• 전자공학회논문지B
• /
• 제31B권12호
• /
• pp.108-115
• /
• 1994

Most pattern classifiers have been designed based on the ML (Maximum Likelihood) training algorithm which is simple and relatively powerful. The ML training is an efficient algorithm to individually estimate the model parameters of each class under the assumption that all class models in a classifier are statistically independent. That assumption, however, is not valid in many real situations, which degrades the performance of the classifier. In this paper, we propose a minimum-error-rate training algorithm based on the MAP (Maximum a Posteriori) approach. The algorithm regards the normalized outputs of the classifier as estimates of the a posteriori probability, and tries to maximize those estimates. According to Bayes decision theory, the proposed algorithm satisfies the condition of minimum-error-rate classificatin. We apply this algorithm to NPM (Neural Prediction Model) for speech recognition, and derive new disrminative training algorithms. Experimental results on ten Korean digits recognition have shown the reduction of 37.5% of the number of recognition errors.

• 대한조선학회지
• /
• 제23권2호
• /
• pp.27-32
• /
• 1986

This paper describes an application of the adaptive finite element computations to a free surface flow problem in a canal. A-posteriori error estimates for the adaptive finite element computations are based on the dual extremum principles. Previously the dual extremum principles were applied to compute the upper and lower bounds of the added mass of two-dimensional cylinders in a canal[1,2]. However, the present method improves the convergence of the computed results by utilizing the local error estimates and by applying the adaptive meshes in the finite element computations. In a test result using triangular elements it is shown that the numerical error in the adaptive finite elements reduces quadratically compared with that in a uniform mesh subdivision.

• 대한기계학회논문집A
• /
• 제26권10호
• /
• pp.2060-2066
• /
• 2002

The closed-loop state and input observer is a pole-placement type observer and estimates unknown state and input variables simultaneously. Pole-placement type observers may have poor transient performance with respect to ill-conditioning factors such as unknown initial estimates, round-off error, etc. For the robust transient performance, the effects of these ill-conditioning factors must be minimized in designing observers. In this paper, the transient performance of the closed-loop state and input observer is investigated quantitatively by considering the error bounds due to ill-conditioning factors. The performance indices are selected from these error bounds and are related to the observer robustness with respect to the ill -conditioning factors. The closed-loop state and input observer with small performance indices is considered as a well-conditioned observer from the transient perspective.

• 대한전자공학회논문지TC
• /
• 제43권3호
• /
• pp.49-58
• /
• 2006

본 논문에서는 시변 주파수 선택적 페이딩 채널에서 OFDM (orthogonal frequency division multiplexing) 시스템을 위한 curve-fitting 채널추정 방법을 제안한다. 제안된 방법은 시간영역 및 주파수영역에서 1차원 curve-fitting을 통하여 smoothing과 interpolation을 순차적으로 수행함으로써 채널추정 정확도를 크게 개선할 수 있다. 먼저, 파일럿 심벌들을 사용하여 LS(least-Square) 추정치를 구하고, 이를 바탕으로 파일럿 밀도가 상대적으로 높은 영역을 선택하여 최소자승오차 기준에 따라 적절한 차수의 다항식으로 1차원 curve-fitting을 수행한다. 다음으로, 이 다항식을 이용하여 주어진 범위 내에 존재하는 LS 추정치들을 smoothing하고, interpolation 또는 prediction을 통하여 데이터 전송을 위하여 필요한 채널추정치들을 구한다. 이어서, 선택된 영역에서 얻어진 채널추정치들을 나머지 영역에서 또 다른 다항식을 사용하여 동일한 과정으로 1차원 curve-fitting을 통하여 smoothing과 interpolation을 수행함으로써 시간영역 및 주파수영역에서의 채널추정을 완료한다. 모의실험을 통하여 다양한 채널환경에서 MSE (mean square error) 및 BER (bit error rate) 성능을 분석한 결과, 제안된 방법이 기존의 채널추정 방법들에 비하여 월등히 우수하며, 최적의 Wiener 필터링 방법보다도 우수함을 보였다.

• 한국컴퓨터정보학회논문지
• /
• 제15권5호
• /
• pp.75-81
• /
• 2010

적응 배열 안테나 시스템은 간섭신호 방향으로 영점을 형성하여 표적신호를 추정한다. 그러나 적응 배열 안테나 시스템에 상관성 간섭신호가 입사된다면 표적신호를 간섭신호로 간주하여 시스템의 성능이 급격히 감소되고, 지향 오차에 민감하여 주빔이 표적신호의 입사방향을 지향하지 못한다. 본 논문에서는 고차제한조건과 지향오차수정방법을 결합하여 빔형성 알고리즘을 제안하였다. 그리고 제안한 알고리즘으로 자유도 손실을 최소화하고 간섭신호와 지향오차를 감소시켰다. 모의실험을 이용하여 제안 알고리즘과 기존 알고리즘의 성능을 비교 분석하였다. 본 논문에서 제안한 알고리즘은 신호 입사각을 정확히 추정하지만 기존의 알고리즘은 약 $0.8^{\circ}$의 오차가 발생하였다.

• Journal of the Korean Data and Information Science Society
• /
• 제19권1호
• /
• pp.25-36
• /
• 2008

Error Correcting Output Codes (ECOC) can improve generalization performance when applied to multicategory classification problem. In this study we propose a new criterion to select hyperparameters included in ECOC scheme. Instead of margins of a data we propose to use the probability of misclassification error since it makes the criterion simple. Using this we obtain an upper bound of leave-one-out error of OVA(one vs all) method. Our experiments from real and synthetic data indicate that the bound leads to good estimates of parameters.

• 제어로봇시스템학회:학술대회논문집
• /
• 제어로봇시스템학회 2000년도 제15차 학술회의논문집
• /
• pp.82-82
• /
• 2000

The way which estimates a position from Navigation is Dead-reckoning Navigation and Radio Navigation. Generally dead-reckoning navigation uses Gyro and odometer as sensor, but these have problems which are an integrating noise and a noise-sensitivity. Gps which is used by radio-navigation has a Troposthetic error and MultiPath ewer and so on. For minimizing a Troposthetic error and a Multipath error, this paper suggests to an algorithm which use vanishing point on CCD camera.

• 한국전산응용수학회:학술대회논문집
• /
• 한국전산응용수학회 2002년도 학술발표대회
• /
• pp.16.1-16
• /
• 2002

• 대한수학회보
• /
• 제39권4호
• /
• pp.705-714
• /
• 2002

We consider a jump-diffusion model generated by a Levy process for an asset price. We present an error estimate for the option prices between the jump-diffusion model and the Black-scholes model when the former converges weakly to the latter.

• Communications for Statistical Applications and Methods
• /
• 제6권3호
• /
• pp.667-676
• /
• 1999

We investigate the effects of omission of initial observations in each individuals in the panel data regression model when the disturbances follow a serially correlated one way error components. We show that the first transformed observation can have a relative large hat matrix diagonal component and a large influence on parameter estimates when the correlation coefficient is large in absolute value.