• 한국음향학회지
• /
• 제23권6호
• /
• pp.473-480
• /
• 2004

본 논문에서는 잡음 환경에서 강인한 음성 인식을 위하여 음성 모델을 기반으로 하는 효과적인 특징 보상 기법을 제안한다. 제안하는 특징 보상 기법은 병렬 결합된 혼합 모델 (PCMM)을 기반으로 한다. 기존의 PCMM 기반의 기법은 시간에 따라 변하는 잡음 환경을 반영하기 위하여 매 음성 입력마다 복잡한 과정의 혼합 모델 결합이 필요하다. 제안하는 기법에서는 다중의 혼합 모델을 보간하는 방법을 채용함으로써 시간에 따라 변하는 배경 잡음에 대응할 수 있다. 보다 신뢰성 있는 혼합 모델 생성을 위하여 데이터 유도 기반의 방법을 도입하고, 실시간 처리를 위하여 프레임에 동기화된 환경 사후 확률 예측 과정을 제안한다. 다중 모델로 인한 연산량 증가를 막기 위하여 혼합 모델을 공유하는 기법을 제안한다. 가우시안 혼합 모델 사이에 통계학적으로 유사한 요소들을 선택하여 공유에 필요한 공통 모델을 생성한다. Aurora 2.0 데이터베이스와 실제 자동차 주행 환경에서 수집된 음성 데이터베이스에 대한 성능 평가를 실시한다. 실험 결과로부터 제안한 기법이 모의 환경과 실제 잡음 환경에서 강인한 음성 인식 성능을 가져오고 연산량 감소에 효과적임을 확인한다.

• 한국정보시스템학회지:정보시스템연구
• /
• 제26권3호
• /
• pp.63-90
• /
• 2017

Purpose This study proposes a novel system trading model using case-based reasoning (CBR) based on absolute similarity threshold. The proposed model is designed to optimize the absolute similarity threshold, feature selection, and instance selection of CBR by using genetic algorithm (GA). With these mechanisms, it enables us to yield higher returns from stock market trading. Design/Methodology/Approach The proposed CBR model uses the absolute similarity threshold varying from 0 to 1, which serves as a criterion for selecting appropriate neighbors in the nearest neighbor (NN) algorithm. Since it determines the nearest neighbors on an absolute basis, it fails to select the appropriate neighbors from time to time. In system trading, it is interpreted as the signal of 'hold'. That is, the system trading model proposed in this study makes trading decisions such as 'buy' or 'sell' only if the model produces a clear signal for stock market prediction. Also, in order to improve the prediction accuracy and the rate of return, the proposed model adopts optimal feature selection and instance selection, which are known to be very effective in enhancing the performance of CBR. To validate the usefulness of the proposed model, we applied it to the index trading of KOSPI200 from 2009 to 2016. Findings Experimental results showed that the proposed model with optimal feature or instance selection could yield higher returns compared to the benchmark as well as the various comparison models (including logistic regression, multiple discriminant analysis, artificial neural network, support vector machine, and traditional CBR). In particular, the proposed model with optimal instance selection showed the best rate of return among all the models. This implies that the application of CBR with the absolute similarity threshold as well as the optimal instance selection may be effective in system trading from the perspective of returns.

• 응용통계연구
• /
• 제34권2호
• /
• pp.153-166
• /
• 2021

약물유전체학 연구의 주요 목표는 고차원의 유전 변수를 기반으로 개인의 약물 반응성을 예측하는 것이다. 변수의 개수가 많기 때문에 변수의 개수를 줄이기 위해서는 변수 선택이 필요하며, 선택된 변수들은 머신러닝 알고리즘을 사용하여 예측 모델을 구축하는데 사용된다. 본 연구에서는 400명의 뇌전증 환자의 차세대 염기서열 분석 데이터에 로지스틱 회귀, ReliefF, TurF, 랜덤 포레스트, LASSO의 조합과 같은 여러 가지 혼합 변수 선택 방법을 적용하였다. 선택된 변수들에 랜덤포레스트, 그래디언트 부스팅, 서포트벡터머신을 포함한 머신러닝 방법들을 적용했고 스태킹을 통해 앙상블 모형을 구축하였다. 본 연구의 결과는 랜덤포레스트와 ReliefF의 혼합 변수 선택 방법을 이용한 스태킹 모형이 다른 모형보다 더 좋은 성능을 보인다는 것을 보여주었다. 5-폴드 교차 검증을 기반으로 하여 적합한 최적 모형의 평균 검증 정확도는 0.727이고 평균 검증 AUC 값은 0.761로 나타났다. 또한, 동일한 변수를 사용할 때 스태킹 모델이 단일 머신러닝 예측 모델보다 성능이 우수한 것으로 나타났다.

