• Journal of the Korean Institute of Intelligent Systems
• /
• v.22 no.2
• /
• pp.205-211
• /
• 2012

The retrieved documents have to be transformed into proper data structure for the clustering algorithms of statistics and machine learning. A popular data structure for document clustering is document-term matrix. This matrix has the occurred frequency value of a term in each document. There is a sparsity problem in this matrix because most frequencies of the matrix are 0 values. This problem affects the clustering performance. The sparseness of document-term matrix decreases the performance of clustering result. So, this research uses the factor score by factor analysis to solve the sparsity problem in document clustering. The document-term matrix is transformed to document-factor score matrix using factor scores in this paper. Also, the document-factor score matrix is used as input data for document clustering. To compare the clustering performances between document-term matrix and document-factor score matrix, this research applies two typed matrices to self organizing map (SOM) clustering.

• Proceedings of the KIEE Conference
• /
• 1990.11a
• /
• pp.421-424
• /
• 1990

An iterative learning control scheme for exact-tracking control and parameter estimation of uncertain robotic systems is presented. In the learning control structure, tracking and feedforward input converge globally and asymptotically as iteration increases. Since convergence of parameter errors depends only on the persistent exciting condition of system trajectories along the iteration independently of length of trajectories, it may be achieved with only system trajectories of small duration. In addition, these learning control schemes are expected to be effectively applicable to time-varying parametric systems as well as time-invariant systems, for the parameter estimation is performed at each fixed time along the iteration. Finally, no usage of acceleration signal and no in version of estimated inertia matrix in the parameter estimator makes these learning control schemes more feasible.

• The Transactions of the Korean Institute of Electrical Engineers
• /
• v.40 no.4
• /
• pp.427-438
• /
• 1991

An iterative learning control scheme for exact-tracking control and parameter estimation of uncertain robotic system is preented. In the learning control structure, the control input converges globally and asymtotically to the desired input as iteration increases. Since convergence of parameter errors depends only on the persistent exciting condition of system trajectories along the iteration independently of the time-duration of trajectories, it may be achieved with system trajectories with small duration. In addition, the proposd learning control schemes are applicable to time-varying parametric systems as well as time-invariant systems, because the parameter estimation is performed at each fixed time along the iteration. In the parameter estimator, the acceleration information as well as the inversion of estimated inertia matrix are not used at all, which makes the proposed learning control schemes more feasible.

• Journal of the Korean Society for Industrial and Applied Mathematics
• /
• v.25 no.4
• /
• pp.162-172
• /
• 2021

This paper proposes stochastic methods to find an approximate solution for the L²-Wasserstein least squares problem of Gaussian measures. The variable for the problem is in a set of positive definite matrices. The first proposed stochastic method is a type of classical stochastic gradient methods combined with projection and the second one is a type of variance reduced methods with projection. Their global convergence are analyzed by using the framework of proximal stochastic gradient methods. The convergence of the classical stochastic gradient method combined with projection is established by using diminishing learning rate rule in which the learning rate decreases as the epoch increases but that of the variance reduced method with projection can be established by using constant learning rate. The numerical results show that the present algorithms with a proper learning rate outperforms a gradient projection method.

• Sleep Medicine and Psychophysiology
• /
• v.8 no.1
• /
• pp.67-71
• /
• 2001

Independent Component Analysis (ICA) is a blind source separation method using unsupervised learning and mutual information theory created in the late eighties and developed in the nineties. It has already succeeded in separating eye movement artifacts from human scalp EEG recording. Several characteristic sleep waves such as sleep spindle, K-complex, and positive occipital sharp transient of sleep (POSTS) can be recorded during sleep EEG recording. They are used as stage determining factors of sleep staging and might be reflections of unknown neural sources during sleep. We applied the ICA method to sleep EEG for sleep waves separation. Eighteen channel scalp longitudinal bipolar montage was used for the EEG recording. With the sampling rate of 256Hz, digital EEG data were converted into 18 by n matrix which was used as a original data matrix X. Independent source matrix U (18 by n) was obtained by independent component analysis method ($U=W{\timex}X$, where W is an 18 by 18 matrix obtained by ICA procedures). ICA was applied to the original EEG containing sleep spindle, K-complex, and POSTS. Among the 18 independent components, those containing characteristic shape of sleep waves could be identified. Each independent component was reconstructed into original montage by the product of inverse matrix of W (inv(W)) and U. The reconstructed EEG might be a separation of sleep waves without other components of original EEG matrix X. This result (might) demonstrates that characteristic sleep waves may be separated from original EEG of unknown mixed neural origins by the Independent Component Analysis (ICA) method.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.18 no.4
• /
• pp.1020-1041
• /
• 2024

