• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.12 no.10
• /
• pp.4930-4951
• /
• 2018

The deep learning based speech enhancement has shown considerable success. However, it still suffers performance degradation under mismatch conditions. In this paper, an adaptation method is proposed to improve the performance under noise mismatch conditions. Firstly, we advise a noise aware training by supplying identity vectors (i-vectors) as parallel input features to adapt deep neural network (DNN) acoustic models with the target noise. Secondly, given a small amount of adaptation data, the noise-dependent DNN is obtained by using $L_2$ regularization from a noise-independent DNN, and forcing the estimated masks to be close to the unadapted condition. Finally, experiments were carried out on different noise and SNR conditions, and the proposed method has achieved significantly 0.1%-9.6% benefits of STOI, and provided consistent improvement in PESQ and segSNR against the baseline systems.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.14 no.7
• /
• pp.2752-2768
• /
• 2020

Phishing websites can have devastating effects on governmental, financial, and social services, as well as on individual privacy. Currently, many phishing detection solutions are evaluated using small datasets and, thus, are prone to sampling issues, such as representing legitimate websites by only high-ranking websites, which could make their evaluation less relevant in practice. Phishing detection solutions which depend only on the URL are attractive, as they can be used in limited systems, such as with firewalls. In this paper, we present a URL-only phishing detection solution based on a convolutional neural network (CNN) model. The proposed CNN takes the URL as the input, rather than using predetermined features such as URL length. For training and evaluation, we have collected over two million URLs in a massive URL phishing detection (MUPD) dataset. We split MUPD into training, validation and testing datasets. The proposed CNN achieves approximately 96% accuracy on the testing dataset; this accuracy is achieved with URL schemes (such as HTTP and HTTPS) removed from the URL. Our proposed solution achieved better accuracy compared to an existing state-of-the-art URL-only model on a published dataset. Finally, the results of our experiment suggest keeping the CNN up-to-date for better results in practice.

• Korean Journal of Head & Neck Oncology
• /
• v.19 no.2
• /
• pp.137-141
• /
• 2003

Introduction: A routine superficial parotidectomy with facial nerve dissection in parotid tumor surgery often results in facial dysfunction, Frey syndrome and defect in operation site. Formal facial nerve dissection has been a recommended procedure, because pleomorphic adenoma is a commonly recurrent tumor in case of inadequate surgical management, however it can not be always reasonable in aspect of postoperative sequelae. Patients and Methods: Through retrospective review of 245 cases parotidectomies and follow up for more than three years, clinicophathologic factors influencing to the selection of surgical procedure were considered to be age, sex, and preoperative pathology confirmed by preoperative MRI and FNA. Results: Five categories were established as follow for surgical decision in parotid tumor surgery. Category 1. Superficial lobe adenoma -- Superficial parotidectomy -- 124 Category 2. Deep lobe adenoma -- Deep parotidectomy -- 39 Category 3. Non pleomorphic adenoma -- Tumorectomy 1.5cm adenoma in young female -- Tumorectomy -- 25 Category 4. Recurrent multicentric tumor -- Parotidectomy+RT -- 9 Category 5. Parotid cancer; Parotidectomy + UND (RND) + RT -- 48 ; CORE (Composite Regional Ear Resection) -- 2 Conclusion: Surgical morbidity and recurrence rate could be minimized by individualizing the surgical procedure according to the category principle based on the clincopathologic features.

• Steel and Composite Structures
• /
• v.37 no.2
• /
• pp.193-209
• /
• 2020

In this paper, the efficiency of five Machine Learning (ML) methods consisting of Deep Learning (DL), Support Vector Machine (SVM), Random Forest (RF), Decision Tree (DT), and Gradient Tree Booting (GTB) for regression and classification of the Ultimate Load Factor (ULF) of nonlinear inelastic steel frames is compared. For this purpose, a two-story, a six-story, and a twenty-story space frame are considered. An advanced nonlinear inelastic analysis is carried out for the steel frames to generate datasets for the training of the considered ML methods. In each dataset, the input variables are the geometric features of W-sections and the output variable is the ULF of the frame. The comparison between the five ML methods is made in terms of the mean-squared-error (MSE) for the regression models and the accuracy for the classification models, respectively. Moreover, the ULF distribution curve is calculated for each frame and the strength failure probability is estimated. It is found that the GTB method has the best efficiency in both regression and classification of ULF regardless of the number of training samples and the space frames considered.