• 한국물환경학회지
• /
• 제38권6호
• /
• pp.292-305
• /
• 2022

The Yeongsan River, one of the four largest rivers in South Korea, has been facing difficulties with water quality management with respect to algal bloom. The algal bloom menace has become bigger, especially after the construction of two weirs in the mainstream of the Yeongsan River. Therefore, the prediction and factor analysis of Chlorophyll-a (Chl-a) concentration is needed for effective water quality management. In this study, Chl-a prediction model was developed, and the performance evaluated using machine and deep learning methods, such as Deep Neural Network (DNN), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost). Moreover, the correlation analysis and the feature importance results were compared to identify the major factors affecting the concentration of Chl-a. All models showed high prediction performance with an R² value of 0.9 or higher. In particular, XGBoost showed the highest prediction accuracy of 0.95 in the test data.The results of feature importance suggested that Ammonia (NH₃-N) and Phosphate (PO₄-P) were common major factors for the three models to manage Chl-a concentration. From the results, it was confirmed that three machine learning methods, DNN, RF, and XGBoost are powerful methods for predicting water quality parameters. Also, the comparison between feature importance and correlation analysis would present a more accurate assessment of the important major factors.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제17권12호
• /
• pp.3383-3397
• /
• 2023

Scene graphs serve as semantic abstractions of images and play a crucial role in enhancing visual comprehension and reasoning. However, the performance of Scene Graph Generation is often compromised when working with biased data in real-world situations. While many existing systems focus on a single stage of learning for both feature extraction and classification, some employ Class-Balancing strategies, such as Re-weighting, Data Resampling, and Transfer Learning from head to tail. In this paper, we propose a novel approach that decouples the feature extraction and classification phases of the scene graph generation process. For feature extraction, we leverage a transformer-based architecture and design an adaptive calibration function specifically for predicate classification. This function enables us to dynamically adjust the classification scores for each predicate category. Additionally, we introduce a Distribution Alignment technique that effectively balances the class distribution after the feature extraction phase reaches a stable state, thereby facilitating the retraining of the classification head. Importantly, our Distribution Alignment strategy is model-independent and does not require additional supervision, making it applicable to a wide range of SGG models. Using the scene graph diagnostic toolkit on Visual Genome and several popular models, we achieved significant improvements over the previous state-of-the-art methods with our model. Compared to the TDE model, our model improved mR@100 by 70.5% for PredCls, by 84.0% for SGCls, and by 97.6% for SGDet tasks.

• 한국정밀공학회지
• /
• 제30권6호
• /
• pp.601-606
• /
• 2013

3D scanning is a technique to measure the 3D shape information of the object. Shape information obtained by 3D scanning is expressed either as point cloud or as polygon mesh type data that can be widely used in various areas such as reverse engineering and quality inspection. 3D scanning should be performed as accurate as possible since the scanned data is highly required to detect the features on an object in order to scan the shape of the object more precisely. In this study, we propose the method on finding the location of feature more accurately, based on the extended Biplane SNAKE with global optimization. In each iteration, we project the feature lines obtained by the extended Biplane SNAKE into each image plane and move the feature lines to the features on each image. We have applied this approach to real models to verify the proposed optimization algorithm.

