• KIPS Transactions on Software and Data Engineering
• /
• v.4 no.3
• /
• pp.143-150
• /
• 2015

As smart devices get popular, research on gesture recognition using their embedded-accelerometer draw attention. As Dynamic Time Warping(DTW), recently, has been used to perform gesture recognition on data sequence from accelerometer, in this paper we propose Feature-Strengthened Gesture Recognition(FsGr) Model which can improve the recognition success rate when DTW is used. FsGr model defines feature-strengthened parts of data sequences to similar gestures which might produce unsuccessful recognition, and performs additional DTW on them to improve the recognition rate. In training phase, FsGr model identifies sets of similar gestures, and analyze features of gestures per each set. During recognition phase, it makes additional recognition attempt based on the result of feature analysis to improve the recognition success rate, when the result of first recognition attempt belongs to a set of similar gestures. We present the performance result of FsGr model, by experimenting the recognition of lower case alphabets.

• Structural Engineering and Mechanics
• /
• v.56 no.4
• /
• pp.549-567
• /
• 2015

A first-order moment method (FORM) reliability analysis is commonly used for structural stability analysis. It requires the values and partial derivatives of the performance to function with respect to the random variables for the design. These calculations can be cumbersome when the performance functions are implicit. A Gaussian process (GP)-based response surface is adopted in this study to approximate the limit state function. By using a trained GP model, a large number of values and partial derivatives of the performance functions can be obtained for conventional reliability analysis with a FORM, thereby reducing the number of stability analysis calculations. This dynamic renewed knowledge source can provide great assistance in improving the predictive capacity of GP during the iterative process, particularly from the view of machine learning. An iterative algorithm is therefore proposed to improve the precision of GP approximation around the design point by constantly adding new design points to the initial training set. Examples are provided to illustrate the GP-based response surface for both structural and non-structural reliability analyses. The results show that the proposed approach is applicable to structural reliability analyses that involve implicit performance functions and structural response evaluations that entail time-consuming finite element analyses.

• Journal of Biomedical Engineering Research
• /
• v.38 no.5
• /
• pp.237-241
• /
• 2017

Early detection of the pulmonary nodule is important for diagnosis and treatment of lung cancer. Recently, CT has been used as a screening tool for lung nodule detection. And, it has been reported that computer aided detection(CAD) systems can improve the accuracy of the radiologist in detection nodules on CT scan. The previous study has been proposed a method using Convolutional Neural Network(CNN) in Lung CAD system. But the proposed model has a limitation in accuracy due to its sparse layer structure. Therefore, we propose a Deep Convolutional Neural Network to overcome this limitation. The model proposed in this work is consist of 14 layers including 8 convolutional layers and 4 fully connected layers. The CNN model is trained and tested with 61,404 regions-of-interest (ROIs) patches of lung image including 39,760 nodules and 21,644 non-nodules extracted from the Lung Image Database Consortium(LIDC) dataset. We could obtain the classification accuracy of 91.79% with the CNN model presented in this work. To prevent overfitting, we trained the model with Augmented Dataset and regularization term in the cost function. With L1, L2 regularization at Training process, we obtained 92.39%, 92.52% of accuracy respectively. And we obtained 93.52% with data augmentation. In conclusion, we could obtain the accuracy of 93.75% with L2 Regularization and Data Augmentation.

• Progress in Medical Physics
• /
• v.30 no.4
• /
• pp.94-103
• /
• 2019

Purpose: To evaluate the clinical feasibility of knowledge-based planning (KBP) for volumetric-modulated arc radiotherapy (VMAT) in spine stereotactic body radiotherapy (SBRT). Methods: Forty-eight VMAT plans for spine SBRT was studied. Two planning target volumes (PTVs) were defined for simultaneous integrated boost: PTV for boost (PTV-B: 27 Gy/3fractions) and PTV elective (PTV-E: 24 Gy/3fractions). The expert VMAT plans were manually generated by experienced planners. Twenty-six plans were used to train the KBP model using Varian RapidPlan. With the trained KBP model each KBP plan was automatically generated by an individual with little experience and compared with the expert plan (closed-loop validation). Twenty-two plans that had not been used for KBP model training were also compared with the KBP results (open-loop validation). Results: Although the minimal dose of PTV-B and PTV-E was lower and the maximal dose was higher than those of the expert plan, the difference was no larger than 0.7 Gy. In the closed-loop validation, D_1.2cc, D_0.35cc, and D_mean of the spinal cord was decreased by 0.9 Gy, 0.6 Gy, and 0.9 Gy, respectively, in the KBP plans (P<0.05). In the open-loop validation, only D_mean of the spinal cord was significantly decreased, by 0.5 Gy (P<0.05). Conclusions: The dose coverage and uniformity for PTV was slightly worse in the KBP for spine SBRT while the dose to the spinal cord was reduced, but the differences were small. Thus, inexperienced planners could easily generate a clinically feasible plan for spine SBRT by using KBP.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.38A no.6
• /
• pp.486-491
• /
• 2013

This paper proposes a fast and efficient Haar-like feature selection algorithm for training classifier used in object detection. Many features selected by Haar-like feature selection algorithm and existing AdaBoost algorithm are either similar in shape or overlapping due to considering only feature's error rate. The proposed algorithm calculates similarity of features by their shape and distance between features. Fast and efficient feature selection is made possible by removing selected features and features with high similarity from feature set. FERET face database is used to compare performance of classifiers trained by previous algorithm and proposed algorithm. Experimental results show improved performance comparing classifier trained by proposed method to classifier trained by previous method. When classifier is trained to show same performance, proposed method shows 20% reduction of features used in classification.

