• Journal of KIISE
• /
• v.42 no.12
• /
• pp.1561-1567
• /
• 2015

Abstract Semi-supervised learning is an area in machine learning that employs both labeled and unlabeled data in order to train a model and has the potential to improve prediction performance compared to supervised learning. Graph-based semi-supervised learning has recently come into focus with two phases: graph construction, which converts the input data into a graph, and label inference, which predicts the appropriate labels for unlabeled data using the constructed graph. The inference is based on the smoothness assumption feature of semi-supervised learning. In this study, we propose an enhanced label inference algorithm by incorporating the importance of each vertex. In addition, we prove the convergence of the suggested algorithm and verify its excellence.

• Korean Journal of Applied Biomechanics
• /
• v.20 no.2
• /
• pp.205-212
• /
• 2010

The purpose of this study was to compare the efficiency of visual perception and impulse according to the three types of Taekwondo players to be able to supply an efficient training method, for this a total of 12 representative Taekwondo players of the Korean National team, 4 poomsae players, 4 kyokpa players and 4 kyorugi players weighting between 68 to 74 kg, and the results from the motion analysis system, eye tracker and Electronic hogu are as follows. For the visual perceptual function, the total body reaction time was slowest for the kyokpa group, and for the visible reaction and vision fixation time was longest of the poomsae group, while the performance movement was fastest for the kyorugi group. As for description of the two kicking motions dollyo chagi and dolgae chagi the longer visual fixation helps the accuracy of the kick. In conclusion, as there was a difference between the groups, this information could help to train the visual perception of players according to what event they are participating in.

• Journal of Powder Materials
• /
• v.19 no.3
• /
• pp.189-195
• /
• 2012

A novel PM (powder metallurgy) steel for automotive power-train gear components was developed to reduce manufacturing cost, while meeting application requirements. The high-density PM steel was manufactured by mixing using special Cr-Mo atomized iron powders, high-pressure compaction, and sintering. Tensile strength, charpy impact, bending fatigue, and contact fatigue tests for the PM steel were carried out and compared to conventional forged steel. Pinion gears for auto-transmission were also manufactured by helical pressing, sintering, and surface densification process. In order to evaluate the durability of the PM parts, auto-transmission durability tests were performed using dynamometer tests. Results showed that the PM steel fulfilled the requirements for pinion gears indicating suitable tensile, bending fatigue, contact fatigue strengths and improved gear tooth profile. The PM gears also showed good performance during the transmission durability tests. As a result, the PM gears showed significant potential to replace the conventional forged steel gears manufactured by tooth machining (hobbing, shaving, and grinding) processes.

• Proceedings of the KSR Conference
• /
• 2004.10a
• /
• pp.1353-1359
• /
• 2004

There are many massive components added on the railway overhead-line. These components cause larger fluctuations of contact forces, which are due to wave reflections and Doppler effects when a high-speed train passes those. In this paper, mathematical formula are derived for the relation between the added mass and contact force fluctuations. Using the derived formula, we calculate a added mass on the overhead-line which cause amplification factor to become 2.5. German design practice requires that amplification factor due to the wave reflection should be less than 2.5 to obtain good current collection performance. To show the validity of the formula, simulation results are compared with the calculation results. Simulation results showed that contact force fluctuations grow rapidly when an added mass is larger than the calculation result. Therefore, the simple form of formula can be used for estimating maximum added mass not to cause large fluctuations of contact forces in early design phase.

• Transactions of the Korean hydrogen and new energy society
• /
• v.23 no.4
• /
• pp.330-336
• /
• 2012

FCEV uses 250 ~ 450 V instead of using 12 V battery. High voltage vehicle can cause electric shock, fire and explosion accident. Therefore, it has potential factors that can cause hazard of safety for users. United states of America and Europe legislate regulations such as ECE R100, FMVSS 305 for regulating electrical safety during driving or after collision. The company manufacturing high voltage components must do advanced R&D about Method for improving and confirming the safety of high voltage. We develop the specific hardware components of high voltage wiring system for the power train system and power supply system of Hyundai Motors FCEV. This paper shows test method of insulative performance for securing the electrical safety of high voltage components such as power cable, connectors and buss-bar, and proposals the guide line value for human safety of FCEV according to the test result of our development components.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.19 no.12
• /
• pp.2418-2425
• /
• 1994

Although the error backpropagation(EBP) algorithm based on the gradient descent method is a widely-used learning algorithm of neural networks, learning sometimes takes a long time to acquire accuracy. This paper develops a novel learning method to alleviate the problems of EBP algorithm such as local minima, slow speed, and size of structure and thus to improve performance by adopting other new networks. Gaussian Potential Function networks(GPFN), in parallel with multilayer neural networks. Empirical simulations show the efficacy of the proposed algorithm in function approximation, which enables us to train networks faster with the better generalization capabilities.

