• 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.

• The Transactions of the Korean Institute of Electrical Engineers B
• /
• v.50 no.1
• /
• pp.10-14
• /
• 2001

The superconducting transmission cable is one of interesting part in power application using high temperature superconducting wire. One important parameter in HTS cable design is transport current sharing because it is related with current transmission capacity and loss. In this paper, we calculate self inductances of each layer and mutual inductances between two layers from magnetic field energy, and current sharing of each layer for 4-layer cable using the electric circuit model which contain inductance and resistance (by joint and AC loss). Also, transport current losses which are calculated by monoblock model and Norris equation are compared. As a results, outer layer has always larger transport current than inner layer, and current capacity of each layer is largely influenced by resistance per unit cable length. As a conclusion, for high current uniformity and low AC loss, we have to decrease inductances themselves or those differences.

• The Transactions of the Korean Institute of Power Electronics
• /
• v.8 no.6
• /
• pp.585-592
• /
• 2003

The parallel operation system of UPS is used to increase reliability of power source at critical load. But parallel UPS system has a few defects, impedance is different from each other and circulating current occurs between UPSs, due to line impedance and parameter variation, though controlled by the same synchronization signal. According to such characteristic of parallel UPS, balanced load-sharing control is the most important technique in parallel UPS operation. In this paper, a novel power deviation compensation algorithm is proposed. it is composed of voltage controller to compensate power deviation that be calculated by using active and reactive current deviation between Inverters on synchronous d-q reference frame.

• Proceedings of the Korea Database Society Conference
• /
• 2010.06a
• /
• pp.235-247
• /
• 2010

Introduction of ICE(Integrated Collaboration Environment) has a great effect on corporate performance. In particular, it is expected that information sharing and collaboration among transaction partners will be an important factor which mediates ICE and its performance. Under such a background, this research intends to identify the influence of ICE by manufacturing industry on corporate performance through role of the parameter of information sharing and collaboration. The result of this research may be summed up as follows: ICE promotes information sharing and collaboration among enterprises, it accelerates integration of supply chain network among enterprises and cooperation. In particular, ICE has a great effect on information sharing. And information sharing has a direct effect on collaboration and performance. However, when we classify corporate performance into the "performance due to introduction", which is direct performance, and the "overall performance", which is indirect performance, the result of this research is as follows: information sharing has a great effect on the performance due to introduction, but does not have a meaningful effect on indirect performance. On the other hand, collaboration does not have any effect on the performance due to introduction, but has a meaningful effect on the overall performance, which is indirect performance. It is necessary to study and analyze additional this result in future.

• Journal of information and communication convergence engineering
• /
• v.6 no.1
• /
• pp.43-46
• /
• 2008

This paper proposes an advanced test circuit for detecting the connectivity between a drum ring of laser printer and PCB. The detection circuit of charge sharing is proposed, which minimizes the influences of internal parasitic capacitances. The test circuit is composed of precharge circuit, analog comparator, level shifter. Its functional operation is verified using $0.6{\mu}m$ 3.3V/40V CMOS process parameter by HSPICE. Access time is100ns. Layout of the drum contact test circuit is $465{\mu}m\;{\times}\;117{\mu}m$.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.3
• /
• pp.389-395
• /
• 2022

Because of improvement of deep learning techniques, deep learning using computer vision such as classification, detection and segmentation has also been used widely at many fields. Expecially, automatic driving is one of the major fields that applies computer vision systems. Also there are a lot of works and researches to combine multiple tasks in a single network. In this study, we propose the network that predicts the individual depth of pedestrians and vehicles. Proposed model is constructed based on YOLOv3 for object detection and Monodepth for depth estimation, and it process object detection and depth estimation consequently using encoder and decoder based on hard parameter sharing. We also used attention module to improve the accuracy of both object detection and depth estimation. Depth is predicted with monocular image, and is trained using self-supervised training method.

• Journal of the Korean Society of Manufacturing Process Engineers
• /
• v.15 no.4
• /
• pp.67-72
• /
• 2016

Planetary gear sets are widely used in power transmission components, which have high efficiency and good durability. Their most important design parameter is the load-sharing characteristics among several planetary gears. In this study, the load sharing of planetary gears was analyzed according to the carrier pinhole position error of planetary gear sets. The loads acting on planetary gears varied with the pinhole position error of the carrier, and the load sharing of planetary gears improved as the input load increased. In addition, the load of the planetary gear with a carrier pinhole position error was relatively higher than that of other planetary gears without carrier pinhole position errors. This trend appeared more clearly in the non-floating-type carrier than the floating-type carrier.

• Structural Engineering and Mechanics
• /
• v.43 no.4
• /
• pp.439-448
• /
• 2012

This paper investigates the load sharing of double circular arc helical gears considering the influence of assembly errors. Based on a load sharing formulae, a three-dimensional finite element tooth contact analysis (TCA) is implemented with commercial software package ANSYS. The finite element grid for the double circular arc gear contact model is automatically generated by using the APDL (ANSYS Parameter Design Language) embedded in ANSYS. The realistic rotation of gears is achieved by using a coupling degree-of-freedom method. Numerical simulations are carried out to exemplify the proposed approach. The distribution of contact stress and bending stress under specific loading conditions are computed and compared with those obtained from Hertz contact theory and empirical formulae to demonstrate the efficiency of the proposed load sharing calculation formulae and TCA approach.

• Proceedings of the KIPE Conference
• /
• 1998.10a
• /
• pp.176-181
• /
• 1998

The parallel inverter is popularly used because of its fault-tolerance capability, high-current outputs at constant voltages and system modularity. The conventional parallel inverter usually employs active and reactive power control of frequency and voltage droop control. However, these approaches have the disadvantages that the response time of parallel inverter control is slow against load and system parameter variation to calculate active, reactive power, frequency and voltage. This paper describes a novel control scheme for power equalization in parallel-connected inverter. The proposed scheme has a fast power balance control response, a simplicity of implementation, and inherent peak current limiting capability since it employees an instantaneous current/voltage control with output voltage and current balance and output voltage regulation. A design procedure for the proposed parallel inverter controller is presented. Furthermore, the proposed control scheme is verified through the experiment in various cases such as the system parameter variation, the control parameter variation and the nonlinear load condition.

• KIEAE Journal
• /
• v.16 no.3
• /
• pp.35-46
• /
• 2016

Purpose: This paper aims to analyse the composition system of architectural elements including shape, kinetic and material elements of kinetic facades and establish the design parameter system as a common conceptual and practical knowledge sharing platform with mechanical and electrical experts. Method: This research has been conducted in a three steps. At first, 120 cases of external shading devices are analyzed and their classification criteria have been established. Secondly geometric, kinetic and material elements are categorized in a common kinetic facade coordinates system considering environmental effects and operation method, and the applicability of combination of each element are tested. Lastly core design parameters for each element have been established in a common office building installation coordinate. Result: Geometry elements are categorized into seven geometric shapes and kinetic elements is categorized into basic linear and rotational motion and combinational folding and rolling motion. The combined set of parameters for three elements composes the whole design parameters for architectural elements of kinetic façade. Design parameters of shape elements are composed of shape, installation and arrangement parameters; design parameters for kinetic elements are composed of axis and range parameters; and design parameters of material elements are composed of thermal, lighting and color parameters.