• KIPS Transactions on Software and Data Engineering
• /
• v.9 no.5
• /
• pp.169-176
• /
• 2020

Recently, deep neural networks (DNNs) have brought revolutions to many mobile and embedded devices by providing human-level machine intelligence for various applications. However, high inference accuracy of such DNNs comes at high computational costs, and, hence, there have been significant efforts to reduce computational overheads of DNNs either by compressing off-the-shelf models or by designing a new small footprint DNN architecture tailored to resource constrained devices. One notable recent paradigm in designing small footprint DNN models is sharing parameters in several layers. However, in previous approaches, the parameter-sharing techniques have been applied to large deep networks, such as ResNet, that are known to have high redundancy. In this paper, we propose a parameter-sharing method for already parameter-efficient small networks such as ShuffleNetV2. In our approach, small templates are combined with small layer-specific parameters to generate weights. Our experiment results on ImageNet and CIFAR100 datasets show that our approach can reduce the size of parameters by 15%-35% of ShuffleNetV2 while achieving smaller drops in accuracies compared to previous parameter-sharing and pruning approaches. We further show that the proposed approach is efficient in terms of latency and energy consumption on modern embedded devices.

• Journal of the Korean Society of Environmental Restoration Technology
• /
• v.10 no.2
• /
• pp.61-70
• /
• 2007

This study aims to clarify the effect of forest management practices(thinning and pruning) on soil physical properties and water quality to get the fundamental information on the facility of purifying water quality after forestry practices. Rainfall, throughfall, stemflow, soil and stream water were sampled at the study site which consists of Pinus rigida in Jinju National University Experimental Forest for 4 years from Mar. 1, 2002 to Nov. 30, 2006. Averaged tree height of the management site increased by 1.6m, compared to the value of the non-management site in Pinus rigida. Increment of averaged D. B. H. at the management site showed 4.2cm more in Pinus rigida compared with that at non-management sites. Mesopore ratios (pF2.7) and total porosities of A layer soil at the management site increased more than those at the non-management sites in both stands. Otherwise, soil bulk density resulted in being reversely. Water qualities of throughfall, stemflow and buffered soil water were influenced more positively by the management practice. The average electrical conductivity of stream water was $32.9{\mu}S/cm$ within the range of non-polluted stream water.

• Journal of Korea Society of Industrial Information Systems
• /
• v.22 no.5
• /
• pp.17-24
• /
• 2017

Recently Deep-learning, Technologies using Large or Deep Artificial Neural Networks, have Shown Remarkable Performance, and the Increasing Size of the Network Contributes to its Performance Improvement. However, the Increase in the Size of the Neural Network Leads to an Increase in the Calculation Amount, which Causes Problems Such as Circuit Complexity, Price, Heat Generation, and Real-time Restriction. In This Paper, We Propose and Test a Method to Reduce the Number of Network Connections by Effectively Pruning the Redundancy in the Connection and Showing the Difference between the Performance and the Desired Range of the Original Neural Network. In Particular, we Proposed a Simple Method to Improve the Performance by Re-learning and to Guarantee the Desired Performance by Allocating the Error Rate per Layer in Order to Consider the Difference of each Layer. Experiments have been Performed on a Typical Neural Network Structure such as FCN (full connection network) and CNN (convolution neural network) Structure and Confirmed that the Performance Similar to that of the Original Neural Network can be Obtained by Only about 1/10 Connection.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2006.11a
• /
• pp.127-132
• /
• 2006

