• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.21 no.2
• /
• pp.189-194
• /
• 2020

This paper analyzes the gradient descent method, which is the one most used for learning neural networks. Learning means updating a parameter so the loss function is at its minimum. The loss function quantifies the difference between actual and predicted values. The gradient descent method uses the slope of the loss function to update the parameter to minimize error, and is currently used in libraries that provide the best deep learning algorithms. However, these algorithms are provided in the form of a black box, making it difficult to identify the advantages and disadvantages of various gradient descent methods. This paper analyzes the characteristics of the stochastic gradient descent method, the momentum method, the AdaGrad method, and the Adadelta method, which are currently used gradient descent methods. The experimental data used a modified National Institute of Standards and Technology (MNIST) data set that is widely used to verify neural networks. The hidden layer consists of two layers: the first with 500 neurons, and the second with 300. The activation function of the output layer is the softmax function, and the rectified linear unit function is used for the remaining input and hidden layers. The loss function uses cross-entropy error.

• Journal of Digital Contents Society
• /
• v.18 no.7
• /
• pp.1443-1450
• /
• 2017

The purpose of this study is to develop an intelligent parking dispatching system based on LoRa network technology. During the local festival, many tourists come into the festival site simultaneously after sunset. To handle the traffic jam and parking dispatching, many traffic management staffs are engaged in the main road to guide the cars to available parking lots. Nevertheless, the traffic problems are more serious at the peak time of festival. Such parking dispatching problems are complex and real-time traffic information dependent. We used Queuing theory to predict inbound traffics and to measure parking service performance. Q-learning algorithm is used to find fastest routes and dispatch the vehicles efficiently to the available parking lots.

• KIPS Transactions on Software and Data Engineering
• /
• v.6 no.12
• /
• pp.573-580
• /
• 2017

Deep Neural Network (DNN) is a large layered neural network which is consisted of a number of layers of non-linear units. Deep Learning which represented as DNN has been applied very successfully in various applications. However, many issues in DNN have been identified through past researches. Among these issues, generalization is the most well-known problem. A Recent study, Dropout, successfully addressed this problem. Also, Dropout plays a role as noise, and so it helps to learn robust feature during learning in DNN such as Denoising AutoEncoder. However, because of a large computations required in Dropout, training takes a lot of time. Since Dropout keeps changing an inter-layer representation during the training session, the learning rates should be small, which makes training time longer. In this paper, using mutation operation, we reduce computation and improve generalization performance compared with Dropout. Also, we experimented proposed method to compare with Dropout method and showed that our method is superior to the Dropout one.

• Journal of IKEEE
• /
• v.24 no.1
• /
• pp.147-153
• /
• 2020

Conventional lane detection algorithms have problems in that the detection rate is lowered in road environments having a large change in curvature and illumination. The probabilistic Hough transform method has low lane detection rate since it exploits edges and restrictive angles. On the other hand, the method using a sliding window can detect a curved lane as the lane is detected by dividing the image into windows. However, the detection rate of this method is affected by road slopes because it uses affine transformation. In order to detect lanes robustly and avoid obstacles, we propose driving assist system using semantic segmentation based on deep learning. The architecture for segmentation is SegNet based on VGG-16. The semantic image segmentation feature can be used to calculate safety space and predict collisions so that we control a vehicle using adaptive-MPC to avoid objects and keep lanes. Simulation results with CARLA show that the proposed algorithm detects lanes robustly and avoids unknown obstacles in front of vehicle.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2018.10a
• /
• pp.58-61
• /
• 2018

In Korea, current medical technology is the highest level in the world. As a result, many doctors have specialized knowledge of various disorders or diseases, and are proceeding in a better way. With such high medical technology, it is possible to increase the probability of success of surgery to provide high reliability to patients. Rehabilitation is also a form of medical treatment that reduces the side effects that occur after surgery that is done for quick cure. However, the situation in this section is slightly different. There are moves to develop rehabilitation devices and operations, but most of them are now dependent on foreign technology. Rehabilitation therapy, which we commonly know, is dominated by behavior. However, it is also a kind of rehabilitation to find out how much the patient's symptoms are improved or recovered. In this paper, we have studied the development of a program by using the Deeplearning method in order to detect the problem of the food swallowing operation by the severity.

