• The Transactions of the Korean Institute of Power Electronics
• /
• v.19 no.3
• /
• pp.290-296
• /
• 2014

This paper presents a design strategy for the control of the Z-source inverter (ZSI). For the Z-network capacitor voltage control, the average current model is derived to describe the dynamics of the voltage control and the controller outputs the average current command for the capacitor. Z-network inductor current reference is derived from the average current model of the Z-network capacitor. The inner current control loop outputs the average voltage command for the Z-network inductor and the shoot-through duty ratio of the ZSI is calculated from the output using the average voltage model of the Z-network inductor. The gain values of the current and voltage controllers are directly obtained by the Z-network parameters and desired bandwidth of each controller without a gain tuning process.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2018.06a
• /
• pp.63-66
• /
• 2018

Recently, a convolutional neural network (CNN) models at single image super-resolution have been very successful. Residual learning improves training stability and network performance in CNN. In this paper, we compare four convolutional neural network models for super-resolution (SR) to learn nonlinear mapping from low-resolution (LR) input image to high-resolution (HR) target image. Four models include general CNN model, global residual learning CNN model, local residual learning CNN model, and the CNN model with global and local residual learning. Experiment results show that the results are greatly affected by how skip connections are connected at the basic CNN network, and network trained with only global residual learning generates highest performance among four models at objective and subjective evaluations.

• The Journal of Korea Robotics Society
• /
• v.17 no.2
• /
• pp.133-141
• /
• 2022

This paper introduces model compression algorithms which make a deep neural network smaller and faster for embedded systems. The model compression algorithms can be largely categorized into pruning, quantization and knowledge distillation. In this study, gradual pruning, quantization aware training, and knowledge distillation which learns the activation boundary in the hidden layer of the teacher neural network are integrated. As a large deep neural network is compressed and accelerated by these algorithms, embedded computing boards can run the deep neural network much faster with less memory usage while preserving the reasonable accuracy. To evaluate the performance of the compressed neural networks, we evaluate the size, latency and accuracy of the deep neural network, DenseNet201, for image classification with CIFAR-10 dataset on the NVIDIA Jetson Xavier.

• The KIPS Transactions:PartC
• /
• v.8C no.1
• /
• pp.68-74
• /
• 2001

This paper mentions a centralized approach based on SNMP protocol and distributed approach based on mobile agent in network management s system. And it presents a few Quantitative models for systematically evaluating those two different approaches. To do this, we propose model m that is applicable under a uniform network environment, and compare network execution times of each paradigms based on parameters from s simulation. The model is then refined to take into account non-uniform networks. We show that it can reduce overall network execution times b by determining the best interaction patterns to perfo$\pi$n network management operations from this model. We believe that the model proposed h here should help us to decide appropriate paradigms and interaction patterns for developing network management applications.

• Journal of Korea Society of Digital Industry and Information Management
• /
• v.11 no.4
• /
• pp.33-45
• /
• 2015

Currently, Internet is used an essential tool in the business area. Despite this importance, there is a risk of network attacks attempting collection of fraudulence, private information, and cyber terrorism. Firewalls and IDS(Intrusion Detection System) are tools against those attacks. IDS is used to determine whether a network data is a network attack. IDS analyzes the network data using various techniques including expert system, data mining, and state transition analysis. This paper tries to compare the performance of two data mining models in detecting network attacks. They are decision tree (C4.5), and neural network (FANN model). I trained and tested these models with data and measured the effectiveness in terms of detection accuracy, detection rate, and false alarm rate. This paper tries to find out which model is effective in intrusion detection. In the analysis, I used KDD Cup 99 data which is a benchmark data in intrusion detection research. I used an open source Weka software for C4.5 model, and C++ code available for FANN model.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.9 no.2
• /
• pp.680-699
• /
• 2015

The use of web service has been increased rapidly, with an increase in the number of available services, finding the exact service is the challenging task. Service discovery is the most significant job to complete the service discoverers needs. In order to achieve the efficient service discovery, we focus on designing a cluster based service discovery model for service registering and service provisioning among all mobile nodes in a mobile ad hoc network (MANETs). A dynamic backbone of nodes (i.e. cluster heads) that forms a service repository to which MANET nodes can publish their services and/or send their service queries. The designed model is based on storing services with their service description on cluster head nodes that are found in accordance with the proposed cluster head election model. In addition to identifying and analyzing the system parameters for finding the effectiveness of our model, this paper studies the stability analysis of the network, overhead of the cluster, and bandwidth utilization and network traffic is evaluated using analytic derivations and experimental evaluation has been done.

• Transactions of the Korean Society of Automotive Engineers
• /
• v.12 no.4
• /
• pp.110-118
• /
• 2004

In this paper, a blackbox approach is carried out to model the nonlinear dynamic bushing model. One-axis durability test is performed to describe the mechanical behavior of typical vehicle elastomeric components. The results of the tests are used to develop an empirical bushing model with an artificial neural network. The back propagation algorithm is used to obtain the weighting factor of the neural network. Since the output for a dynamic system depends on the histories of inputs and outputs, Narendra's algorithm of ‘NARMAX’ form is employed in the neural network bushing module. A numerical example is carried out to verify the developed bushing model.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.8 no.9
• /
• pp.3110-3125
• /
• 2014

Network coding improves TCP's performance in lossy wireless networks. However, the complex congestion window evolution of network coded TCP (TCP-NC) makes the analysis of end-to-end throughput challenging. This paper analyzes the evolutionary process of TCP-NC against lossy links. An analytic model is established by applying a two-dimensional Markov chain. With maximum window size, end-to-end erasure rate and redundancy parameter as input parameters, the analytic model can reflect window evolution and calculate end-to-end throughput of TCP-NC precisely. The key point of our model is that by the novel definition of the states of Markov chain, both the number of related states and the computation complexity are substantially reduced. Our work helps to understand the factors that affect TCP-NC's performance and lay the foundation of its optimization. Extensive simulations on NS2 show that the analytic model features fairly high accuracy.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.16 no.6
• /
• pp.785-790
• /
• 2006

In RFID-based SCM, The traceability and product information is the important target data. In this paper, efficient items traceability model and the integrated model of the product between RFID network and GDS(Global Data Synchronization) network are studied. Information consists of the dynamic data generated from RFID network and static data generated from GDS Network. The integrated model will provide the interoperability between 2 kinds of networks.

• KIPS Transactions on Software and Data Engineering
• /
• v.6 no.4
• /
• pp.203-210
• /
• 2017

In this paper, we propose an effective neural network model for image caption generation and model transfer. This model is a kind of multi-modal recurrent neural network models. It consists of five distinct layers: a convolution neural network layer for extracting visual information from images, an embedding layer for converting each word into a low dimensional feature, a recurrent neural network layer for learning caption sentence structure, and a multi-modal layer for combining visual and language information. In this model, the recurrent neural network layer is constructed by LSTM units, which are well known to be effective for learning and transferring sequence patterns. Moreover, this model has a unique structure in which the output of the convolution neural network layer is linked not only to the input of the initial state of the recurrent neural network layer but also to the input of the multimodal layer, in order to make use of visual information extracted from the image at each recurrent step for generating the corresponding textual caption. Through various comparative experiments using open data sets such as Flickr8k, Flickr30k, and MSCOCO, we demonstrated the proposed multimodal recurrent neural network model has high performance in terms of caption accuracy and model transfer effect.