• Journal of IKEEE
• /
• v.20 no.3
• /
• pp.303-306
• /
• 2016

In this paper, we propose a method to accelerate convolutional neural network by utilizing a GPGPU. Convolutional neural network is a sort of the neural network learning features of images. Convolutional neural network is suitable for the image processing required to learn a lot of data such as images. The convolutional layer of the conventional CNN required a large number of multiplications and it is difficult to operate in the real-time on the embedded environment. In this paper, we reduce the number of multiplications through Winograd convolution operation and perform parallel processing of the convolution by utilizing SIMT-based GPGPU. The experiment was conducted using ModelSim and TestDrive, and the experimental results showed that the processing time was improved by about 17%, compared to the conventional convolution.

• International Journal of Advanced Culture Technology
• /
• v.7 no.4
• /
• pp.321-326
• /
• 2019

As AI technology is recently introduced into various fields, it is being applied to the fashion field. This paper proposes a system for recommending cody clothes suitable for a user's selected clothes. The proposed system consists of user app, cody recommendation module, and server interworking of each module and managing database data. Cody recommendation system classifies clothing images into 80 categories composed of feature combinations, selects multiple representative reference images for each category, and selects 3 full body cordy images for each representative reference image. Cody images of the representative reference image were determined by analyzing the user's preference using Google survey app. The proposed algorithm classifies categories the clothing image selected by the user into a category, recognizes the most similar image among the classification category reference images, and transmits the linked cody images to the user's app. The proposed system uses the ResNet-50 model to categorize the input image and measures similarity using ORB and HOG features to select a reference image in the category. We test the proposed algorithm in the Android app, and the result shows that the recommended system runs well.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.38 no.5
• /
• pp.443-453
• /
• 2020

The automatic land category extraction method was proposed, and the accuracy was evaluated by learning the aerial photo characteristics by land category in the border area with various restrictions on the acquisition of geospatial data. As experimental data, this study used four years' worth of published aerial photos as well as serial cadastral maps from the same time period. In evaluating the results of land category extraction by learning features from different temporal and spatial ranges of aerial photos, it was found that land category extraction accuracy improved as the temporal and spatial ranges increased. Moreover, the greater the diversity and quantity of provided learning images, the less the results were affected by the quality of images at a specific time to be extracted, thus generally demonstrating accurate and practical land category feature extraction.

• Journal of Digital Convergence
• /
• v.19 no.12
• /
• pp.353-358
• /
• 2021

Parkinson's disease is the second most common degenerative brain disease after Alzheimer's in old age. Symptoms of Parkinson's disease are factors that reduce the quality of life in daily life, such as shaking hands, slowing behavior and cognitive function. Parkinson's disease that can slow the progression of the disease through early diagnosis. To diagnoze Parkinson's disease early, an algorithm was implemented to extract features using wav2vec and to diagnose the presence or absence of Parkinson's disease with deep learning(ANN). As a results of the experiment, the accuracy was 97.47%. It was better than the results of diagnosing Parkinson's disease using the existing neural network. The audio voice file could simply reduce the experiment process and obtain improved results.

• Journal of the Korea Society of Computer and Information
• /
• v.26 no.10
• /
• pp.19-26
• /
• 2021

The performance of deep learning-based malware detection and classification models depends largely on how to construct a feature set to be applied to training. In this paper, we propose an approach to select the optimal feature set to maximize detection performance for CNN-based Android malware detection. The features to be included in the feature set were selected through the Chi-Square test algorithm, which is widely used for feature selection in machine learning and deep learning. To validate the proposed approach, the CNN model was trained using 36 characteristics selected for the CICANDMAL2017 dataset and then the malware detection performance was measured. As a result, 99.99% of Accuracy was achieved in binary classification and 98.55% in multiclass classification.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.15 no.2
• /
• pp.658-672
• /
• 2021

Since the signal with strong power need be demodulated first for successive interference cancellation (SIC) receiver in non-orthogonal multiple access (NOMA) systems, the base station (BS) need inform the near user terminal (UT), which has allocated higher power, of the far UT's modulation mode. To avoid unnecessary signaling overhead of control channel, a blind detection algorithm of NOMA signal modulation mode is designed in this paper. Taking the joint constellation density diagrams of NOMA signal as the detection features, the deep residual network is built for classification, so as to detect the modulation mode of NOMA signal. In view of the fact that the joint constellation diagrams are easily polluted by high intensity noise and lose their real distribution pattern, the wavelet denoising method is adopted to improve the quality of constellations. The simulation results represent that the proposed algorithm can achieve satisfactory detection accuracy in NOMA systems. In addition, the factors affecting the recognition performance are also verified and analyzed.