Quantization in video coding plays an important role in controlling the bit-rate of compressed video bit-streams. It has been used as an important control means to adjust the amount of bit-streams to at]owed bandwidth of delivery networks and storage. Due to the dependent nature of video coding, dependent quantization has been proposed and applied for MPEG-2 video coding to better maintain the quality of reconstructed frame for given constraints of target bit-rate. Since Scalable Video Coding (SVC) being currently standardized exhibits highly dependent coding nature not only between frames but also lower and higher scalability layers where the dependent quantization can be effectively applied, in this paper, we propose a dependent quantization scheme for SVC and compare its performance in visual qualities and bit-rates with the current JSVM reference software for SVC. The proposed technique exploits the frame dependences within each GOP of SVC scalability layers to formulate dependent quantization. We utilize Lagrange optimization, which is widely accepted in R-D (rate-distortion) based optimization, and construct trellis graph to find the optimal cost path in the trellis by minimizing the R-D cost. The optimal cost path in the trellis graph is the optimal set of quantization parameters (QP) for frames within a GOP. In order to reduce the complexity, we employ pruning procedure using monotonicity property in the trellis optimization and cut the frame dependency into one GOP to decrease dependency depth. The optimal Lagrange multiplier that is used for SVC is equal to H.264/AVC which is also used in the mode prediction of the JSVM reference software. The experimental result shows that the dependent quantization outperforms the current JSVM reference software encoder which actually takes a linear increasing QP in temporal scalability layers. The superiority of the dependent quantization is achieved up to 1.25 dB increment in PSNR values and 20% bits saving for the enhancement layer of SVC.

• Korean Journal of Environment and Ecology
• /
• v.23 no.2
• /
• pp.208-219
• /
• 2009

This study is to establish a management method for conservation through comparison and analysis on vegetation structures of Pinus densiflora forest around Beopjusa area for past 17-year. The spatial range of the study was $3.6km^2$ from maintenance office to Beopjusa area. The analysis results of the actual vegetation showed that the ratio of vegetation were composed of 64.7% of Pinus densiflora forest, 3.2% of mixed forest of P. densiflora and deciduous broadleaf trees and 5.9% of deciduous broadleaf tree community out of overall area, 360ha. The type of P. densiflora forest were categorized into four communities; community having high potential of succession, community having low potential of it, the community being in the process of succession and community being in the process of natural selection. The succession tendency was in order of the community having low potential of succession(P. densiflora forest), having high potential of it(P. densiflora forest which is deciduous broadleaf trees are dominating in sub-canopy layer), being in the process of succession(P. densiflora-Prunus sargentii and P. densiflora-Quercus serrata community) and being in the process of natural selection(Q. serrata-P. densiflora and Q. aliena-P. densiflora community). In terms of vegetation management, P. densiflora forest having high potential of succession was needed to remove deciduous broadleaf trees in the sub-canopy layer and the community being in the process of succession was required to be pruning the branch in the canopy layer. Lastly, the community being in the process of natural selection was suggested to let it be in succession, since it is hard to be in the status of P. densiflora Forest.

• Horticultural Science & Technology
• /
• v.33 no.4
• /
• pp.479-485
• /
• 2015

This study was conducted to study the factors inducing micro-cracking and determine proper practices to reduce its occurrence in 'Mansoo' pear fruits. Micro-cracking was observed on ripe fruits. Occurrence of micro-cracking was closely related to sun duration time in August. Micro-cracking occurred severely with shorter sun duration, but weakly with longer sun duration and continuous sunlight. Micro-cracking fruits occurred more frequently in trees with a pergola training system than in those with Y-trellis, but there was no difference between the irrigated and non-irrigated groups. While no micro-cracking was observed without fruit bagging, micro-cracking occurred in fruits with black and yellow outer paper bagging at rates of 62.2 and 17.3%, respectively. Our results suggest that the light condition was the most important cause of micro-cracking because the occurrence of micro-cracking in 'Mansoo' fruits was affected by factors including sun duration, trellis system and fruit bag color. We suggested that micro-cracking could be reduced in 'Mansoo' fruit by optimizing the light conditions with the proper trellis system, bagging color and summer pruning.

• Journal of Korean Society of Forest Science
• /
• v.88 no.3
• /
• pp.364-373
• /
• 1999

This study aims to clarify the effect of forest management practices(thinning and pruning) on soil physical properties and water quality to get the fundamental information on the facility of purifying water quality after forestry practices. Rainfall, throughfall, stemflow, soil and stream water were sampled at the study sites which consist of Abies holophylla and Pinus koraiensis, in Kwangnung Experimental Forest for 6 months from March 1 to August 7, 1998. Average tree height of the management site increased by 1.8m and 0.6m more than that of the non-management site in Abies holophylla and Pinus koraiensis, respectively. Increment of average D.B.H. at the management site showed 4.7cm and 1.4cm more in Abies holophylla and Pinus koraiensis compared with that at non-management sites. Coarse(less than pF2.7) and total porosities of A layer soil at the management site increased more than those at the non-management sites in both stands. Otherwise, soil bulk density resulted in being reversely. Water qualities of throughfall, stemflow and soil water were buffered more by the management practice in both.