• Proceedings of the KIEE Conference
• /
• 1995.07b
• /
• pp.705-709
• /
• 1995

The objective of this study is to design a multi-layer neural network which controls the position of excavator's attachment. In this paper, a dynamic controller has been developed based on an error back-propagation(BP) neural network. Since the neural network can model an arbitrary nonlinear mapping, it was used as a commanded feedforward input generator. A PD feedback controller is used in parallel with the feedforward neural network to train the system. The neural network was trained by the current state of the excavator as well as the PD feedback error. By using the BP network as a feedforward controller, no a priori knowledge on system dynamics is need. Computer simulation results demonstrate such powerful characteristics of the proposed controller as adaptation to changing environment, robustness to disturbancen and performance improvement with the on-line learning in the position control of excavator attachment.

• Proceedings of the KSR Conference
• /
• 2003.05a
• /
• pp.567-573
• /
• 2003

Korean High Speed Train (KHST) has been tested on high speed line in JungBu site since it was developed in 2002. The data acquisition system was developed to accomplish successfully this on-line test for proving the dynamic Performance of KHST. This system was consist of the personal computers based on National Instrument PXI modules and the test programs based on Labview 6i. This paper shows that this system is efficient to acquire the test data through the multi-channels connected the accelerometers which located in long distance places and flexible to change and add channels for data acquisition. The dynamic analysis of an on-line test is very complicate because the environmental conditions, as examples radius of curve, inclination of the track, tunnels, bridges, and so forth, and running conditions, as examples driving, braking, the number of working motors, and so forth, have an effect on the results. Therefor, the analysis method is important and this paper proposes the efficient procedure graphically, showing the proposed method simplify the accelerations of 5th bogie frame acquired during the on-line test for KHST.

• ETRI Journal
• /
• v.36 no.4
• /
• pp.662-672
• /
• 2014

Human action recognition is used in areas such as surveillance, entertainment, and healthcare. This paper proposes a system to recognize both single and continuous human actions from monocular video sequences, based on 3D human modeling and cyclic hidden Markov models (CHMMs). First, for each frame in a monocular video sequence, the 3D coordinates of joints belonging to a human object, through actions of multiple cycles, are extracted using 3D human modeling techniques. The 3D coordinates are then converted into a set of geometrical relational features (GRFs) for dimensionality reduction and discrimination increase. For further dimensionality reduction, k-means clustering is applied to the GRFs to generate clustered feature vectors. These vectors are used to train CHMMs separately for different types of actions, based on the Baum-Welch re-estimation algorithm. For recognition of continuous actions that are concatenated from several distinct types of actions, a designed graphical model is used to systematically concatenate different separately trained CHMMs. The experimental results show the effective performance of our proposed system in both single and continuous action recognition problems.

• Journal of Institute of Control, Robotics and Systems
• /
• v.16 no.5
• /
• pp.433-439
• /
• 2010

This paper proposes a beacon color code scheduling algorithm for the localization of multiple robots in a multi-block workspace. With the developments of intelligent robotics and ubiquitous technology, service robots are applicable for the wide area such as airports and train stations where multiple indoor GPS systems are required for the localization of the mobile robots. Indoor localization schemes using ultrasonic sensors have been widely studied due to its cheap price and high accuracy. However, ultrasonic sensors have some shortages of short transmission range and interferences with other ultrasonic signals. In order to use multiple robots in wide workspace concurrently, it is necessary to resolve the interference problem among the multiple robots in the localization process. This paper proposes an indoor localization system for concurrent multiple robots localization in a wide service area which is divided into multi-block for the reliable sensor operation. The beacon color code scheduling algorithm is developed to avoid the signal interferences and to achieve efficient localization with high accuracy and short sampling time. The performance of the proposed localization system is verified through the simulations and the real experiments.