• Journal of the Korea Convergence Society
• /
• v.9 no.12
• /
• pp.47-52
• /
• 2018

Named Entity Recognition is a system that extracts entity names such as Persons(PS), Locations(LC), and Organizations(OG) that can have a unique meaning from a document and determines the categories of extracted entity names. Recently, Bi-LSTM-CRF, which is a combination of CRF using the transition probability between output data from LSTM-based Bi-LSTM model considering forward and backward directions of input data, showed excellent performance in the study of object name recognition using deep-learning, and it has a good performance on the efficient embedding vector creation by character and word unit and the model using CNN and LSTM. In this research, we describe the Bi-LSTM-CNN-CRF model that enhances the features of the Korean named entity recognition system and propose a method for constructing the traditional culture corpus. We also present the results of learning the constructed corpus with the feature augmentation model for the recognition of Korean object names.

• The Journal of the Korea Contents Association
• /
• v.21 no.2
• /
• pp.587-593
• /
• 2021

In this thesis, we will break away from the classic DDoS response system for large-scale denial-of-service attacks that develop day by day, and effectively endure intelligent denial-of-service attacks by utilizing artificial intelligence-based technology, one of the core technologies of the 4th revolution. A possible service model development plan was proposed. That is, a method to detect denial of service attacks and minimize damage through machine learning artificial intelligence learning targeting a large amount of data collected from multiple security devices and web servers was proposed. In particular, the development of a model for using artificial intelligence technology is to detect a Western service attack by focusing on the fact that when a service denial attack occurs while repeating a certain traffic change and transmitting data in a stable flow, a different pattern of data flow is shown. Artificial intelligence technology was used. When a denial of service attack occurs, a deviation between the probability-based actual traffic and the predicted value occurs, so it is possible to respond by judging as aggressiveness data. In this paper, a service denial attack detection model was explained by analyzing data based on logs generated from security equipment or servers.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.24 no.4
• /
• pp.501-507
• /
• 2020

This paper proposes a new frame synchronization technique based on convolutional neural network (CNN). The conventional frame synchronizers usually find the matching instance through correlation between the received signal and the preamble. The proposed method converts the 1-dimensional correlator ouput into a 2-dimensional matrix. The 2-dimensional matrix is input to a convolutional neural network, and the convolutional neural network finds the frame arrival time. Specifically, in additive white gaussian noise (AWGN) environments, the received signals are generated with random arrival times and they are used for training data of the CNN. Through computer simulation, the false detection probabilities in various signal-to-noise ratios are investigated and compared between the proposed CNN-based technique and the conventional one. According to the results, the proposed technique shows 2dB better performance than the conventional method.

• Journal of Advanced Navigation Technology
• /
• v.26 no.3
• /
• pp.179-184
• /
• 2022

The accurate wing flexure model is required to improve the transfer alignment performance of guided weapon system mounted on a wing of fighter aircraft or armed helicopter. In order to solve this problem, mechanical or stochastical modeling methods have been studying, but modeling accuracy is too low to be applied to weapon systems. The deep learning techniques that have been studying recently are suitable for nonlinear. However, operating fighter aircraft for deep-learning modeling to secure a large amount of data is practically difficult. In this paper, it was used to generate amount of flexure data samples that are similar to the actual flexure data. And it was confirmed that generated data is similar to the actual data by utilizing "measures of similarity" which measures how much alike the two data objects are.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.39 no.6
• /
• pp.381-392
• /
• 2021

In this study, we perform deep learning-based object detection analysis on eight types of buildings defined by the digital map topography standard code, leveraging images taken with UAV (Unmanned Aerial Vehicle). Image labeling was done for 509 images taken by UAVs and the YOLO (You Only Look Once) v5 model was applied to proceed with learning and inference. For experiments and analysis, data were analyzed by applying an open source-based analysis platform and algorithm, and as a result of the analysis, building objects were detected with a prediction probability of 88% to 98%. In addition, the learning method and model construction method necessary for the high accuracy of building object detection in the process of constructing and repetitive learning of training data were analyzed, and a method of applying the learned model to other images was sought. Through this study, a model in which high-efficiency deep neural networks and spatial information data are fused will be proposed, and the fusion of spatial information data and deep learning technology will provide a lot of help in improving the efficiency, analysis and prediction of spatial information data construction in the future.