• 정보보호학회논문지
• /
• 제31권6호
• /
• pp.1227-1236
• /
• 2021

딥러닝 분류 모델에 대한 공격 중 하나인 적대적 공격은 입력 데이터에 인간이 구별할 수 없는 섭동을 추가하여 딥러닝 분류 모델이 잘못 분류하도록 만드는 공격이며, 다양한 적대적 공격 알고리즘이 존재한다. 이에 따라 적대적 데이터를 탐지하는 연구는 많이 진행되었으나 적대적 데이터가 어떤 적대적 공격 알고리즘에 의해 생성되었는지 분류하는 연구는 매우 적게 진행되었다. 적대적 공격을 분류할 수 있다면, 공격 간의 차이를 분석하여 더욱 견고한 딥러닝 분류 모델을 구축할 수 있을 것이다. 본 논문에서는 공격 대상 딥러닝 모델이 예측하는 클래스를 기반으로 은닉층의 출력값에서 특징을 추출하고 추출된 특징을 입력으로 하는 랜덤 포레스트 분류 모델을 구축하여 적대적 공격을 탐지 및 분류하는 모델을 제안한다. 실험 결과 제안한 모델은 최신의 적대적 공격 탐지 및 분류 모델보다 정상 데이터의 경우 3.02%, 적대적 데이터의 경우 0.80% 높은 정확도를 보였으며, 기존 연구에서 분류하지 않았던 새로운 공격을 분류한다.

• 산업경영시스템학회지
• /
• 제37권1호
• /
• pp.1-8
• /
• 2014

This paper proposes an emotion classifier from EEG signals based on Bayes' theorem and a machine learning using a perceptron convergence algorithm. The emotions are represented on the valence and arousal dimensions. The fast Fourier transform spectrum analysis is used to extract features from the EEG signals. To verify the proposed method, we use an open database for emotion analysis using physiological signal (DEAP) and compare it with C-SVC which is one of the support vector machines. An emotion is defined as two-level class and three-level class in both valence and arousal dimensions. For the two-level class case, the accuracy of the valence and arousal estimation is 67% and 66%, respectively. For the three-level class case, the accuracy is 53% and 51%, respectively. Compared with the best case of the C-SVC, the proposed classifier gave 4% and 8% more accurate estimations of valence and arousal for the two-level class. In estimation of three-level class, the proposed method showed a similar performance to the best case of the C-SVC.

• 대한전기학회:학술대회논문집
• /
• 대한전기학회 2002년도 합동 추계학술대회 논문집 정보 및 제어부문
• /
• pp.521-524
• /
• 2002

This paper deals with the rule generation from data for control system and data mining using rough set. If the cores and reducts are searched for without consideration of the frequency of data belonging to the same equivalent class, the unnecessary attributes may not be discarded, and the resultant rules don't represent well the characteristics of the data. To improve this, we handle the inconsistent data with a probability measure defined by support, As a result the effect of uncertainty in knowledge reduction can be reduced to some extent. Also we construct the rule base in a hierarchical structure by applying core as the classification criteria at each level. If more than one core exist, the coverage degree is used to select an appropriate one among then to increase the classification rate. The proposed method gives more proper and effective rule base in compatibility and size. For some data mining example the simulations are performed to show the effectiveness of the proposed method.

• 대한수학회보
• /
• 제61권3호
• /
• pp.867-873
• /
• 2024

A pseudo-Riemannian solvmanifold is a solvable Lie group endowed with a left invariant pseudo-Riemannian metric. In this short note, we investigate the nilpotency of the Ricci operator of pseudo-Riemannian solvmanifolds. We focus on a special class of solvable Lie groups whose Lie algebras can be expressed as a one-dimensional extension of a nilpotent Lie algebra ℝD⋉n, where D is a derivation of n whose restriction to the center of n has at least one real eigenvalue. The main result asserts that every solvable Lie group belonging to this special class admits a left invariant pseudo-Riemannian metric with nilpotent Ricci operator. As an application, we obtain a complete classification of three-dimensional solvable Lie groups which admit a left invariant pseudo-Riemannian metric with nilpotent Ricci operator.

• 센서학회지
• /
• 제30권5호
• /
• pp.286-289
• /
• 2021

Image classification is one of the fundamental applications of computer vision. It enables a system to identify an object in an image. Recently, image classification applications have broadened their scope from computer applications to edge devices. The convolutional neural network (CNN) is the main class of deep learning neural networks that are widely used in computer tasks, and it delivers high accuracy. However, CNN algorithms use a large number of parameters and incur high computational costs, which hinder their implementation in edge hardware devices. To address this issue, this paper proposes a lightweight image classifier that provides good accuracy while using fewer parameters. The proposed image classifier diverts the input into three paths and utilizes different scales of receptive fields to extract more feature maps while using fewer parameters at the time of training. This results in the development of a model of small size. This model is tested on the CIFAR-10 dataset and achieves an accuracy of 90% using .26M parameters. This is better than the state-of-the-art models, and it can be implemented on edge devices.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제13권5호
• /
• pp.2558-2573
• /
• 2019

We present a deep CNN-based approach for classifying artistic media from artwork images. We aim to classify most frequently used artistic media including oilpaint brush, watercolor brush, pencil and pastel, etc. For this purpose, we extend VGGNet, one of the most widely used CNN structure, by substituting its last layer with a fully convolutional layer, which reveals class activation map (CAM), the region of classification. We build two artwork image datasets: YMSet that collects more than 4K artwork images for four most frequently used artistic media from various internet websites and WikiSet that collects almost 9K artwork images for ten most frequently used media from WikiArt. We execute a human baseline experiment to compare the classification performance. Through our experiments, we conclude that our classifier is superior in classifying artistic media to human.

