• Database Research
• /
• v.34 no.3
• /
• pp.137-147
• /
• 2018

A key to a well-performing human activity recognition (HAR) system through machine learning technique is the availability of a substantial amount of labeled data. Collecting sufficient labeled data is an expensive and time-consuming task. To build a HAR system in a new environment (i.e., the target domain) with very limited labeled data, it is unfavorable to naively exploit the data or trained classifier model from the existing environment (i.e., the source domain) as it is due to the domain difference. While traditional machine learning approaches are unable to address such distribution mismatch, transfer learning approach leverages the utilization of knowledge from existing well-established source domains that help to build an accurate classifier in the target domain. In this work, we propose a transfer learning approach to create an accurate HAR classifier with very limited data through the multitask neural network. The classifier loss function minimization for source and target domain are treated as two different tasks. The knowledge transfer is performed by simultaneously minimizing the loss function of both tasks using a single neural network model. Furthermore, we utilize the unlabeled data in an unsupervised manner to help the model training. The experiment result shows that the proposed work consistently outperforms existing approaches.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2008.05a
• /
• pp.694-698
• /
• 2008

In hydrology, it is a difficult task to deal with multivariate time series such as modeling streamflows of an entire complex river system. Normal distribution based model such as MARMA (Multivariate Autorgressive Moving average) has been a major approach for modeling the multivariate time series. There are some limitations for the normal based models. One of them might be the unfavorable data-transformation forcing that the data follow the normal distribution. Furthermore, the high dimension multivariate model requires the very large parameter matrix. As an alternative, one might be decomposing the multivariate data into independent components and modeling it individually. In 1985, Lins used Principal Component Analysis (PCA). The five scores, the decomposed data from the original data, were taken and were formulated individually. The one of the five scores were modeled with AR-2 while the others are modeled with AR-1 model. From the time series analysis using the scores of the five components, he noted "principal component time series might provide a relatively simple and meaningful alternative to conventional large MARMA models". This study is inspired from the researcher's quote to develop a multivariate simulation model. The multivariate simulation model is suggested here using Principal Component Analysis (PCA) and Independent Component Analysis (ICA). Three modeling step is applied for simulation. (1) PCA is used to decompose the correlated multivariate data into the uncorrelated data while ICA decomposes the data into independent components. Here, the autocorrelation structure of the decomposed data is still dominant, which is inherited from the data of the original domain. (2) Each component is resampled by block bootstrapping or K-nearest neighbor. (3) The resampled components bring back to original domain. From using the suggested approach one might expect that a) the simulated data are different with the historical data, b) no data transformation is required (in case of ICA), c) a complex system can be decomposed into independent component and modeled individually. The model with PCA and ICA are compared with the various statistics such as the basic statistics (mean, standard deviation, skewness, autocorrelation), and reservoir-related statistics, kernel density estimate.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.39C no.6
• /
• pp.503-511
• /
• 2014

In this paper, a novel, real-time video stitching algorithm in an H.264/AVC compressed domain is proposed. This enables viewers to watch multiple video contents using a single device. The basic concept of this paper is that the server is asked to combine multiple streams into one bit-stream based in a compressed domain. In other words, this paper presents a new compressed domain combiner that works in boundary macroblocks of input videos with re-calculating intra prediction mode, intra prediction MVD, a re-allocation of the coefficient table, and border extension methods. The rest of the macroblocks of the input video data are achieved simply by copying them. Simulation experiments have demonstrated the possibility and effectiveness of the proposed algorithm by showing that it is able to generate more than 103 frames per second, stitching four 480p-sized images into each frame.

• Proceedings of the Korea Society of Information Technology Applications Conference
• /
• 2005.11a
• /
• pp.177-178
• /
• 2005

Wireless 1394 over IEEE802.15.3 must allow a data reserved for delivery over a wired 1394 network to be delivered over an IEEE802.15.3 wireless network through bridging IEEE 1394 to IEEE802.15.3. Isochronous transfers on the 1394 bus guarantee timely delivery of data. Specifically, isochronous transfers are scheduled by the bus so that they occur once every $125\;{\mu}s$ and require clock time synchronization to complete the real-time data transfer. IEEE1394.1 and Protocol Adaptation Layer for IEEE1394 over IEEE802.15.3 specify clock time synchronization for a wired 1394 bus network to a wired 1394 bus network and wireless 1394 nodes, which are IEEE802.15.3 nodes handling 1394 applications, over IEEE802.15.3. Thus, the clock time synchronizations are just defined within a homogeneous network environment like IEEE1394 or IEEE802.15.3 until now. This paper proposes new clock time synchronization method for wireless 1394 heterogeneous networks between 1394 and 802.15.3. If new method is adopted for various wireless 1394 products, consumer electronics devices such as DTV and Set-top Box or PC devices on a 1394 bus network can transmit real time data to the AV devices on the other 1394 bus in a different place via IEEE 802.15.3.

