• Convergence Security Journal
• /
• v.20 no.2
• /
• pp.111-121
• /
• 2020

Deep neural networks (DNNs) provide good performance for machine learning tasks such as image recognition and object recognition. However, DNNs are vulnerable to an adversarial example. An adversarial example is an attack sample that causes the neural network to recognize it incorrectly by adding minimal noise to the original sample. However, the disadvantage is that it takes a long time to generate such an adversarial example. Therefore, in some cases, an attack may be necessary that quickly causes the neural network to recognize it incorrectly. In this paper, we propose a fast misclassification sample that can rapidly attack neural networks. The proposed method does not consider the distortion of the original sample when adding noise. We used MNIST and CIFAR10 as experimental data and Tensorflow as a machine learning library. Experimental results show that the fast misclassification sample generated by the proposed method can be generated with 50% and 80% reduced number of iterations for MNIST and CIFAR10, respectively, compared to the conventional Carlini method, and has 100% attack rate.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2022.07a
• /
• pp.11-14
• /
• 2022

본 논문에서는 회귀 문제에서 예측값들의 분산을 줄이기 위한 딥뉴럴 네트워크 구조를 제안한다. 일반적인 회귀 문제에서 딥뉴럴 네트워크 학습 시, 하나의 입력에 대한 레이블 값을 이용하여 학습한다. 본 눈문에서는 하나의 입력에 대한 레이블 값뿐만 아니라 두 입력에 대한 레이블 값들의 차이를 학습시키는 딥뉴럴 네트워크 구조를 제안한다. 통계학 이론을 통하여 예측값들의 분산이 줄어든다는 것을 증명한다. 또한, 배관 곡관의 감육두께를 예측하는 문제를 통해 제안된 네트워크의 성능을 검증한다. 일반적인 딥뉴럴 네트워크 구조를 이용하였을 때에 비하여 제안한 네트워크 구조를 이용하였을 때, 회귀 문제의 예측값들의 분산이 감소함을 확인한다.

• The Journal of Korean Institute of Electromagnetic Engineering and Science
• /
• v.29 no.7
• /
• pp.550-559
• /
• 2018

The effectiveness of deep neural networks (DNNs) for detection and classification of micro-Doppler signals generated by human walking and background noise sources is investigated. Previous research included a complex process for extracting meaningful features that directly affect classifier performance, and this feature extraction is based on experiences and statistical analysis. However, because a DNN gradually reconstructs and generates features through a process of passing layers in a network, the preprocess for feature extraction is not required. Therefore, binary classifiers and multiclass classifiers were designed and analyzed in which multilayer perceptrons (MLPs) and DNNs were applied, and the effectiveness of DNNs for recognizing micro-Doppler signals was demonstrated. Experimental results showed that, in the case of MLPs, the classification accuracies of the binary classifier and the multiclass classifier were 90.3% and 86.1%, respectively, for the test dataset. In the case of DNNs, the classification accuracies of the binary classifier and the multiclass classifier were 97.3% and 96.1%, respectively, for the test dataset.

• Convergence Security Journal
• /
• v.21 no.2
• /
• pp.57-66
• /
• 2021

Deep neural networks (DNNs) provide excellent performance for image, speech, and pattern recognition. However, DNNs sometimes misrecognize certain adversarial examples. An adversarial example is a sample that adds optimized noise to the original data, which makes the DNN erroneously misclassified, although there is nothing wrong with the human eye. Therefore studies on defense against adversarial example attacks are required. In this paper, we have experimentally analyzed the success rate of detection for adversarial examples by adjusting various parameters. The performance of the ensemble defense method was analyzed using fast gradient sign method, DeepFool method, Carlini & Wanger method, which are adversarial example attack methods. Moreover, we used MNIST as experimental data and Tensorflow as a machine learning library. As an experimental method, we carried out performance analysis based on three adversarial example attack methods, threshold, number of models, and random noise. As a result, when there were 7 models and a threshold of 1, the detection rate for adversarial example is 98.3%, and the accuracy of 99.2% of the original sample is maintained.

• Convergence Security Journal
• /
• v.21 no.1
• /
• pp.191-200
• /
• 2021

Artificial intelligence technology is developing in an environment where a lot of data is produced due to the development of computing technology, a cloud environment that can store data, and the spread of personal mobile phones. Among these artificial intelligence technologies, the deep neural network provides excellent performance in image recognition and image classification. There have been many studies on image detection for forest fires and fire prevention using such a deep neural network, but studies on detection of cigarette smoking were insufficient. Meanwhile, military units are establishing surveillance systems for various facilities through CCTV, and it is necessary to detect smoking near ammunition stores or non-smoking areas to prevent fires and explosions. In this paper, by reflecting experimentally optimized numerical values such as activation function and learning rate, we did the detection of smoking pictures and non-smoking pictures in two cases. As experimental data, data was constructed by crawling using pictures of smoking and non-smoking published on the Internet, and a machine learning library was used. As a result of the experiment, when the learning rate is 0.004 and the optimization algorithm Adam is used, it can be seen that the accuracy of 93% and F1-score of 94% are obtained.

