• Convergence Security Journal
• /
• v.23 no.1
• /
• pp.27-33
• /
• 2023

Digitalization due to COVID-19 has rapidly developed artificial intelligence-based voice recognition technology. However, this technology causes unfair social problems, such as race and gender discrimination if datasets are biased against some groups, and degrades the reliability and security of artificial intelligence services. In this work, we compare and analyze accuracy-based unfairness in biased data environments using VGGNet (Visual Geometry Group Network), ResNet (Residual Neural Network), and MobileNet, which are representative CNN (Convolutional Neural Network) models of artificial intelligence. Experimental results show that ResNet34 showed the highest accuracy for women and men at 91% and 89.9%in Top1-accuracy, while ResNet18 showed the slightest accuracy difference between genders at 1.8%. The difference in accuracy between genders by model causes differences in service quality and unfair results between men and women when using the service.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.16 no.6
• /
• pp.2044-2059
• /
• 2022

Recent works have shown that the geometric constraint can be harnessed to boost the performance of CNN-based camera localization. However, the existing strategies are limited to imposing image-level constraint between pose pairs, which is weak and coarse-gained. In this paper, we introduce a pixel-level epipolar geometry constraint to vanilla localization framework without the ground-truth 3D information. Dubbed EpiLoc, our method establishes the geometric relationship between pixels in different images by utilizing the epipolar geometry thus forcing the network to regress more accurate poses. We also propose a variant called EpiSingle to cope with non-sequential training images, which can construct the epipolar geometry constraint based on a single image in a self-supervised manner. Extensive experiments on the public indoor 7Scenes and outdoor RobotCar datasets show that the proposed pixel-level constraint is valuable, and helps our EpiLoc achieve state-of-the-art results in the end-to-end camera localization task.

• Journal of the Korean Society of Manufacturing Technology Engineers
• /
• v.8 no.5
• /
• pp.70-75
• /
• 1999

Adaptive control in the robotic GMA(Gas Metal Arc) welding is employed to monitor the information about weld characteristics and process paramters as well as modification of those parameters to hold weld quality within the acceptable limits. Typical characteristics are the bead geometry composition micrrostructure appearance and process parameters which govern the quality of the final weld. The main objectives of this paper are to realize the mapping characteristicso f penetration through the learning. After learning the neural network can predict the pene-traition desired from the learning mapping characteristic. The design parameters of the neural network estimator(the number of hidden layers and the number of nodes in a layer) were chosen from an error analysis. partial-penetration single-pass bead-on-plate welds were fabricated in 12mm mild steel plates in order to verify the performance of the neural network estimator. The experimental results show that the proposed neural network estimator can predict the penetration with reasonable accuracy and gurarantee the uniform weld quality.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.9 no.2
• /
• pp.483-503
• /
• 2015

This paper analyzes the optimal user density and power allocation for Device-to-Device (D2D) communication underlaying cellular networks on multiple bands with the target of maximizing the D2D transmission capacity. The entire network is modeled by Poisson point process (PPP) which based on stochastic geometry. Then in order to ensure the outage probabilities of both cellular and D2D communication, a sum capacity optimization problem for D2D system on multiple bands is proposed. Using convex optimization, the optimal D2D density is obtained in closed-form when the D2D transmission power is determined. Next the optimal D2D transmission power is obtained in closed-form when the D2D density is fixed. Based on the former two conclusions, an iterative algorithm for the optimal D2D density and power allocation on multiple bands is proposed. Finally, the simulation results not only demonstrate the D2D performance, density and power on each band are constrained by cellular communication as well as the interference of the entire system, but also verifies the superiority of the proposed algorithm over sorting-based and removal algorithms.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.53 no.2
• /
• pp.133-140
• /
• 2016

In active sonar field, a target detection and localization based on a distributed sensor network has been much studied for the underwater surveillance of the coast. Zhou et al. proposed a target localization method utilizing the positions of target-detected sensors in distributed sensor network which consists of detection-only sensors. In contrast with a conventional method, Zhou's method dose not require to estimate the propagation model parameters of detection signal. Also it needs the lower computational complexity, and to transmit less data between network nodes. However, it has large target localization error. So it has been modified for reducing localization error by Ryu. Modified Zhou's method has better estimation performance than Zhou's method, but still relatively large estimation error. In this paper, a target localization method based on modified Zhou's method is proposed for reducing the localization error. The proposed method utilizes the geometry of the positions of target-detected sensors and a line that represents the bearing of target, a line can be found by modified Zhou's method. This paper shows that the proposed method has better target position estimation performance than Zhou's and modified Zhou's method by computer simulations.