• 스마트미디어저널
• /
• 제12권7호
• /
• pp.52-58
• /
• 2023

Alzheimer's disease is one type of dementia, the symptoms can be treated by detecting the disease at its early stages. Recently, many computer-aided diagnosis using magnetic resonance image(MRI) have shown a good results in the classification of AD. Taken these MRI images and feed to Free surfer software to extra the features. In consideration, using T1-weighted images and classifying using the convolution neural network (CNN) model are proposed. In this paper, taking the subjects from ADNI of subcortical and cortical features of 190 subjects. Consider the study to reduce the complexity of the model by using the single layer in the Res-Net, VGG, and Alex Net. Multi-class classification is used to classify four different stages, CN, EMCI, LMCI, AD. The following experiment shows for respective classification Res-Net, VGG, and Alex Net with the best accuracy with VGG at 96%, Res-Net, GoogLeNet and Alex Net at 91%, 93% and 89% respectively.

• 한국방송∙미디어공학회:학술대회논문집
• /
• 한국방송∙미디어공학회 2020년도 추계학술대회
• /
• pp.244-247
• /
• 2020

Data augmentation has been helpful in improving the performance in deep learning, when we have a limited data and random erasing is one of the augmentations that have shown impressive performance in deep learning in multiple domains. But the main issue is that sometime it loses good features when randomly selected region is erased by some random values, that does not improve performance as it should. We target that problem in way that good features should not be lost and also want random erasing at the same time. For that purpose, we introduce new augmentation technique named Intra-Class Random Erasing (ICRE) that focuses on data to learn robust features of the same class samples by randomly exchanging randomly selected region. We perform multiple experiments by using different models including resnet18, VGG16 over variety of the datasets including ESC10, UrbanSound8K. Our approach has shown effectiveness over others methods including random erasing.

• 치과방사선
• /
• 제22권1호
• /
• pp.149-160
• /
• 1992

The purpose of this study was to localize the inferior alveolar canals in relation to the root apices of the mandibular third molars, according to their positions and degrees of impaction using vertical tube shift technique. One hundred three mandibular third molars, from 95 persons consisted of 57 males and 38 females, were clinically and radiographically investigated. The mandibular third molars had no pericoronitis and periapical lesions, and showed an evidence of complete root formation. The obtained results were as follows: 1. In localiztion of the inferior alveolar canals in relation to the root apices of the mandibular third molars, the inferior alveolar canal was located at the buccal side of the root apices of mandibular third molar in 77.7%, below the root apices in 16.5%, and the lingual side of root apices in 5.8%. 2. The positions of the mandibular third molars according to the Winter's Classification were as follows; 36.9% in Class Ⅰ, 21.3% in Class Ⅱ, 14.7% in Class Ⅲ, 4.8% in Class Ⅳ, 1.9% in Class Ⅴ, 17.5% in Class Ⅵ, 2.9% in Class Ⅶ. In localization of the inferior alveolar canals in relation to the root apices of the mandibular third molars according to the Winter's Classification, 92.1 % of Class Ⅰ, 86.4% of Class Ⅱ, 80.0% of Class Ⅲ, and 100.0% of Class Ⅳ and Ⅴ were located at the buccal side. In Class Ⅵ, however, 33.3% was located at the buccal side, 44.5% below the root apices, and 22.2% at the lingual side. 3. The degree of impaction was revealed to be 53.4% in Degree Ⅰ, 36.9% in Degree Ⅱ, and 9.7% in Degree Ⅲ. In localization of the inferior alveolar canals in relation to the root apices of mandibular third molars according to degree of impaction, 98.2% of Degree Ⅰ was located at the buccal side. In Degree Ⅱ, 60.5% was located at the bucal side, 31.6% below the root apices, and 7.9% at the lingual side. In Degree Ⅲ, 30.0% was located at the buccal side, 40.0% below the root apices, and 30.0% at the lingual side.

• 정보관리학회지
• /
• 제35권2호
• /
• pp.37-62
• /
• 2018

문헌정보학 분야의 국내 학술지 논문으로 구성된 문헌집합을 대상으로 기계학습에 기초한 자동분류의 성능에 영향을 미치는 요소들을 검토하였다. 특히, "정보관리학회지"에 수록된 논문에 주제 범주를 자동 할당하는 분류 성능 측면에서 용어 가중치부여 기법, 학습집합 크기, 분류 알고리즘, 범주 할당 방법 등 주요 요소들의 특성을 다각적인 실험을 통해 살펴보았다. 결과적으로 분류 환경 및 문헌집합의 특성에 따라 각 요소를 적절하게 적용하는 것이 효과적이며, 보다 단순한 모델의 사용으로 상당히 좋은 수준의 성능을 도출할 수 있었다. 또한, 국내 학술지 논문의 분류는 특정 논문에 하나 이상의 범주를 할당하는 복수-범주 분류(multi-label classification)가 실제 환경에 부합한다고 할 수 있다. 따라서 이러한 환경을 고려하여 단순하고 빠른 분류 알고리즘과 소규모의 학습집합을 사용하는 최적의 분류 모델을 제안하였다.