of output when forward and backward propagations are executed at the same time. By doing so, the proposed system maximizes parallelism of BP algorithm, minimizes th enumber of PEs. And it reduces the execution time by 2 times through making idle PEs participate in forward-backward propagation at the same time.

• 한국음향학회지
• /
• 제28권6호
• /
• pp.514-519
• /
• 2009

PAST (projection approximation subspace tracking)는 여러 연구자들에 의해 비교 연구가 되는 대표적인 신호 부공간을 추정하는 알고리즘이다. 이 방법은 신호 부공간 추정에 있어 상대적으로 낮은 계산 복잡도를 요구하기 때문에 인기가 있는 방법이 되었다. 그러나 이 방법은 개선의 여지가 많아서 추정 정확도나 공간의 직교성 등에서 계속 개선된 알고리즘이 연구되고 있다. 본 논문은 기존에 연구된 PAST 알고리즘의 추정 정확도 개선을 위해 연구된 알고리즘 중 하나인 FB-PAST(Forward-Backward PAST) 알고리즘에 사용할 직교성 알고리즘을 제안한다.

• 산업공학
• /
• 제17권spc호
• /
• pp.141-151
• /
• 2004

As today's business environment has become more and more competitive, forward as well as backward flows of products among members belonging to a supply chain have been increased. The backward flows of products, which are common in most industries, result from increasing amount of products that are returned, recalled, or need to be repaired. Effective management for the backward flows of products has become an important issue for businesses because of opportunities for simultaneously enhancing profitability and customer satisfaction from returned products. Since third party logistics service providers (3PLs) are playing an important role in reverse logistics operations, they should perform two simultaneous logistics operations for a number of different clients who want to improve their logistics operations for both forward and reverse flows. In this case, distribution networks have been independently designed with respect to either forward or backward flows so far. This paper proposes a mixed integer programming model for the design of network integrating both forward and reverse logistics. Since the network design problem belongs to a class of NP-hard problems, we present an efficient heuristic algorithm based on genetic algorithm (GA), of which the performance is compared to the lower bound by Lagrangian relaxation. Finally, the validity of proposed algorithm is tested using numerical examples.

• 정보처리학회논문지C
• /
• 제12C권3호
• /
• pp.437-442
• /
• 2005

인터넷 망의 종단간 혼잡상태를 유추할 수 있는 가장 일반적인 방법은 RTT (Round Trip Time)값을 측정하는 것이며, 이 방법으로 망의 흔잡 정도를 측정할 수 있다. 그러나 RTT는 패킷의 왕복 시간만을 측정하기 때문에 패킷의 송수신시 순방향과 역방향에서 어느 정도의 혼잡과 전송지연이 발생하였는가는 알 수가 없다. 본 논문에서는 패킷이 전송될 때 순방향/역방향 전송 지연을 계산하여, 망의 흔잡 상태를 정확하게 유추할 수 있는 새로운 알고리즘을 제시한다. 본 알고리즘에서는 여러 RTT 값들 중에서 가장 작은 RTT 값을 기준으로 하여 기준이 되는 순방향/역방향 전송 시간을 결정하고, 이 값과 각 패킷이 전송될 때 측정된 전송 시간을 비교하여 순방향/역방향 전송 지연 시간을 계산한다. 본 연구에서는 NS-2에서의 시뮬레이션과 실제 네트워크 상에의 측정을 통하여 제안된 방법의 올바른 동작을 확인하였다.

• Industrial Engineering and Management Systems
• /
• 제12권4호
• /
• pp.380-388
• /
• 2013

Large variation in electrocardiogram (ECG) waveforms continues to present challenges in defining R-wave locations in ECG signals. This research presents a procedure to extract the R-wave locations by forward-backward (FB) algorithm and classify the arrhythmic beat conditions by using RR intervals. The FB algorithm shows forward and backward searching rules from QRS onset and eliminates lower-amplitude signals near the baseline using a statistical process control concept. The proposed algorithm was trained the optimal parameters by using MIT-BIH arrhythmia database (MITDB), and it was verified by actual Holter ECG signals from a local hospital. The signals are classified into normal (N) and three arrhythmia beat types including premature ventricular contraction (PVC), ventricular flutter/fibrillation (VF), and second-degree heart block (BII) beat. This work produces 98.54% accuracy in the detection of R-wave location; 98.68% for N beats; 91.17% for PVC beats; and 87.2% for VF beats in the collected Holter ECG signals, and the results are better than what are reported in literature.

