• Journal of Korea Multimedia Society
• /
• v.23 no.1
• /
• pp.1-7
• /
• 2020

Clustering methods essentially take a two-step approach; extracting feature vectors for dimensionality reduction and then employing clustering algorithm on the extracted feature vectors. However, for clustering images, the traditional clustering methods such as stacked auto-encoder based k-means are not effective since they tend to ignore the local information. In this paper, we propose a method first to effectively reduce data dimensionality using convolutional auto-encoder to capture and reflect the local information and then to accurately cluster similar data samples by using a hierarchical clustering approach. The experimental results confirm that the clustering results are improved by using the proposed model in terms of clustering accuracy and normalized mutual information.

• Journal of the Korean Data and Information Science Society
• /
• v.25 no.5
• /
• pp.1069-1077
• /
• 2014

A clubfoot is a kind of congenital deformity of foot, which is internally rotated at the ankle. In this paper, we are going to cluster the curves of relative differences between regular and operated feet. Since these curves are irregular and sparsely sampled, general clustering models could not be applied. So the clustering model for sparsely sampled functional data by James and Sugar (2003) are applied and parameters are estimated using EM algorithm. The number of clusters is determined by the distortion function (Sugar and James, 2003) and two clusters of the curves are found.

• Communications for Statistical Applications and Methods
• /
• v.29 no.6
• /
• pp.695-708
• /
• 2022

Recently, several studies have been conducted using state space model. In this study, a dynamic linear model with state space model form is applied to stock data. The monthly returns for 135 Korean stocks are fitted to a dynamic linear model, to obtain an estimate of the time-varying 𝛽-coefficient time-series. The model formula used for the return is a capital asset pricing model formula explained in economics. In particular, the transition equation of the state space model form is appropriately modified to satisfy the assumptions of the error term. k-shape clustering is performed to classify the 135 estimated 𝛽 time-series into several groups. As a result of the clustering, four clusters are obtained, each consisting of approximately 30 stocks. It is found that the distribution is different for each group, so that it is well grouped to have its own characteristics. In addition, a common pattern is observed for each group, which could be interpreted appropriately.

• Journal of Digital Convergence
• /
• v.18 no.3
• /
• pp.189-194
• /
• 2020

Recently, various clustering techniques have been studied to efficiently handle data generated by heterogeneous IoT devices. However, existing clustering techniques are not suitable for mobile IoT devices because they focus on statically dividing networks. This paper proposes a probabilistic deep learning-based dynamic clustering model for collecting and analyzing information on IoT devices using edge networks. The proposed model establishes a subnet by applying the frequency of the attribute values collected probabilistically to deep learning. The established subnets are used to group information extracted from seeds into hierarchical structures and improve the speed and accuracy of dynamic clustering for IoT devices. The performance evaluation results showed that the proposed model had an average 13.8 percent improvement in data processing time compared to the existing model, and the server's overhead was 10.5 percent lower on average than the existing model. The accuracy of extracting IoT information from servers has improved by 8.7% on average from previous models.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.31 no.9C
• /
• pp.853-858
• /
• 2006

TMHMM(Tied Mixture Hidden Markov Model) is an important approach to reduce the number of free parameters in speech recognition. However, this model suffers from a degradation in recognition accuracy due to its GPDF (Gaussian Probability Density Function) clustering error. This paper proposes a clustering algorithm, called HCNN(Homogeneous Centroid Neural network), to cluster acoustic feature vectors in TMHMM. Moreover, the HCNN uses the heterogeneous distance measure to allocate more code vectors in the heterogeneous areas where probability densities of different states overlap each other. When applied to Korean digit isolated word recognition, the HCNN reduces the error rate by 9.39% over CNN clustering, and 14.63% over the traditional K-means clustering.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.13 no.5
• /
• pp.512-519
• /
• 2003

In this paper, we propose a neuro-fuzzy modeling to improve the performance using the hierarchical clustering and Gaussian Mixture Model(GMM). The hierarchical clustering algorithm has a property of producing unique parameters for the given data because it does not use the object function to perform the clustering. After optimizing the obtained parameters using the GMM, we apply them as initial parameters for Adaptive Network-based Fuzzy Inference System. Here, the number of fuzzy rules becomes to the cluster numbers. From this, we can improve the performance index and reduce the number of rules simultaneously. The proposed method is verified by applying to a neuro-fuzzy modeling for Box-Jenkins s gas furnace data and Sugeno's nonlinear system, which yields better results than previous oiles.

