• The KIPS Transactions:PartB
• /
• v.13B no.4 s.107
• /
• pp.439-448
• /
• 2006

Previous recognition/clustering algorithms such as Kohonen SOM(Self-Organizing Map), MLP(Multi-Layer Percecptron) and SVM(Support Vector Machine) might not adapt to unexpected input pattern. And it's recognition rate depends highly on the complexity of own training patterns. We could make up for and improve the weak points with lowering complexity of original problem without losing original characteristics. There are so many ways to lower complexity of the problem, and we chose a kernel concepts as an approach to do it. In this paper, using a kernel concepts, original data are mapped to hyper-dimension space which is near infinite dimension. Therefore, transferred data into the hyper-dimension are distributed spasely rather than originally distributed so as to guarantee the rate to be risen. Estimating ratio of recognition is based on a new similarity-probing and learning method that are proposed in this paper. Using CEDAR DB which data is written in cursive letters, 0 to 9, we compare a recognition/clustering performance of kSOM that is proposed in this paper with previous SOM.

• The KIPS Transactions:PartD
• /
• v.19D no.1
• /
• pp.21-28
• /
• 2012

Predication of protein interaction sites for monomer structures can reduce the search space for protein docking and has been regarded as very significant for predicting unknown functions of proteins from their interacting proteins whose functions are known. In the other hand, the prediction of interaction sites has been limited in crystallizing weakly interacting complexes which are transient and do not form the complexes stable enough for obtaining experimental structures by crystallization or even NMR for the most important protein-protein interactions. This work reports the calculation of 3D surface patches of complex structures and their properties and a machine learning approach to build a predictive model for the 3D surface patches in interaction and non-interaction sites using support vector machine. To overcome classification problems for class imbalanced data, we employed an under-sampling technique. 9 properties of the patches were calculated from amino acid compositions and secondary structure elements. With 10 fold cross validation, the predictive model built from SVM achieved an accuracy of 92.7% for classification of 3D patches in interaction and non-interaction sites from 147 complexes.

• Convergence Security Journal
• /
• v.23 no.4
• /
• pp.61-70
• /
• 2023

Clustering is an unsupervised learning method that involves grouping data based on features such as distance metrics, using data without known labels or ground truth values. This method has the advantage of being applicable to various types of data, including images, text, and audio, without the need for labeling. Traditional clustering techniques involve applying dimensionality reduction methods or extracting specific features to perform clustering. However, with the advancement of deep learning models, research on deep clustering techniques using techniques such as autoencoders and generative adversarial networks, which represent input data as latent vectors, has emerged. In this study, we propose a deep clustering technique based on deep learning. In this approach, we use an autoencoder to transform the input data into latent vectors, and then construct a vector space according to the cluster structure and perform k-means clustering. We conducted experiments using the MNIST and Fashion-MNIST datasets in the PyTorch machine learning library as the experimental environment. The model used is a convolutional neural network-based autoencoder model. The experimental results show an accuracy of 89.42% for MNIST and 56.64% for Fashion-MNIST when k is set to 10.

• Journal of the Korean Society of Manufacturing Process Engineers
• /
• v.19 no.9
• /
• pp.47-52
• /
• 2020

With the 4th industrial revolution, condition monitoring using machine learning techniques has become popular among researchers. An overload due to complex operations causes several irregularities in MOSFETs. This study investigated the acquired voltage to analyze the overcurrent effects on MOSFETs using a failure mode effect analysis (FMEA). The results indicated that the voltage pattern changes greatly when the current is beyond the threshold value. Several features were extracted from the collected voltage signals that indicate the health state of a switched-mode power supply (SMPS). Then, the data were reduced to a smaller sample space by using a principal component analysis (PCA). A robust machine learning algorithm, the support vector machine (SVM), was used to classify different health states of an SMPS, and the classification results are presented for different parameters. An SVM approach assisted by a PCA algorithm provides a strong fault diagnosis framework for an SMPS.

• Proceedings of the KSRS Conference
• /
• v.2
• /
• pp.635-638
• /
• 2006

Regression of conventional prediction techniques in data mining uses the model which is generated from the training step. This model is applied to new input data without any change. If this model is applied directly to time series, the rate of prediction accuracy will be decreased. This paper proposes an incremental regression for time series prediction like typhoon track prediction. This technique considers the characteristic of time series which may be changed over time. It is composed of two steps. The first step executes a fractional process for applying input data to the regression model. The second step updates the model by using its information as new data. Additionally, the model is maintained by only recent data in a queue. This approach has the following two advantages. It maintains the minimum information of the model by using a matrix, so space complexity is reduced. Moreover, it prevents the increment of error rate by updating the model over time. Accuracy rate of the proposed method is measured by RME(Relative Mean Error) and RMSE(Root Mean Square Error). The results of typhoon track prediction experiment are performed by the proposed technique IMLR(Incremental Multiple Linear Regression) is more efficient than those of MLR(Multiple Linear Regression) and SVR(Support Vector Regression).