• The Journal of Society for e-Business Studies
• /
• v.25 no.4
• /
• pp.61-75
• /
• 2020

The scope of data for understanding or predicting stock prices has been continuously widened from traditional structured format data to unstructured data. This study investigates whether commentary data collected from SNS may affect future stock prices. From "Stock Discussion Room" in Naver, we collect 20 stocks' commentary data for six months, and test whether this data have prediction power with respect to one-hour ahead price direction and price range. Deep neural network such as LSTM and CNN methods are employed to model the predictive relationship. Among the 20 stocks, we find that future price direction can be predicted with higher than the accuracy of 50% in 13 stocks. Also, the future price range can be predicted with higher than the accuracy of 50% in 16 stocks. This study validate that the investors' sentiment reflected in SNS community such as Naver's "Stock Discussion Room" may affect the demand and supply of stocks, thus driving the stock prices.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2022.11a
• /
• pp.615-617
• /
• 2022

본 논문에서는 얼굴사진 기반 감정인식 심층망, 음성사운드를 기반한 감정인식 심층망을 결합한 앙상블 네트워크 구축을 위한 사전연구로서 얼굴사진 기반 감정을 인식하는 기존 딥뉴럴 네트워크 모델들을 입력 데이터 처리 방법에 따라 분류하고, 각 방법의 특성을 분석한다. 또한, 얼굴사진 외관 특성을 기반한 감정인식 네트워크를 여러 구조로 구성하고, 구성된 방법의 성능을 비교하여, 우수 성능을 보이는 네트워크를 선정하여 추후 앙상블 네트워크의 구성 네트워크로 사용하고자 한다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2020.11a
• /
• pp.8-11
• /
• 2020

4 차 산업혁명 시대의 도래와 함께 AI, ICT 기술의 융합이 진행됨에 따라, 유저 레벨의 디바이스에서도 AI 서비스의 요청이 실현되었다. 이미지 처리와 관련된 AI 서비스는 피사체 판별, 불량품 검사, 자율주행 등에 이용되고 있으며, 특히 Deep Convolutional Neural Network (DCNN)은 이미지의 특색을 파악하는 데 뛰어난 성능을 보여준다. 하지만, 이미지의 크기가 커지고, 신경망이 깊어짐에 따라 연산 처리에 있어 낮은 데이터 지역성과 빈번한 메모리 참조를 야기했다. 이에 따라, 기존의 계층적 시스템 구조는 DCNN 을 scalable 하고 빠르게 처리하는 데 한계를 보인다. 본 연구에서는 DCNN 의 scalable 하고 빠른 처리를 위해 3 차원 메모리 구조의 Processing-In-Memory (PIM) 가속기를 제안한다. 이를 위해 기존 3 차원 메모리인 Hybrid Memory Cube (HMC)에 하드웨어 및 소프트웨어 모듈을 추가로 구성하였다. 구체적으로, Processing Element (PE)간 데이터를 공유할 수 있는 공유 캐시 및 소프트웨어 스택, 파이프라인화된 곱셈기 및 듀얼 프리페치 버퍼를 구성하였다. 이를 유명 DCNN 알고리즘 LeNet, AlexNet, ZFNet, VGGNet, GoogleNet, RestNet 에 대해 성능 평가를 진행한 결과 기존 HMC 대비 40.3%의 속도 향상을 29.4%의 대역폭 향상을 보였다.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.15 no.3
• /
• pp.465-472
• /
• 2020

Recently, various deep neural networks based methods have been proposed for audio event detection. In this study, we improved the performance of audio event detection by adopting an attention approach to a baseline CRNN. We applied context gating at the input of the baseline CRNN and added an attention layer at the output. We improved the performance of the attention based CRNN by using the audio data of strong labels in frame units as well as the data of weak labels in clip levels. In the audio event detection experiments using the audio data from the Task 4 of the DCASE 2018/2019 Challenge, we could obtain maximally a 66% relative increase in the F-score in the proposed attention based CRNN compared with the baseline CRNN.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.25 no.3
• /
• pp.427-432
• /
• 2021

In this paper, we conduct performance analysis in automatic modulation classification of unknown communication signal to identify its modulation types based on deep neural network. The modulation classification performance was verified using time domain digital sample data of the modulated signal, frequency domain data to which FFT was applied, and time and frequency domain mixed data as neural network input data. For 11 types of analog and digitally modulated signals, the modulation classification performance was verified in various SNR environments ranging from -20 to 18 dB and reason for false classification was analyzed. In addition, by checking the learning speed according to the type of input data for neural network, proposed method is effective for constructing an practical automatic modulation recognition system that require a lot of time to learn.