• Korean Journal of Environment and Ecology
• /
• v.23 no.6
• /
• pp.528-534
• /
• 2009

Styan's Grasshopper Warblers (Locustella pleskei) are vulnerable species distributed in East-Asia only. Its known breeding sites in Korea are remote islets including Hongdo Islet, Chilbal Islet, Mara Islet, and Chuja Islets, and, therefore, the breeding biology of this species is still poorly known. This study was conducted on Mara Islet (N $33^{\circ}$ 06', E $126^{\circ}$ 16') from May to September 2008 to investigate the breeding status and breeding site characteristics of the grasshopper warblers. A total of 11 breeding pairs and their nests were found on trees and shrubs at artificially planted forests and hedges of Pinus thunbergii. The grasshopper warblers preferentially utilized the shrub trees for nesting places, and major nesting trees were Camellia japonica, Pittosporum tobira and Pinus thunbergii as nesting trees. Average heights of nesting trees and nests were $2.77{\pm}1.10m$ and $1.75{\pm}0.56m$, respectively. The grasshopper warblers selected lower shrubs and trees for nesting than randomly selected ones around them, probably to avoid strong and prevailed winds in flat and un vegetated environments on Mara Islet. The shape of nests was a round bowl-type, and measurements of nests were $11.9{\pm}0.5cm$ in exterior nest diameter, $11.1{\pm}1.1cm$ in height of exterior nest, $5.8{\pm}0.4cm$ in interior nest depth, and 6.0cm in interior nest diameter. It incubated eggs until the early August on Mara islet, and incubation periods of Mara Islet was possibly later than that of other areas. Furthermore, the clutch size in the study area was three, and they laid smaller number of eggs than normal clutch size (4~5 eggs) reported in other areas. Although we could not observe any nest predator on this species in the study area, selective cutting and pruning of trees will diminish dense shrub layer of forests. Therefore, it may affect the breeding of this threatened species which prefers dense shrubs of artificially planted forest of Pinus thunbergii. This study suggests that detailed and consistent further research on breeding biology and habitats of the grasshopper warblers are needed to conserve and manage of Pinus thunbergii forests on Mara Islet as an important breeding site of Styan's Grasshopper Warblers.

• Journal of agriculture & life science
• /
• v.50 no.1
• /
• pp.105-115
• /
• 2016

This study was carried out to establish the optimum forest management plan for the Cryptomeria japonica plantations in southern inland and Jeju island in Korea. Sixty seven circular sample plots of 0.04ha were established and we surveyed vegetation structure and growth characteristics from three layers(upper, middle, and lower). As a result of cluster analysis obtained by importance values of each tree species, the community type of C. japonica stands were classified into C. japonica group(C1) and C. japonica-C obtusa group. C. obtusa community were also sbudivided into P. thunbergii-Q. serrata group(C2) and Q. serrata-C obtusa group(C3). In tree layers importance value(IV) of C. japonica were 97.2% in C1, 80.7% in C2, and 47.6% in C3 and in sub-tree layers IV were 8.9% in C1, 15.2% in C2, and 5.7% in C3. Especially in C3 there are bamboo species (Smilacina japonica var. lutecarpa and Pseudosasa japonica) it is necessary for us to control them. In shrub layers C. japonica were found in C1(9.2%) and C2(7.0%), but except for C3. In tree layer species diversity indices of each community ranged from the lowest 0.059 in C1 to the highest 0.548 in C3. Dominance ranged from 0.958 in C1 to 0.393 in C3 which may caused by interspecific competition. Current annual increment of diameter growth ranged from 7.01mm/yr to 8.04mm/yr. As a result of our study we recommend the application of proper thinning and pruning for C1 and C2.