• 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.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2022.05a
• /
• pp.517-519
• /
• 2022

In 5G network environment, proactive mobility management is essential as 5G mobile networks provide new services with ultra-low latency through dense deployment of small cells. The importance of a system that actively controls device handover is emerging and it is essential to predict mobile trajectory during handover. Sequence-to-sequence model is a kind of deep learning model where it converts sequences from one domain to sequences in another domain, and mainly used in natural language processing. In this paper, we developed a system for predicting mobile trajectory in a wireless network environment using sequence-to-sequence model. Handover speed can be increased by utilize our sequence-to-sequence model in actual mobile network environment.

• ETRI Journal
• /
• v.45 no.4
• /
• pp.666-677
• /
• 2023

In the last few years, convolutional neural networks (CNNs) have demonstrated good performance while solving various computer vision problems. However, since CNNs exhibit high computational complexity, signal processing is performed on the server side. To reduce the computational complexity of CNNs for edge computing, a lightweight algorithm, such as a MobileNet, is proposed. Although MobileNet is lighter than other CNN models, it commonly achieves lower classification accuracy. Hence, to find a balance between complexity and accuracy, additional hyperparameters for adjusting the size of the model have recently been proposed. However, significantly increasing the number of parameters makes models dense and unsuitable for devices with limited computational resources. In this study, we propose a novel MobileNet architecture, in which the number of parameters is adaptively increased according to the importance of feature maps. We show that our proposed network achieves better classification accuracy with fewer parameters than the conventional MobileNet.

• Journal of Broadcast Engineering
• /
• v.25 no.2
• /
• pp.242-251
• /
• 2020

Recently, many deep convolutional neural networks for image super-resolution have been studied. Existing deep learning-based super-resolution algorithms are architecture that up-samples the resolution at the end of the network. The post-upsampling architecture has an inefficient structure at large scaling factor result of predicting a lot of information for mapping from low-resolution to high-resolution at once. In this paper, we propose a single image super-resolution using Channel Attention Residual Dense Block based on an iterative up-down sampling architecture. The proposed algorithm efficiently predicts the mapping relationship between low-resolution and high-resolution, and shows up to 0.14dB performance improvement and enhanced subjective image quality compared to the existing algorithm at large scaling factor result.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.19 no.6
• /
• pp.1307-1312
• /
• 2015

As smart-phones become popular, mobile data traffic has been dramatically increasing and intensive researches on the next-generation mobile communication network is in progress to meet the increasing demand for mobile data traffic. In particular, heterogeneous network (HetNet) is attracting much interest because it can significantly enhance the network capacity by increasing the spatial reuse with macro and small cells. In the HetNet, we have several problems such as load imbalance and interference because of the difference in transmit power between macro and small cells and cell range expansion (CRE) can mitigate the problems. In this paper, we propose a new cell selection scheme with adaptive cell range expansion bias (CREB) for ultra dense HetNet and we analyze the performance of the proposed scheme in terms of average cell transmission rate through system-level simulations and compare it with those of other schemes.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2015.05a
• /
• pp.63-66
• /
• 2015

As smart-phones become popular, mobile data traffic has been dramatically increasing and intensive researches on the next-generation mobile communication network is in progress to meet the increasing demand for mobile data traffic. In particular, heterogeneous network (HetNet) is attracting much interest because it can significantly enhance the network capacity by increasing the spatial resue with macro and small cells. In the HetNet, we have several problems such as load imbalance and interference because of the difference in transmit power between macro and small cells and cell range expansion (CRE) can mitigate the problems. In this paper, we propose a new cell selection scheme with adaptive cell range expansion bias (CREB) for ultra dense HetNet and we analyze the performance of the proposed scheme in terms of average cell transmission rate through system-level simulations and compare it with those of other schemes.