• 대한전자공학회:학술대회논문집
• /
• 대한전자공학회 2004년도 ICEIC The International Conference on Electronics Informations and Communications
• /
• pp.442-446
• /
• 2004

This paper propose scene change detection algorithm using coefficient of forward prediction macro-block, backward prediction macro-block, and intra-coded macro-block on getting motion estimation. Proposed method detect scene change with correlation according picture type forward two picture or forward and backward two picture on video sequences. Proposed algorithm is high accuracy and can detect all scene change on video, and detect to occur scene change on P, B, I-picture.

• 제어로봇시스템학회논문지
• /
• 제8권11호
• /
• pp.984-990
• /
• 2002

We propose a new type of nonlinear fixed interval smoother to which an existing nonlinear smoother is modified. The nonlinear smoother is derived from two-filter formulas. For the backward filter. the propagation and the update equation of error states are derived. In particular, the modified update equation of the backward filter uses the estimated error terms from the forward filter. Data fusion algorithm, which combines the forward filter result and the backward filter result, is altered into the compatible form with the new type of the backward filter. The proposed algorithm is more efficient than the existing one because propagation in backward filter is very simple from the implementation point of view. We apply the proposed nonlinear smoothing algorithm to off-line navigation system and show the proposed algorithm estimates position, and altitude fairly well through the computer simulation.

• 한국통신학회논문지
• /
• 제21권6호
• /
• pp.1425-1429
• /
• 1996

누락펄스가 발생할 시 PRI 분석을 위한 전형적인 sequence search 알고리즘은 harmonic 문제를 발생시킨다. 본 논문에서 이러한 harmonic 문제를 해결할 수 있는 PRI 분석 알고리즘을 제안한다. 분석해야 할 전체 PRI 범위를 harmonic이 없도록 subband 로 나눈 후, sequence search를 forward 및 backward로 행한다. 전형적인 sequence search 알고리즘에 비하여 제안한 알고리즘의 성능이 증대된다.

• 전자공학회논문지B
• /
• 제32B권10호
• /
• pp.1326-1337
• /
• 1995

In this paper, we transformed the bilinear filter into an equivalent linear multichannel filter and derived QR decomposition based recursive least squares algorithms for bilinear lattice filters. We also defined order update relation of the forward and the backward input vectors by using the channel decomposition. The forward and the backward data matrices were defined by using the forward and the backward input vectors and orthogonalized with the QR decomposition. we can obtain the lattice equations of the bilinear filters by using the channel decomposition. we can be derived the lattice equations of the bilinear filters using this decomposition process which are the same as the lattice equations derived by Baik, we can use the coefficient transformation algorithm proposed by Baik. We derived the equation error and the output error algorithm of the QRD based RLS bilinear lattice algorithm. Also, we evaluated the performance of the proposed algorithms through the system identification of the bilinear system.

• 대한산업공학회지
• /
• 제42권5호
• /
• pp.314-326
• /
• 2016

The purpose of variable selection techniques is to select a subset of relevant variables for a particular learning algorithm in order to improve the accuracy of prediction model and improve the efficiency of the model. We conduct an empirical analysis to evaluate and compare seven well-known variable selection techniques for multiple linear regression model, which is one of the most commonly used regression model in practice. The variable selection techniques we apply are forward selection, backward elimination, stepwise selection, genetic algorithm (GA), ridge regression, lasso (Least Absolute Shrinkage and Selection Operator) and elastic net. Based on the experiment with 49 regression data sets, it is found that GA resulted in the lowest error rates while lasso most significantly reduces the number of variables. In terms of computational efficiency, forward/backward elimination and lasso requires less time than the other techniques.