• Journal of Information Processing Systems
• /
• 제10권3호
• /
• pp.443-458
• /
• 2014

An extreme learning machine (ELM) is a recently proposed learning algorithm for a single-layer feed forward neural network. In this paper we studied the ensemble of ELM by using a bagging algorithm for facial expression recognition (FER). Facial expression analysis is widely used in the behavior interpretation of emotions, for cognitive science, and social interactions. This paper presents a method for FER based on the histogram of orientation gradient (HOG) features using an ELM ensemble. First, the HOG features were extracted from the face image by dividing it into a number of small cells. A bagging algorithm was then used to construct many different bags of training data and each of them was trained by using separate ELMs. To recognize the expression of the input face image, HOG features were fed to each trained ELM and the results were combined by using a majority voting scheme. The ELM ensemble using bagging improves the generalized capability of the network significantly. The two available datasets (JAFFE and CK+) of facial expressions were used to evaluate the performance of the proposed classification system. Even the performance of individual ELM was smaller and the ELM ensemble using a bagging algorithm improved the recognition performance significantly.

• Journal of Information Processing Systems
• /
• 제19권5호
• /
• pp.673-687
• /
• 2023

Convolutional neural network-based deep learning technology is the most commonly used in image identification, but it requires large-scale data for training. Therefore, application in specific fields in which data acquisition is limited, such as in the military, may be challenging. In particular, the identification of ground weapon systems is a very important mission, and high identification accuracy is required. Accordingly, various studies have been conducted to achieve high performance using small-scale data. Among them, the ensemble method, which achieves excellent performance through the prediction average of the pre-trained models, is the most representative method; however, it requires considerable time and effort to find the optimal combination of ensemble models. In addition, there is a performance limitation in the prediction results obtained by using an ensemble method. Furthermore, it is difficult to obtain the ensemble effect using models with imbalanced classification accuracies. In this paper, we propose a transfer learning-based feature fusion technique for heterogeneous models that extracts and fuses features of pre-trained heterogeneous models and finally, fine-tunes hyperparameters of the fully connected layer to improve the classification accuracy. The experimental results of this study indicate that it is possible to overcome the limitations of the existing ensemble methods by improving the classification accuracy through feature fusion between heterogeneous models based on transfer learning.

• Journal of the Korean Data and Information Science Society
• /
• 제27권2호
• /
• pp.549-557
• /
• 2016

In this paper we propose multiclassification method for large data sets by ensembling least squares support vector machines (LS-SVM) with principal components instead of raw input vector. We use the revised one-vs-all method for multiclassification, which is one of voting scheme based on combining several binary classifications. The revised one-vs-all method is performed by using the hat matrix of LS-SVM ensemble, which is obtained by ensembling LS-SVMs trained using each random sample from the whole large training data. The leave-one-out cross validation (CV) function is used for the optimal values of hyper-parameters which affect the performance of multiclass LS-SVM ensemble. We present the generalized cross validation function to reduce computational burden of leave-one-out CV functions. Experimental results from real data sets are then obtained to illustrate the performance of the proposed multiclass LS-SVM ensemble.

• 한국정보과학회논문지:소프트웨어및응용
• /
• 제32권12호
• /
• pp.1229-1237
• /
• 2005

분류 성능을 향상시키기 위해서 다수의 분류기들을 결합하는 연구가 활발히 진행되고 있다. 우수한 앙상블 분류기를 회득하기 위해서는 정확하고 다양한 개별 분류기를 구축해야 한다. 기존에는 Bagging이나 Boosting 등의 앙상블 학습 기법을 이용하거나 획득된 개별 분류기의 학습 데이타에 대한 다양성을 측정하였지만 유전 발현 데이타와 같이 학습 데이타가 적은 경우 한계가 있다. 본 논문에서는 유전자 프로그래밍으로부터 획득된 규칙의 구조적 다양성을 분석하여 결합하는 앙상블 기법을 제안한다. 유전자 프로그래밍으로 해석 가능한 분류 규칙을 생성하고 그들 사이의 다양성을 측정한 뒤, 이들 중 다양한 규칙의 집합을 결합하여 분류를 수행한다. 유전 발현 데이타로부터 림프종 암, 폐 암, 난소 암 등을 분류하는 문제를 대상으로 실험하여 제안하는 방법의 유용성을 검증하였다. 앙상블 시 분류 규칙 사이의 다양성을 분석하여 결합한 결과, 다양성을 고려하지 않을 때보다 높은 분류 성능을 획득하였고, 개별 분류 규칙들 사이의 다양성에 따라서 정분류율이 증가하는 것도 확인하였다.

• 대한산업공학회지
• /
• 제28권1호
• /
• pp.112-127
• /
• 2002

A useful pattern is a pattern that contributes much to learning. For a classification problem those patterns near the class boundary surfaces carry more information to the classifier. For a regression problem the ones near the estimated surface carry more information. In both cases, the usefulness is defined only for those patterns either without error or with negligible error. Using only the useful patterns gives several benefits. First, computational complexity in memory and time for learning is decreased. Second, overfitting is avoided even when the learner is over-sized. Third, learning results in more stable learners. In this paper, we propose a pattern 'utility index' that measures the utility of an individual pattern. The utility index is based on the bias and variance of a pattern trained by a network ensemble. In classification, the pattern with a low bias and a high variance gets a high score. In regression, on the other hand, the one with a low bias and a low variance gets a high score. Based on the distribution of the utility index, the original training set is divided into a high-score group and a low-score group. Only the high-score group is then used for training. The proposed method is tested on synthetic and real-world benchmark datasets. The proposed approach gives a better or at least similar performance.