• Journal of the Korea Convergence Society
• /
• v.9 no.9
• /
• pp.47-52
• /
• 2018

Wi-Fi enabled Internet of Things (IoTs) had a substantial impact on society, economy and industry. However wireless connectivity technologies in unlicensed band such as Wi-Fi are vulnerable to interferences. They also face difficulty providing wireless connectivity over long range in dense networks due to the dynamically changed interference effect and asymmetric interference conditions. In this paper, robust acknowledgement transmission scheme is proposed for long range IoTs. According to the proposed scheme, it is possible to control the transmission rate of the transmission success rate of the response frame by adjusting the transmission rate of the response frame when the interference is present asymmetrically. It is also possible to use higher data rate when high quality link is guaranteed. The evaluation results demonstrated the proposed scheme improves the aggregate throughput by at most 9 Mbps when 20 MHz bandwidth transmission mode was adopted.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2018.10a
• /
• pp.464-466
• /
• 2018

In this paper, a dense cellular network is considered in which each base station equipped with multiple antennas simultaneously communicates with multiple single-antenna users. Based on limited feedback, each user feeds back its quantized channel state information (CSI) to its associated transmitter, and the transmitter broadcasts multiple data streams prepared for the scheduled users using a space-division multiple access scheme. As the amount of CSI is limited at the transmitter, the downlink throughput increases with the number feedback bits. However, the increased number of feedback bits requires the correspondingly increased amount of uplink resources. Thus, an appropriate balance between the downlink throughput and the uplink resource usage should be considered in realistic systems. A net spectral efficiency defined in this context is used in this paper, and the optimal number of feedback bits that maximizes the net spectral efficiency is analyzed. This paper particularly focuses on the case when the received signal power is much smaller than the noise power.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.8 no.6
• /
• pp.1369-1375
• /
• 2007

The recent, a large capacity of telecommunication networks is required in order to it is in proportion to capacity of information communication increase and to satisfy a demand because of the demand about Internet, a multimedia service of internet, Video of internet protocol(VoIP), Audio/Video streaming. As a result, DWDM(Dense Wavelength Division Multiplexing)technologies are emerging to be a prevailing the method of solving it without additional optical fiber network building and high-speed equipment. Therefore this thesis proposed the optical filter of fiber/multilayer slab coupled structure combining it to multilayer slab waveguide by polishing the cladding of one side of fiber to design the optical filter having these functions. When a separation distance of fiber and slab was $3{\mu}m$, The optical filter proposed as the simulation result was satisfied with a DWDM filter characteristic with FWHM of 0.1nm on TM mode and TE mode as 32nm polarization independence in a communication window of $1.3{\mu}m$ when center wavelength was each ${\lambda}_0=1.274755{\mu}m$ and ${\lambda}_0=1.30591{\mu}m$.

• Imaging Science in Dentistry
• /
• v.52 no.3
• /
• pp.239-244
• /
• 2022

Purpose: Despite the proliferation of numerous morphometric and anthropometric methods for sex identification based on linear, angular, and regional measurements of various parts of the body, these methods are subject to error due to the observer's knowledge and expertise. This study aimed to explore the possibility of automated sex determination using convolutional neural networks(CNNs) based on lateral cephalometric radiographs. Materials and Methods: Lateral cephalometric radiographs of 1,476 Iranian subjects (794 women and 682 men) from 18 to 49 years of age were included. Lateral cephalometric radiographs were considered as a network input and output layer including 2 classes(male and female). Eighty percent of the data was used as a training set and the rest as a test set. Hyperparameter tuning of each network was done after preprocessing and data augmentation steps. The predictive performance of different architectures (DenseNet, ResNet, and VGG) was evaluated based on their accuracy in test sets. Results: The CNN based on the DenseNet121 architecture, with an overall accuracy of 90%, had the best predictive power in sex determination. The prediction accuracy of this model was almost equal for men and women. Furthermore, with all architectures, the use of transfer learning improved predictive performance. Conclusion: The results confirmed that a CNN could predict a person's sex with high accuracy. This prediction was independent of human bias because feature extraction was done automatically. However, for more accurate sex determination on a wider scale, further studies with larger sample sizes are desirable.