• Journal of Internet Computing and Services
• /
• v.24 no.5
• /
• pp.17-27
• /
• 2023

Today, as AI (Artificial Intelligence) technology develops and its practicality increases, it is widely used in various application fields in real life. At this time, the AI model is basically learned based on various statistical properties of the learning data and then distributed to the system, but unexpected changes in the data in a rapidly changing data situation cause a decrease in the model's performance. In particular, as it becomes important to find drift signals of deployed models in order to respond to new and unknown attacks that are constantly created in the security field, the need for lifecycle management of the entire model is gradually emerging. In general, it can be detected through performance changes in the model's accuracy and error rate (loss), but there are limitations in the usage environment in that an actual label for the model prediction result is required, and the detection of the point where the actual drift occurs is uncertain. there is. This is because the model's error rate is greatly influenced by various external environmental factors, model selection and parameter settings, and new input data, so it is necessary to precisely determine when actual drift in the data occurs based only on the corresponding value. There are limits to this. Therefore, this paper proposes a method to detect when actual drift occurs through an Anomaly analysis technique based on XAI (eXplainable Artificial Intelligence). As a result of testing a classification model that detects DGA (Domain Generation Algorithm), anomaly scores were extracted through the SHAP(Shapley Additive exPlanations) Value of the data after distribution, and as a result, it was confirmed that efficient drift point detection was possible.

• Journal of the Korean Society of Hazard Mitigation
• /
• v.3 no.3 s.10
• /
• pp.151-163
• /
• 2003

In this study, the algorithm of groundwater flow process was established for koreanized groundwater program development dealing with the geographic and geologic conditions of the aquifer have dynamic behaviour in groundwater flow system. All the input data settings of the 3-DFM model which is developed in this study are organized in Korean, and the model contains help function for each input data. Thus, it is designed to get detailed information about each input parameter when the mouse pointer is placed on the corresponding input parameter. This model also is designed to easily specify the geologic boundary condition for each stratum or initial head data in the work sheet. In addition, this model is designed to display boxes for input parameter writing for each analysis condition so that the setting for each parameter is not so complicated as existing MODFLOW is when steady and unsteady flow analysis are performed as well as the analysis for the characteristics of each stratum. Descriptions for input data are displayed on the right side of the window while the analysis results are displayed on the left side as well as the TXT file for this results is available to see. The model developed in this study is a numerical model using finite differential method, and the applicability of the model was examined by comparing and analyzing observed and simulated groundwater heads computed by the application of real recharge amount and the estimation of parameters. The 3-DFM model is applied in this study to Sehwa-ri, and Songdang-ri area, Jeju, Korea for analysis of groundwater flow system according to pumping, and obtained the results that the observed and computed groundwater head were almost in accordance with each other showing the range of 0.03 - 0.07 error percent. It is analyzed that the groundwater flow distributed evenly from Nopen-orum and Munseogi-orum to Wolang-bong, Yongnuni-orum, and Songja-bong through the computation of equipotentials and velocity vector using the analysis result of simulation which was performed before the pumping started in the study area. These analysis results show the accordance with MODFLOW's.

• Journal of KIISE:Software and Applications
• /
• v.35 no.12
• /
• pp.780-788
• /
• 2008

Modeling hand poses and tracking its movement are one of the challenging problems in computer vision. There are two typical approaches for the reconstruction of hand poses in 3D, depending on the number of cameras from which images are captured. One is to capture images from multiple cameras or a stereo camera. The other is to capture images from a single camera. The former approach is relatively limited, because of the environmental constraints for setting up multiple cameras. In this paper we propose a method of reconstructing 3D hand poses from a 2D input image sequence captured from a single camera by means of Belief Propagation in a graphical model and recognizing a finger clicking motion using a hidden Markov model. We define a graphical model with hidden nodes representing joints of a hand, and observable nodes with the features extracted from a 2D input image sequence. To track hand poses in 3D, we use a Belief Propagation algorithm, which provides a robust and unified framework for inference in a graphical model. From the estimated 3D hand pose we extract the information for each finger's motion, which is then fed into a hidden Markov model. To recognize natural finger actions, we consider the movements of all the fingers to recognize a single finger's action. We applied the proposed method to a virtual keypad system and the result showed a high recognition rate of 94.66% with 300 test data.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.24 no.12A
• /
• pp.1922-1931
• /
• 1999

A multicast traffic’s feature is the function of providing a point to multipoints cell transmission, which is emerging from the main function of ATM switch. However, when a conventional point-to-point switch executes a multicast function, the excess load is occurred because unicast cell as well as multicast cell passed the copy network. Additionally, due to the excess load, multicast cells collide with other cells in a switch. Thus a deadlock that losses cells raises, extremely diminishes the performance of switch. An input queued switch also has a defect of the HOL (Head of Line) blocking that less lessens the performance of the switch. In the proposed multicast switch, we use shared memory switch to reduce HOL blocking and deadlock. In order to decrease switch’s complexity and cell's processing time, to improve a throughput, we utilize the method that routes a cell on a separated paths by traffic pattern and the scheduling algorithm that processes a maximum 2N cell at once in the control part. Besides, when cells is congested at an output port, a cell loss probability increases. Thus we use the Output Memory (OM) to reduce the cell loss probability. And we make use of the method that stores the assigned memory (UM, MM) with a cell by a traffic pattern and clears the cell of the Output memory after a fixed saving time to improve the memory utilization rate. The performance of the proposed switch is executed and compared with the conventional policy under the burst traffic condition through both the analysis based on Markov chain and simulation.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.35 no.6B
• /
• pp.930-939
• /
• 2010