• Wind and Structures
• /
• v.39 no.4
• /
• pp.271-285
• /
• 2024

The multiple-fan wind tunnel is an important facility for reproducing target wind field. Existing control methods for the multiple-fan wind tunnel can generate wind speeds that satisfy the target statistical characteristics of a wind field (e.g., power spectrum). However, the frequency-domain features cannot well represent the nonstationary winds of extreme storms (e.g., downburst). Therefore, this study proposes a multiple-fan wind tunnel control scheme based on Deep Reinforcement Learning (DRL), which will completely transform into a data-driven closed-loop control problem, to reproduce the target wind field in the time domain. Specifically, the control scheme adopts the Deep Deterministic Policy Gradient (DDPG) paradigm in which the strong fitting ability of Deep Neural Networks (DNN) is utilized. It can establish the complex relationship between the target wind speed time series and the current control strategy in the DRL-agent. To address the fluid memory effect of the wind field, this study innovatively designs the system state and control reward to improve the reproduction performance based on historical data. To validate the performance of the model, we established a simplified flow field based on Navier Stokes equations to simulate a two-dimensional numerical multiple-fan wind tunnel environment. Using the strategy of DRL decision maker, we generated a wind speed time series with minor error from the target under low Reynolds number conditions. This is the first time that the AI methods have been used to generate target wind speed time series in a multiple-fan wind tunnel environment. The hyperparameters in the DDPG paradigm are analyzed to identify a set of optimal parameters. With these efforts, the trained DRL-agent can simultaneously reproduce the wind speed time series in multiple positions.

• Journal of the Korean Society for Precision Engineering
• /
• v.17 no.10
• /
• pp.125-131
• /
• 2000

By using the test of signal error, model structure of an electrohydraulic servo system is determined. For determining parameter of the electrohydraulic servo system, using time discrete model of parametric method, parameters in time discrete model are searched by the least square method. By bilinear transform, we have found the model of electrohydraulic servo system in s domain. Afterwards, we have compared experimental data with simulation data by MATLAB having the identified parameter. As the result, experimental data is agreed with simulation data very well.

• Journal of the Korean Society for Precision Engineering
• /
• v.19 no.4
• /
• pp.87-92
• /
• 2002

By using the signal error test, model structure of total antilock braking system consisting of electromagnetic system and hydraulic system is determined as 9th order system. For determining parameters of the ABS, using time discrete model of parametric method, parameters in time discrete model are searched by least square method. By bilinear transform, we have found the model of ABS in s domain. Afterward, experimental output data is compared with simulated output data by MATLAB haying identified parameter. As the result, experimental data is agreed with simulated data very well.

• Proceedings of the KSR Conference
• /
• 2002.10a
• /
• pp.471-477
• /
• 2002

In this paper, it is proposed instruction code satisfied algorithm that is able to prevent data collision when transponder access in the area of recognition system that is operated to single channel. Differ from absolute collision is used to in the time domain procedure, instruction code satisfied algorithm, transmits data which don't generate collision and must satisfy instruction code. So, to prevent data collision, transponder is of great if it read the entire instruction code. Consequently, it is applied to algorithm that made the system of wireless recognition13.56 Mhz. As a result, for the time of one bit data transmission had 14${\mu}$s difference, it is proved the prevention of data loss in experiment.

• Journal of Gifted/Talented Education
• /
• v.24 no.5
• /
• pp.851-876
• /
• 2014

As an empirical test of a model of giftedness with learning disabilities (Song & Porath, 2011), this paper investigated domain thoughts of gifted students without learning disabilities and gifted students with learning disabilities (GLD) in reading, writing, and math. Gifted students in each group were interviewed and the data were analyzed for domain thoughts. The results showed that the former group of gifted students exhibited domain thoughts in a more balanced manner, whereas GLD students showed large discrepancies between domain thoughts; they showed weak specific domain thoughts in contrast to strong other domain thoughts. They also showed ambivalent attitudes even in a domain activity; they presented positive and negative thoughts at the same time. With a comprehensive explanation of the differences between the two groups of gifted students through a cognitive mechanism presented in the model of GLD model, this paper provides new approaches for identification and education of gifted students and GLD students.

• Journal of Korean Society for Atmospheric Environment
• /
• v.32 no.4
• /
• pp.372-383
• /
• 2016

This study aimed at analyzing the sensitivity of ozone simulation in accordance with the meteorological nudging for a high nocturnal ozone episode. To demonstrate the effectiveness of nudging methods (e.g., nudging techniques and application domains), the following six experiments were designed: (1) control without nudging, (2) experiment with application of observation nudging to all domains (domain 1~4), and (3)~(6) experiments with application of grid nudging to domain 1, domain 1~2, domain 1~3 and all domains, respectively. As a result, the meteorological nudging had a direct (improvement of input data) and indirect (estimate natural emission) effect on ozone simulation. Nudging effects during the daytime were greater than those during the nighttime due to low accuracy of wind direction during the nighttime. On comparison of the nudging techniques, the experiments in which grid nudging was applied showed more improved results than the experiments in which observation nudging was applied. At this time point, the simulated concentrations were generally similar to the observed concentrations due to the increase in the nudging effect when grid nudging was applied up to the sub-domain. However, for high nocturnal ozone uptakes, the experiment in which grid nudging was applied do domain 1~3 showed better results than the other experiments. This is because, when grid nudging was applied to the high resolution domain (e.g., domain 4 with 1 km), the local characteristics were removed due to the smoothing effects of meteorological conditions.