With the development of covert communication technologies, the number of covert communication technologies using blockchain as a carrier is increasing. However, using the transaction amount of digital currency as a carrier for covert communication has problems such as low embedding rate, large consumption of transaction amount, and easy detection. In this paper, firstly, by experimentally analyzing the distribution of bitcoin transaction amounts, we determine the most suitable range of amounts for matrix decomposition. Secondly, we design a novel matrix decomposition method that can successfully decompose a large amount matrix into two small amount matrices and utilize the elements in the small amount matrices for covert communication. Finally, we analyze the feasibility of the novel matrix decomposition method in this scheme in detail from four aspects, and verify it by experimental comparison, which proves that our scheme not only improves the embedding rate and reduces the consumption of transaction amount, but also has a certain degree of resistance to detection.

• The KIPS Transactions:PartB
• /
• v.10B no.4
• /
• pp.397-402
• /
• 2003

This paper proposes an efficient fixed-point (FP) algorithm for improving performances of the independent component analysis (ICA) based on neural networks. The proposed algorithm is the FP algorithm based on Newton method for ICA using the adaptive learning parameters. The purpose of this algorithm is to improve the separation speed and performance by using the learning parameters in Newton method, which is based on the first order differential computation of entropy optimization function. The learning rate and the moment are adaptively adjusted according to an updating state of inverse mixing matrix. The proposed algorithm has been applied to the fingerprints and the images generated by random mixing matrix in the 8 fingerprints of 256${\times}$256-pixel and the 10 images of 512$\times$512-pixel, respectively. The simulation results show that the proposed algorithm has the separation speed and performance better than those using the conventional FP algorithm based on Newton method. Especially, the proposed algorithm gives relatively larger improvement degree as the problem size increases.

• Structural Engineering and Mechanics
• /
• v.12 no.5
• /
• pp.527-540
• /
• 2001

A new sort of learning algorithm named whole learning algorithm is proposed to simulate the nonlinear and dynamic behavior of RC members for the estimation of structural integrity. A mathematical technique to solve the multi-objective optimization problem is applied for the learning of the feedforward neural network, which is formulated so as to minimize the Euclidean norm of the error vector defined as the difference between the outputs and the target values for all the learning data sets. The change of the outputs is approximated in the first-order with respect to the amount of weight modification of the network. The governing equation for weight modification to make the error vector null is constituted with the consideration of the approximated outputs for all the learning data sets. The solution is neatly determined by means of the Moore-Penrose generalized inverse after summarization of the governing equation into the linear simultaneous equations with a rectangular matrix of coefficients. The learning efficiency of the proposed algorithm from the viewpoint of computational cost is verified in three types of problems to learn the truth table for exclusive or, the stress-strain relationship described by the Ramberg-Osgood model and the nonlinear and dynamic behavior of RC members observed under an earthquake.

• Educational Technology International
• /
• v.19 no.1
• /
• pp.35-60
• /
• 2018

The purpose of this study was to analyze the types of lesson and its effectiveness with digital textbook. To address those goals, we had observed five classes of the primary school, which designated as a research pilot school for digital textbook. Based on the result of observation, 3 types of lesson with digital textbook were categorized: Teacher-directed lecture (type 1), Blended learning (type 2), and Flipped learning (type 3). Depending on the type of lesson was analyzed the positive and negative effectiveness by means of matrix analysis method. As a result, in Teacher-directed lecture (type 1), there was found out the participation of the lesson in atmosphere of stable and comfortable as positive experience, also digital textbook operating immature and boring as negative experience. In Blended learning (type 2), there was found out the fun by sharing the product and peer feedback, and flow by learning transfer as positive experience, also digital textbook operating immature and understanding the difference between assignments as negative experience. In Flipped learning (type 3), there was shown the positive attitude and ownership in the lesson as positive experience, also distracting and boring in the lesson when learner was excluded in participation as negative experience. Based on the results, we suggested some strategies for improving positive experience and protecting negative experience in the lesson with using digital textbook.

• 제어로봇시스템학회:학술대회논문집
• /
• 2000.10a
• /
• pp.482-482
• /
• 2000

The essential requirements of neural network for human skill transfer are fast convergence, high storage capacity, and strong noise immunity. Bidirectional associative memory(BAM) suffering from low storage capacity and abundance of spurious memories is rarely used for skill transfer application though it has fast and wide association characteristics for visual data. This paper suggests generalization of classical BAM structure and new learning algorithm which uses supervised learning to guarantee perfect recall starting with correlation matrix. The generalization is validated to accelerate convergence speed, to increase storage capacity, to lessen spurious memories, to enhance noise immunity, and to enable multiple association using simulation work.