• Journal of IKEEE
• /
• v.22 no.3
• /
• pp.693-699
• /
• 2018

CNN based object recognitions have reported splendid results. However, the recognition of mobile apps raises an interesting challenge that recognition performance of similar widgets is not consistent. In order to improve the performance, we propose a noble method utilizing relations between input widgets. The recognition process flows from the Faster R-CNN based recognition to enhancement using a relation recognizer. The relations are represented as vector translation between objects in a relation space. Experiments on 323 apps show that our method significantly enhances the Faster R-CNN only approach.

• Journal of Power Electronics
• /
• v.9 no.2
• /
• pp.232-242
• /
• 2009

In this paper, a matrix converter with improved low frequency output performance is proposed by achieving a one-step commutation owing to a general commutation circuit applicable to n-phase to m-phase matrix converters. The commutation circuit consists of simple resister and capacitor components, leading to a very stable, reliable and robust operation. Also, it requires no extra sensing information to achieve commutation, allowing for a one-step commutation like a conventional dead time commutation. With the dead time commutation strategy applied, the distortion caused by commutation delay is analyzed and compensated, therefore leading to better output linear behavior. In this paper, detailed commutation procedures of the R-C commutation circuit are analyzed. A selection of specific semiconductor switches and commutation circuit components is also provided. Finally, the effectiveness of the proposed commutation method is verified through a two-phase to single-phase matrix converter and the feasibility of the compensation approach is shown by an open loop space vector modulated three-phase matrix converter with a passive load.

• Journal of Power Electronics
• /
• v.14 no.5
• /
• pp.967-979
• /
• 2014

This paper focuses on speed and current sensor fault detection and isolation (FDI) for induction motor (IM) drives. A new, accurate and high-efficiency FDI approach is proposed so that a system can continue operating with good performance even in the presence of speed sensor faults, current sensor faults or both. The proposed three paralleled adaptive observers are capable of current sensor fault detection and localization. By using observers, the rotor flux and rotor speed can be estimated which allows the system to run under the speed sensorless vector control mode when a speed sensor fault occurs. In order to detect speed sensor faults, a threshold-based scheme is proposed. To verify the feasibility and effectiveness of the proposed FDI strategy, experiments are carried out under different conditions based on a dSPACE DS1104 induction motor drive platform.

• Proceedings of the KIEE Conference
• /
• 1998.07g
• /
• pp.2360-2362
• /
• 1998

This paper presents a new collison-free trajectory problem for SCARA robot manipulator. we use artificial potential field for collison detection and avoidance. The potential function is typically defined as the sum of attractive potential pulling the robot toward the goal configuration and a repulsive potential pushing the robot away from the obstacles. In here, end-effector of manipulator is represented as a particle in configuration space and moving obstacles is simply represented, too. we consider not fixed obstacle but moving obstacle in random. So, we propose new distance function of artificial potential field with moving obstacle for SCARA robot. At every sampling time, the artificial potential field is update and the force driving manipulator is derived from the gradient vector of artificial potential field. To real-time path planning, we apply very simple modeling to obstacle. Some simulation results show the effectiveness of the proposed approach.

• The Transactions of the Korean Institute of Electrical Engineers A
• /
• v.48 no.3
• /
• pp.234-240
• /
• 1999

This paper presents a new approach for large-scale generator maintenance scheduling optimizations. The generator preventive maintenance scheduling problems are typical discrete dynamic n-dimensional vector optimization ones with several inequality constraints. The considered objective function to be minimized a subset of{{{{ { R}^{n } }}}} space is the variance (i.g., second-order momentum) of operating reserve margin to levelize risk or reliability during a year. By its nature of the objective function, the optimal solution can only be obtained by enumerating all combinatorial states of each variable, a task which leads to computational explosion in real-world maintenance scheduling problems. This paper proposes a new priority search mechanism based on each generator's discrete sensitivity value which was analytically developed in this study. Unlike the conventional capacity-based priority search, it can prevent the local optimal trap to some extents since it changes dynamically the search tree in each iteration. The proposed method have been applied to two test systems (i.g., one is a sample system with 10 generators and the other is a real-world lage scale power system with 280 generators), and the results anre compared with those of the conventional capacith-based search method and combinatorial optimization method to show the efficiency and effectiveness of the algorithm.