• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• fall
• /
• pp.80-81
• /
• 2021

딥 러닝 모델 사용에 있어서, 일반적인 사용자가 이용할 수 있는 하드웨어 리소스는 제한적이기 때문에 기존 모델을 경량화 할 수 있는 프루닝 방법을 통해 제한적인 리소스를 효과적으로 활용할 수 있도록 한다. 그 방법으로, 여러 딥 러닝 모델들 중 비교적 파라미터 수가 많은 것으로 알려진 GAN 아키텍처에 네트워크 프루닝을 적용함으로써 비교적 무거운 모델을 적은 파라미터를 통해 학습할 수 있는 방법을 제시한다. 또한, 본 논문을 통해 기존의 SRGAN 논문에서 가장 효과적인 결과로 제시했던 16 개의 residual block 의 개수를 실제로 줄여 봄으로써 기존 논문에서 제시했던 결과와의 차이에 대해 서술한다.

• The Journal of the Acoustical Society of Korea
• /
• v.39 no.4
• /
• pp.292-302
• /
• 2020

Machine learning algorithms have made immense contributions in various fields including sonar and radar applications. Recently developed Cycle-Consistency Generative Adversarial Network (CycleGAN), a variant of GAN has been successfully used for unpaired image-to-image translation. We present a modified CycleGAN for translation of underwater ship engine sounds with high perceptual quality. The proposed network is composed of an improved generator model trained to translate underwater audio from one vessel type to other, an improved discriminator to identify the data as real or fake and a modified cycle-consistency loss function. The quantitative and qualitative analysis of the proposed CycleGAN are performed on publicly available underwater dataset ShipsEar by evaluating and comparing Mel-cepstral distortion, pitch contour matching, nearest neighbor comparison and mean opinion score with existing algorithms. The analysis results of the proposed network demonstrate the effectiveness of the proposed network.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2021.07a
• /
• pp.145-146
• /
• 2021

본 연구에서는 카메라의 촬영 시점에 의해서 발생되는 원근감이 광학 흐름 생성에 어떠한 영향을 주는지 살펴보고 광학 흐름 기반 이상행동 탐지 솔루션의 성능을 고도화하기 위해 기존 광학 흐름 영상으로부터 소실점 기반 가중치 행렬을 계산하여 원근감에 따른 광학 흐름 정도를 평활하는 기법에 대해서 연구한다. 카메라의 뷰포인트에 따라 원근감의 발생 정도나 객체의 크기 및 움직임의 정도가 달라지게 되며, 이는 원본 영상 프레임을 광학 흐름의 크기와 방향성으로 표현하는 영상 변환 네트워크를 가진 생성적 적대 신경망을 학습할 때 정상적인 행동 패턴의 범위를 결정짓는 데 방해가 될 수 있다. 이러한 문제를 해결하기 위하여 데이터셋의 배경으로부터 소실점을 추출하고 원근감에 따라 결정되는 광학 흐름의 크기를 평활하는 기법을 개발하여 기존 모델의 성능과 비교하였으며, 프레임 단위의 정확도 성능이 5.75% 향상된 것으로 확인되었다.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.32 no.4
• /
• pp.647-660
• /
• 2022

Recent works demonstrate that the semi-supervised anomaly detection method functions quite well in the environment with normal data and some anomalous data. However, abnormal data shortages can occur in an environment where it is difficult to reserve anomalous data, such as an unknown attack in the cyber security fields. In this paper, we propose ADA-PH(Abnormal Data Augmentation Method using Perturbation based on Hypersphere), a novel anomalous data augmentation method that is applicable in an environment where abnormal data is insufficient to secure the performance of the semi-supervised anomaly detection method. ADA-PH generates abnormal data by perturbing samples located relatively far from the center of the hypersphere. With the network intrusion detection datasets where abnormal data is rare, ADA-PH shows 23.63% higher AUC performance than anomaly detection without data augmentation and even performs better than the other augmentation methods. Also, we further conduct quantitative and qualitative analysis on whether generated abnormal data is anomalous.