• 제어로봇시스템학회:학술대회논문집
• /
• 2003.10a
• /
• pp.592-597
• /
• 2003

In this paper, some differential geometric conditions for the observer using a recurrent neural network are provided in terms of a stochastic nonlinear system control. In the stochastic nonlinear system, it is necessary to make an additional condition for observation of stochastic nonlinear system, called perfect filtering condition. In addition, we provide a observer using a recurrent neural network for the observation of a stochastic nonlinear system with the proposed observation conditions. Computer simulation shows that the control performance of the stochastic nonlinear system with a observer using a recurrent neural network satisfying the proposed conditions is more efficient than the conventional observer as Kalman filter

• Proceedings of the Korean Institute of Navigation and Port Research Conference
• /
• v.1
• /
• pp.505-510
• /
• 2006

SVLBI (Space Very Long Baseline Interferometry) has some important potential applications in geodesy and geodynamics, for which one of the most difficult tasks is to precisely determine the orbit of SVLBI satellite. This paper studies several technologies which possibly will be able to determine the orbit of space VLBI satellite. And then, according to the sorts and characteristicsof satellite and the requirements for geodetic study and the geometry of GNSS (GPS, GALILEO) satellite to track the space VLBI satellite, the six Keplerian elements of SVLBI satellite (TEST-SVLBI) are determined. A program is designed to analyze the coverage area of the space of different heights by the stations of the network, with which the tracking network of TEST-SVLBI is designed. The efficiency of tracking TEST-SVLBI by the network is studied, and the results are presented.

• Tunnel and Underground Space
• /
• v.25 no.5
• /
• pp.450-461
• /
• 2015

Extensive numerical experiments have been carried out to investigate effect of block size and fracture geometry on hydraulic characteristics of fractured rock masses based on connected pipe flow in DFN systems. Using two fracture sets, a total of 72 2-D fracture configurations were generated with different combinations of fracture size distribution and deterministic fracture density. The directional block conductivity including the theoretical block conductivity, principal conductivity tensor and average block conductivity for each generated fracture network system were calculated using the 2-D equivalent pipe network method. There exist significant effects of block size, orientation, density and size of fractures in a fractured rock mass on its hydraulic behavior. We have been further verified that it is more difficult to reach the REV size for the fluid flow network with decreasing intersection angle of two fracture sets, fracture plane density and fracture size distribution.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.11 no.11
• /
• pp.5301-5323
• /
• 2017

Small cell deployment offers a low-cost solution for the boosted traffic demand in heterogeneous cellular networks (HCNs). Besides improved spatial spectrum efficiency and energy efficiency, future HCNs are also featured with the trend of network architecture convergence and feasibility for flexible mobile applications. To achieve these goals, dual connectivity (DC) is playing a more and more important role to support control/user-plane splitting, which enables maintaining fixed control channel connections for reliability. In this paper, we develop a tractable framework for the downlink SINR analysis of DC assisted HCN. Based on stochastic geometry model, the data-control joint coverage probabilities under multi-frequency and single-frequency tiering are derived, which involve quick integrals and admit simple closed-forms in special cases. Monte Carlo simulations confirm the accuracy of the expressions. It is observed that the increase in mobility robustness of DC is at the price of control channel SINR degradation. This degradation severely worsens the joint coverage performance under single-frequency tiering, proving multi-frequency tiering a more feasible networking scheme to utilize the advantage of DC effectively. Moreover, the joint coverage probability can be maximized by adjusting the density ratio of small cell and macro cell eNBs under multi-frequency tiering, though changing cell association bias has little impact on the level of the maximal coverage performance.

• International Journal of Computer Science & Network Security
• /
• v.24 no.2
• /
• pp.53-58
• /
• 2024

This paper presents a novel miniaturized 3-D cubic antenna for use in wireless sensor network (WSN) application. The geometry of this antenna is designed as a cube including a meander dipole antenna. A truly omnidirectional pattern is produced by this antenna in both E-plane and H-plane, which allows for non-intermittent communication that is orientation independent. The operating frequency lies in the ISM band (centered in 2.45 GHz). The dimensions of this ultra-compact cubic antenna are 1.25*1.12*1cm3 which features a length dimension λ/11. The coefficient which presents the overall antenna structure is Ka=0.44. The cubic shape of the antenna is allowing for smart packaging, as sensor equipment may be easily integrated into the cube hallow interior. The major constraint of WSN is the energy consumption. The power consumption of radio communication unit is relatively high. So it is necessary to design an antenna which improves the energy efficiency. The parameters considered in this work are the resonant frequency, return loss, efficiency, bandwidth, radiation pattern, gain and the electromagnetic field of the proposed antenna. The specificity of this geometry is that its size is relatively small with an excellent gain and efficiency compared to previously structures (reported in the literature). All results of the simulations were performed by CST Microwave Studio simulation software and validated with HFSS. We used Advanced Design System (ADS) to validate the equivalent scheme of our conception. Input here the part of summary.