• Journal of Industrial Technology
• /
• v.28 no.A
• /
• pp.185-189
• /
• 2008

Practical pattern recognition systems should overcome very large class problem. Sometimes it is almost impossible to build every model for every class due to memory and time constraints. For this case, grouping similar models will be helpful. In this paper, we propose GMC(Group Model Clustering) to build a large class Chinese character recognition system. We built hidden Markov models for 10% of total classes, then classify the rest of classes into already trained group classes. Finally group models are trained using group model clustered data. Recognition is performed using only group models, in order to achieve reduced model size and improved recognition speed.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.11 no.12
• /
• pp.4991-4996
• /
• 2010

Clustering method in wireless sensor network is the technique that forms the cluster to aggregate the data and transmit them at the same time that they can use the energy efficiently. Even though cluster group model is based on clustering, it differs from previous method that reducing the total energy consumption by separating energy overload to cluster group head and cluster head. In this thesis, I calculate the optimal cluster group number and cluster number in this kind of cluster group model according to threshold of energy consumption model. By using that I can minimize the total energy consumption in sensor network and maximize the network lifetime. I also show that proposed cluster group model is better than previous clustering method at the point of network energy efficiency.

• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.60 no.1
• /
• pp.184-192
• /
• 2011

In this study, we introduce a identification methodology for FCM-based fuzzy model. The two underlying design mechanisms of such networks involve Fuzzy C-Means (FCM) clustering method and Particle Swarm Optimization(PSO). The proposed algorithm is based on FCM clustering method for efficient processing of data and the optimization of model was carried out using PSO. The premise part of fuzzy rules does not construct as any fixed membership functions such as triangular, gaussian, ellipsoidal because we build up the premise part of fuzzy rules using FCM. As a result, the proposed model can lead to the compact architecture of network. In this study, as the consequence part of fuzzy rules, we are able to use four types of polynomials such as simplified, linear, quadratic, modified quadratic. In addition, a Weighted Least Square Estimation to estimate the coefficients of polynomials, which are the consequent parts of fuzzy model, can decouple each fuzzy rule from the other fuzzy rules. Therefore, a local learning capability and an interpretability of the proposed fuzzy model are improved. Also, the parameters of the proposed fuzzy model such as a fuzzification coefficient of FCM clustering, the number of clusters of FCM clustering, and the polynomial type of the consequent part of fuzzy rules are adjusted using PSO. The proposed model is illustrated with the use of Automobile Miles per Gallon(MPG) and Boston housing called Machine Learning dataset. A comparative analysis reveals that the proposed FCM-based fuzzy model exhibits higher accuracy and superb predictive capability in comparison to some previous models available in the literature.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.24 no.4
• /
• pp.443-449
• /
• 2014

Preferences of users on movies are observables of various factors that are related with user attributes and movie features. For movie recommendation, analysis methods for relation among users, movies, and preference patterns are mandatory. As a relational analysis tool, we focus on the Infinite Relational Model (IRM) which was introduced as a tool for multiple concept search. We show that IRM-based co-clustering on preference patterns and movie descriptors can be used as the first tool for movie recommender methods, especially content-based filtering approaches. By introducing a set of well-defined tag sets for movies and doing three-way co-clustering on a movie-rating matrix and a movie-tag matrix, we discovered various explainable relations among users and movies. We suggest various usages of IRM-based co-clustering, espcially, for incremental and dynamic recommender systems.