• Convergence Security Journal
• /
• v.23 no.4
• /
• pp.61-70
• /
• 2023

Clustering is an unsupervised learning method that involves grouping data based on features such as distance metrics, using data without known labels or ground truth values. This method has the advantage of being applicable to various types of data, including images, text, and audio, without the need for labeling. Traditional clustering techniques involve applying dimensionality reduction methods or extracting specific features to perform clustering. However, with the advancement of deep learning models, research on deep clustering techniques using techniques such as autoencoders and generative adversarial networks, which represent input data as latent vectors, has emerged. In this study, we propose a deep clustering technique based on deep learning. In this approach, we use an autoencoder to transform the input data into latent vectors, and then construct a vector space according to the cluster structure and perform k-means clustering. We conducted experiments using the MNIST and Fashion-MNIST datasets in the PyTorch machine learning library as the experimental environment. The model used is a convolutional neural network-based autoencoder model. The experimental results show an accuracy of 89.42% for MNIST and 56.64% for Fashion-MNIST when k is set to 10.

• Journal of the Korea Convergence Society
• /
• v.12 no.1
• /
• pp.49-55
• /
• 2021

Recently, with the development of database, it is possible to store a lot of data generated in finance, security, and networks. These data are being analyzed through classifiers based on machine learning. The main problem at this time is data imbalance. When we train imbalanced data, it may happen that classification accuracy is degraded due to over-fitting with majority class data. To overcome the problem of data imbalance, oversampling strategy that increases the quantity of data of minority class data is widely used. It requires to tuning process about suitable method and parameters for data distribution. To improve the process, In this study, we propose a strategy to explore and optimize oversampling combinations and ratio based on various methods such as synthetic minority oversampling technique and generative adversarial networks through genetic algorithms. After sampling credit card fraud detection which is a representative case of data imbalance, with the proposed strategy and single oversampling strategies, we compare the performance of trained classifiers with each data. As a result, a strategy that is optimized by exploring for ratio of each method with genetic algorithms was superior to previous strategies.

• Journal of the Korean Society of Radiology
• /
• v.81 no.6
• /
• pp.1290-1304
• /
• 2020

Medical image analyses have been widely used to differentiate normal and abnormal cases, detect lesions, segment organs, etc. Recently, owing to many breakthroughs in artificial intelligence techniques, medical image analyses based on deep learning have been actively studied. However, sufficient medical data are difficult to obtain, and data imbalance between classes hinder the improvement of deep learning performance. To resolve these issues, various studies have been performed, and data augmentation has been found to be a solution. In this review, we introduce data augmentation techniques, including image processing, such as rotation, shift, and intensity variation methods, generative adversarial network-based method, and image property mixing methods. Subsequently, we examine various deep learning studies based on data augmentation techniques. Finally, we discuss the necessity and future directions of data augmentation.

• Journal of the Korea Computer Graphics Society
• /
• v.24 no.2
• /
• pp.1-9
• /
• 2018

In this paper, we propose a method to restore corrupted black and white facial images to color. Previous studies have shown that when coloring damaged black and white photographs, such as old ID photographs, the area around the damaged area is often incorrectly colored. To solve this problem, this paper proposes a method of restoring the damaged area of input photo first and then performing colorization based on the result. The proposed method consists of two steps: BEGAN (Boundary Equivalent Generative Adversarial Networks) model based restoration and CNN (Convolutional Neural Network) based coloring. Our method uses the BEGAN model, which enables a clearer and higher resolution image restoration than the existing methods using the DCGAN (Deep Convolutional Generative Adversarial Networks) model for image restoration, and performs colorization based on the restored black and white image. Finally, we confirmed that the experimental results of various types of facial images and masks can show realistic color restoration results in many cases compared with the previous studies.