• 한국군사과학기술학회지
• /
• 제23권2호
• /
• pp.139-146
• /
• 2020

The camouflage pattern was difficult to distinguish from the surrounding background, so it was difficult to classify the object and the background image when the color image is used as the training data of deep-learning. In this paper, we proposed a red-green color blindness image transformation method using the principle that people of red-green blindness distinguish green color better than ordinary people. Experimental results show that the camouflage soldier's recognition performance improved by proposed a deep learning model of the ensemble technique using the imitated red-green-blind image data and the original color image data.

• 한국음향학회지
• /
• 제42권4호
• /
• pp.305-312
• /
• 2023

기포 크기 분포를 음파 감쇄 손실을 이용하여 역산하기 위해 Physics-Informed Neural Network(PINN)을 사용하였다. 역산에 사용되는 선형시스템을 풀기 위해 이미지 처리 분야에서 선형시스템 문제를 해결한 Adaptive Learned Iterative Shrinkage Thresholding Algorithm(Ada-LISTA)를 PINN의 신경망 구조로 이용하였다. 더 나아가, PINN의 손실함수에 선형시스템 기반의 정규항을 포함함으로써 PINN의 해가 기포 물리 법칙을 만족하여 더 높은 일반화 성능을 가지도록 하였다. 그리고 기포 추정값의 불확실성을 계산하기 위해 딥앙상블 기법을 이용하였다. 서로 다른 초기값을 갖는 20개의 Ada-LISTA는 같은 훈련데이터를 이용하여 학습되었다. 이 후 테스트시 훈련데이터와 다른 경향의 감쇄 손실을 입력으로 사용하여 기포 크기 분포를 추정하였고, 추정값과 이에 대한 불확실성을 20개 추정값의 평균과 분산으로 각각 구하였다. 그 결과 딥앙상블이 적용된 Ada-LISTA는 기존 볼록 최적화 기법인 CVX보다 기포 크기 분포를 역산하는데 더 우수한 성능을 보였다.

• 한국전자거래학회지
• /
• 제25권3호
• /
• pp.1-13
• /
• 2020

기계학습(machine learning)이란 주어진 데이터에 대한 일반화 과정으로부터 특정 문제를 해결할 수 있는 모델(model) 생성 기술을 의미한다. 우수한 성능의 모델을 생성하기 위해서는 양질의 학습데이터와 일반화 과정을 위한 학습 알고리즘이 준비되어야 한다. 성능 개선을 위한 한 가지 방법으로서 앙상블(Ensemble) 기법은 단일 모델(single model)을 생성하기보다 다중 모델을 생성하며, 이는 배깅(Bagging), 부스팅(Boosting), 스태킹(Stacking) 학습 기법을 포함한다. 본 논문은 기존 스태킹 기법을 개선한 다중 스태킹 앙상블(Multiple Stacking Ensemble) 학습 기법을 제안한다. 다중 스태킹 앙상블 기법의 학습 구조는 딥러닝 구조와 유사하고 각 레이어가 스태킹 모델의 조합으로 구성되며 계층의 수를 증가시켜 각 계층의 오분류율을 최소화하여 성능을 개선한다. 4가지 유형의 데이터셋을 이용한 실험을 통해 제안 기법이 기존 기법에 비해 분류 성능이 우수함을 보인다.

• 대한조선학회논문집
• /
• 제56권6호
• /
• pp.488-496
• /
• 2019

The estimation of block erection work time at a dock is one of the important factors when establishing or managing the total shipbuilding schedule. In order to predict the work time, it is a natural approach that the existing block erection data would be used to solve the problem. Generally the work time per unit is the product of coefficient value, quantity, and product value. Previously, the work time per unit is determined statistically by unit load data. However, we estimate the work time per unit through work time coefficient value from series ships using machine learning. In machine learning, the outcome depends mainly on how the training data is organized. Therefore, in this study, we use 'Feature Engineering' to determine which one should be used as features, and to check their influence on the result. In order to get the coefficient value of each block, we try to solve this problem through the Ensemble learning methods which is actively used nowadays. Among the many techniques of Ensemble learning, the final model is constructed by Stacking Ensemble techniques, consisting of the existing Ensemble models (Decision Tree, Random Forest, Gradient Boost, Square Loss Gradient Boost, XG Boost), and the accuracy is maximized by selecting three candidates among all models. Finally, the results of this study are verified by the predicted total work time for one ship among the same series.

• 한국정보과학회논문지:소프트웨어및응용
• /
• 제30권3_4호
• /
• pp.251-258
• /
• 2003

Support Vector Machine(SVM)은 이론상으로 좋은 일반화 성능을 보이지만, 실제적으로 구현된 SVM은 이론적인 성능에 미치지 못한다. 주 된 이유는 시간, 공간상의 높은 복잡도로 인해 근사화된 알고리듬으로 구현하기 때문이다. 본 논문은 SVM의 분류성능을 향상시키기 위해 Bagging(Bootstrap aggregating)과 Boosting을 이용한 SVM 앙상블 구조의 구성을 제안한다. SVM 앙상블의 학습에서 Bagging은 각각의 SVM의 학습데이타는 전체 데이타 집합에서 임의적으로 일부 추출되며, Boosting은 SVM 분류기의 에러와 연관된 확률분포에 따라 학습데이타를 추출한다. 학습단계를 마치면 다수결 (Majority voting), 최소자승추정법(LSE:Least Square estimation), 2단계 계층적 SVM등의 기법에 개개의 SVM들의 출력 값들이 통합되어진다. IRIS 분류, 필기체 숫자인식, 얼굴/비얼굴 분류와 같은 여러 실험들의 결과들은 제안된 SVM 앙상블의 분류성능이 단일 SVM보다 뛰어남을 보여준다.