• 한국멀티미디어학회논문지
• /
• 제12권1호
• /
• pp.50-57
• /
• 2009

본 논문은 3차원 다각형 모델에서 특징 선을 추출하기 위한 방법에 대해 제안한다. 이산 곡면으로 이루어진 다각형 모델에서 특징 선을 추출하기 위하여 기존 방법에서는 전역적인 음함수 곡면 맞춤 기법(Implicit Surface Fitting)을 이용하여 모델의 꼭지점에서 곡률과 곡률 미분 값을 측정하였다. 이러한 방법은 다각형 모델의 꼭지점에서 음함수 곡면으로 정확하게 투영할 수 있도록 사용자의 정의 파라미타를 찾아야 하며, 특징 추출을 위한 많은 계산 시간을 요구한다. 그러나 제안 방법은 지역적 음함수 곡면 맞춤 기법을 이용하여 모델의 꼭지점에 근사된 곡면을 통해 미분 정보를 측정한다. 측정된 미분 정보를 통해 쉽게 각각의 모서리에서 제로-클로싱을 통해 특징 점을 추출하고, 곡률 방향을 따라 추출된 점들을 연결하여 특징 선을 생성한다. 여러 가지 다각형 모델에서 실험을 하였고 기존 방법보다 빠르며 높은 품질의 특징 선을 추출한다.

• 대한기계학회논문집A
• /
• 제24권9호
• /
• pp.2235-2245
• /
• 2000

The necessity of using rapid prototyping(RP) for short-run manufacturing is continuously driving a development of a cost-effective technique that will produce completely-finished quality parts in a very short time. To meet these demands, the improvements in production speed, accuracy, materials, aid cost are crucial. Thus, a new hybrid-RP system performing both deposition and machining in a station is proposed. For the new hybrid RP process to maintain the same degree of process automation as in currently available processes like SLA or FDNI, a sophisticated process planning system is developed. In the process planner, CAD models(STEP AP203) are partitioned into 3D manufacturable volumes called 'Ueposition feature segment"(DFS) after machining features called "machining feature segmenf'(MFS) are extracted from the initial CAD model. Once MFS and DFS are identified, the process planner arranges them into a chain of processes and automatically generates machining information for each DFS and MFS. The goal of this paper is to present a framework for a process planning system for hybrid RP processes and to outline the geometric algorithms involved in developing such an environment.

• Structural Engineering and Mechanics
• /
• 제54권5호
• /
• pp.891-908
• /
• 2015

For a systematic study on wind-induced vibration characteristics of large hyperbolic cooling towers with different feature sizes, the pressure measurement tests are finished on the rigid body models of three representative cooling towers with the height of 155 m, 177 m and 215 m respectively. Combining the refined frequency-domain algorithm of wind-induced responses, the wind-induced average response, resonant response, background response, coupling response and wind vibration coefficients of large cooling towers with different feature sizes are obtained. Based on the calculating results, the parametric analysis on wind-induced vibration of cooling towers is carried out, e.g. the feature sizes, damping ratio and the interference effect of surrounding buildings. The discussion shows that the increase of feature sizes makes wind-induced average response and fluctuating response larger correspondingly, and the proportion of resonant response also gradually increased, but it has little effect on the wind vibration coefficient. The increase of damping ratio makes resonant response and the wind vibration coefficient decreases obviously, which brings about no effect on average response and background response. The interference effect of surrounding buildings makes the fluctuating response and wind vibration coefficient increased significantly, furthermore, the increase ranges of resonant response is greater than background response.

• 한국멀티미디어학회논문지
• /
• 제14권4호
• /
• pp.504-516
• /
• 2011

In this paper, we will present a method to detect human hand and recognize hand gesture. For detecting the hand region, we use the feature of human skin color and hand feature (with boundary complexity) to detect the hand region from the input image; and use algorithm of optical flow to track the hand movement. Hand gesture recognition is composed of two parts: 1. Posture recognition and 2. Motion recognition, for describing the hand posture feature, we employ the Fourier descriptor method because it's rotation invariant. And we employ PCA method to extract the feature among gesture frames sequences. The HMM method will finally be used to recognize these feature to make a final decision of a hand gesture. Through the experiment, we can see that our proposed method can achieve 99% recognition rate at environment with simple background and no face region together, and reduce to 89.5% at the environment with complex background and with face region. These results can illustrate that the proposed algorithm can be applied as a production.