• The Korean Journal of Applied Statistics
• /
• v.32 no.6
• /
• pp.899-908
• /
• 2019

Massive in-depth neural networks with numerous parameters are powerful machine learning methods, but they have overfitting problems due to the excessive flexibility of the models. Dropout is one methods to overcome the problem of oversized neural networks. It is also an effective method that randomly drops input and hidden nodes from the neural network during training. Every sample is fed to a thinned network from an exponential number of different networks. In this study, instead of feeding one sample for each thinned network, two or more samples are used in fitting for one thinned network known as a Hybrid Dropout. Simulation results using real data show that the new method improves the stability of estimates and reduces the minimum error for the verification data.

• Journal of KIISE
• /
• v.43 no.8
• /
• pp.902-909
• /
• 2016

Question answering system (QA system) is a system that finds an actual answer to the question posed by a user, whereas a typical search engine would only find the links to the relevant documents. Recent works related to the open domain QA systems are receiving much attention in the fields of natural language processing, artificial intelligence, and data mining. However, the prior works on QA systems simply replace all words that are not in the training data with a single token, even though such unseen words are likely to play crucial roles in differentiating the candidate answers from the actual answers. In this paper, we propose a method to compute vectors of such unseen words by taking into account the context in which the words have occurred. Next, we also propose a model which utilizes inverse document frequencies (IDF) to efficiently process unseen words by expanding the system's vocabulary. Finally, we validate that the proposed method and model improve the performance of a QA system through experiments.

• Journal of KIISE
• /
• v.43 no.8
• /
• pp.927-932
• /
• 2016

With the explosive increase in the number of online medical documents, the demand for question-answering systems is increasing. Recently, question-answering models based on machine learning have shown high performances in various domains. However, many question-answering models within the medical domain are still based on information retrieval techniques because of sparseness of training data. Based on various information retrieval techniques, we propose an answer snippet retrieval model for question-answering systems of medical documents. The proposed model first searches candidate answer sentences from medical documents using a cluster-based retrieval technique. Then, it generates reliable answer snippets using a re-ranking model of the candidate answer sentences based on various sentence retrieval techniques. In the experiments with BioASQ 4b, the proposed model showed better performances (MAP of 0.0604) than the previous models.

• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.59 no.2
• /
• pp.436-444
• /
• 2010

In this study, we introduce and discuss a concept of a granular-oriented radial basis function neural networks (GRBF NNs). In contrast to the typical architectures encountered in radial basis function neural networks(RBF NNs), our main objective is to develop a design strategy of GRBF NNs as follows : (a) The architecture of the network is fully reflective of the structure encountered in the training data which are granulated with the aid of clustering techniques. More specifically, the output space is granulated with use of K-Means clustering while the information granules in the multidimensional input space are formed by using a so-called context-based Fuzzy C-Means which takes into account the structure being already formed in the output space, (b) The innovative development facet of the network involves a dynamic reduction of dimensionality of the input space in which the information granules are formed in the subspace of the overall input space which is formed by selecting a suitable subset of input variables so that the this subspace retains the structure of the entire space. As this search is of combinatorial character, we use the technique of genetic optimization to determine the optimal input subspaces. A series of numeric studies exploiting some nonlinear process data and a dataset coming from the machine learning repository provide a detailed insight into the nature of the algorithm and its parameters as well as offer some comparative analysis.

• The KIPS Transactions:PartB
• /
• v.13B no.7 s.110
• /
• pp.653-662
• /
• 2006

In ubiquitous computing that is to build computing environments to provide proper services according to user's context, human being's emotion recognition based on facial expression is used as essential means of HCI in order to make man-machine interaction more efficient and to do user's context-awareness. This paper addresses a problem of rigidly basic emotion recognition in context-sensitive facial expressions through a new Bayesian classifier. The task for emotion recognition of facial expressions consists of two steps, where the extraction step of facial feature is based on a color-histogram method and the classification step employs a new Bayesian teaming algorithm in performing efficient training and test. New context-sensitive Bayesian learning algorithm of EADF(Extended Assumed-Density Filtering) is proposed to recognize more exact emotions as it utilizes different classifier complexities for different contexts. Experimental results show an expression classification accuracy of over 91% on the test database and achieve the error rate of 10.6% by modeling facial expression as hidden context.