Recently, many researches for making more comfortable input device based on gaze detection have been vigorously performed in human computer interaction. However, these previous researches are difficult to be used in IPTV environment because these methods need additional wearing devices or do not work at a distance. To overcome these problems, we propose a new way of controlling IPTV interface by using a detected face and eye positions in single static camera. And although face or eyes are not detected successfully by using Adaboost algorithm, we can control IPTV interface by using motion vectors calculated by pyramidal KLT (Kanade-Lucas-Tomasi) feature tracker. These are two novelties of our research compared to previous works. This research has following advantages. Different from previous research, the proposed method can be used at a distance about 2m. Since the proposed method does not require a user to wear additional equipments, there is no limitation of face movement and it has high convenience. Experimental results showed that the proposed method could be operated at real-time speed of 15 frames per second. Wd confirmed that the previous input device could be sufficiently replaced by the proposed method.

• Journal of Korea Society of Industrial Information Systems
• /
• v.23 no.6
• /
• pp.67-77
• /
• 2018

Currently, there are many problems due to the decline in school-age population. Moreover, Korea has the largest number of universities compared to the population, and the university enrollment rate is also the highest in the world. As a result, the minimum student retention rate required for the survival of each university is becoming increasingly important. The purpose of this study was to examine the effects of reducing the number of graduates of education and the social climate that prioritizes employment. And to determine what the basic direction is for students to manage the student retention rate, which can be maintained from admission to graduation, to determine the optimal input variables, Based on the input parameters, we will make associative analysis using apriori algorithm to collect training data that is most suitable for maintenance rate management and make base data for development of the most efficient Deep Learning module based on it. The accuracy of Deep Learning was 75%, which is a measure of graduation using decision trees. In decision tree, factors that determine whether to graduate are graduated from general high school and students who are female and high in residence in urban area have high probability of graduation. As a result, the Deep Learning module developed rather than the decision tree was identified as a model for evaluating the graduation of students more efficiently.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.33 no.6
• /
• pp.1099-1110
• /
• 2023

The recognition system of autonomous driving and robot navigation performs vision work such as object recognition, tracking, and lane detection after multi-sensor fusion to improve performance. Currently, research on a deep learning model based on the fusion of a camera and a lidar sensor is being actively conducted. However, deep learning models are vulnerable to adversarial attacks through modulation of input data. Attacks on the existing multi-sensor-based autonomous driving recognition system are focused on inducing obstacle detection by lowering the confidence score of the object recognition model.However, there is a limitation that an attack is possible only in the target model. In the case of attacks on the sensor fusion stage, errors in vision work after fusion can be cascaded, and this risk needs to be considered. In addition, an attack on LIDAR's point cloud data, which is difficult to judge visually, makes it difficult to determine whether it is an attack. In this study, image scaling-based camera-lidar We propose an attack method that reduces the accuracy of LCCNet, a fusion model (camera-LiDAR calibration model). The proposed method is to perform a scaling attack on the point of the input lidar. As a result of conducting an attack performance experiment by size with a scaling algorithm, an average of more than 77% of fusion errors were caused.

• 전자공학회논문지 IE
• /
• v.43 no.4
• /
• pp.21-27
• /
• 2006

This paper is for the segmentation and volume calculation of the white matter and gray matter from brain MRI. We segment white matter, gray matter and CSF from the Brain image in the normal and abnormal person, and calculate the volume of segmented tissue. In this paper, we present a new method of extracting white matter, gray matter and CSF and calculation its volume from MR images for brain. And we have developed the determining method of threshold that can extract white matter and gray matter from MR image for brain through the analysis of gray values represented by ratio of each component. We proposed the calculation method of volume for white matter and gray matter by using number of extracted pixels in each slice. This algorithm input CSF/Head volume ratio and age of patient and calculates discriminant value through discriminant expression, classifies normal and abnormal using calculated discriminant value. As a result, we could blow that white matter and gray matter volume decrease and CSF volume increase as we grow gold.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.28 no.12C
• /
• pp.1250-1257
• /
• 2003

Generally images are classified into two regions: edge and flat region. While low frequency components are popular in the flat region, high frequency components are quite important in the edge region. Therefore, deinterlacing algorithm that considers the characteristic of each region can be more efficient. In this paper, an image is divided into edge region and flat region by the local variance. And then, for each region, frequency modularized neural network is assigned. Using this structure, each modularized neural network can learn only its region intensively and avoid the complexity of learning caused by the data of different region. Using the local AC data for the input of neural network can prevent the degradation of the performance of teaming due to the average intensity values of image that disturbs the effective learning. The proposed method shows the improved performance compared with previous algorithms in the simulation.

• The Transactions of the Korea Information Processing Society
• /
• v.1 no.2
• /
• pp.215-224
• /
• 1994

In this paper, the rotation and size invariant fingerprint recognition using the neural network EART (Extended Adaptive Resonance Theory) is studied ($515{\times}512$) gray level fingerprint images are converted into the binary thinned images based on the adaptive threshold and a thinning algorithm. From these binary thinned images, we extract the ending points and the bifurcation points, which are the most useful critical feature points in the fingerprint images, using the $3{\times}3$ MASK. And we convert the number of these critical points and the interior angles of convex polygon composed of the bifurcation points into the 40*10 critical using the weighted code which is invariant of rotation and size as the input of EART. This system produces very good and efficient results for the rotation and size variations without the restoration of the binary thinned fingerprints.