• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.33 no.4
• /
• pp.621-631
• /
• 2023

Adversarial training improves the robustness of deep neural networks for adversarial examples. However, the previous adversarial training method focuses only on the adversarial loss function, ignoring that even a small perturbation of the input layer causes a significant change in the hidden layer features. Consequently, the accuracy of a defended model is reduced for various untrained situations such as clean samples or other attack techniques. Therefore, an architectural perspective is necessary to improve feature representation power to solve this problem. In this paper, we apply an attention module that generates an attention map of an input image to a general model and performs PGD adversarial training upon the augmented model. In our experiments on the CIFAR-10 dataset, the attention augmented model showed higher accuracy than the general model regardless of the network structure. In particular, the robust accuracy of our approach was consistently higher for various attacks such as PGD, FGSM, and BIM and more powerful adversaries. By visualizing the attention map, we further confirmed that the attention module extracts features of the correct class even for adversarial examples.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.33 no.6
• /
• pp.1087-1098
• /
• 2023

The rapid advancement of artificial intelligence (AI) technology has led to its proactive utilization across various fields. However, this widespread adoption of AI-based systems has raised concerns about the increasing threat of attacks on these systems. In particular, deep neural networks, commonly used in deep learning, have been found vulnerable to adversarial attacks that intentionally manipulate input data to induce model errors. In this study, we propose a method to protect image classification models from visually imperceptible One-Pixel attacks, where only a single pixel is altered in an image. The proposed defense technique utilizes an autoencoder model to remove potential threat elements from input images before forwarding them to the classification model. Experimental results, using the CIFAR-10 dataset, demonstrate that the autoencoder-based defense approach significantly improves the robustness of pretrained image classification models against One-Pixel attacks, with an average defense rate enhancement of 81.2%, all without the need for modifications to the existing models.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.11 no.10
• /
• pp.5132-5148
• /
• 2017

Cyber attacks are evolving commensurate with recent developments in information security technology. Intrusion detection systems collect various types of data from computers and networks to detect security threats and analyze the attack information. The large amount of data examined make the large number of computations and low detection rates problematic. Feature selection is expected to improve the classification performance and provide faster and more cost-effective results. Despite the various feature selection studies conducted for intrusion detection systems, it is difficult to automate feature selection because it is based on the knowledge of security experts. This paper proposes a feature selection technique to overcome the performance problems of intrusion detection systems. Focusing on feature selection, the first phase of the proposed system aims at constructing a feature subset using a sequential forward floating search (SFFS) to downsize the dimension of the variables. The second phase constructs a classification model with the selected feature subset using a random forest classifier (RFC) and evaluates the classification accuracy. Experiments were conducted with the NSL-KDD dataset using SFFS-RF, and the results indicated that feature selection techniques are a necessary preprocessing step to improve the overall system performance in systems that handle large datasets. They also verified that SFFS-RF could be used for data classification. In conclusion, SFFS-RF could be the key to improving the classification model performance in machine learning.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.33 no.3
• /
• pp.449-458
• /
• 2023

Artificial Intelligence is providing convenience in various fields using big data and deep learning technologies. However, deep learning technology is highly vulnerable to adversarial examples, which can cause misclassification of classification models. This study proposes a method to detect and purification various adversarial attacks using StarGAN. The proposed method trains a StarGAN model with added Categorical Entropy loss using adversarial examples generated by various attack methods to enable the Discriminator to detect adversarial examples and the Generator to purification them. Experimental results using the CIFAR-10 dataset showed an average detection performance of approximately 68.77%, an average purification performance of approximately 72.20%, and an average defense performance of approximately 93.11% derived from restoration and detection performance.

• Journal of the Korean Society of Visualization
• /
• v.22 no.1
• /
• pp.49-60
• /
• 2024

The purpose of this study establishes the deep neural network (DNN) and Decoder models to predict the flow and thermal fields of three-dimensional wavy wings as a passive flow control. The wide ranges of the wavy geometric parameters of wave amplitude and wave number are considered for the various the angles of attack and the aspect ratios of a wing. The huge dataset for training and test of the deep learning models are generated using computational fluid dynamics (CFD). The DNN and Decoder models exhibit quantitatively accurate predictions for aerodynamic coefficients and Nusselt numbers, also qualitative pressure, limiting streamlines, and Nusselt number distributions on the surface. Particularly, Decoder model regenerates the important flow features of tiny vortices in the valleys, which makes a delay of the stall. Also, the spiral vortical formation is realized by the Decoder model, which enhances the lift.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.26 no.5
• /
• pp.1235-1241
• /
• 2016

Personalized App recommendation system is recently famous since the number of various apps that can be used in smart phones that increases exponentially. However, the site users using google play site with malwares have experienced severe damages of privacy exposure and extortion as well as a simple damage of satisfaction descent at the same time. In addition, Sybil attack (Sybil) manipulating the score (rating) of each app with falmay also present because of the social networks development. Up until now, the sybil detection studies and malicious apps studies have been conducted independently. But it is important to determine finally the existence of intelligent attack with Sybil and malware simultaneously when we consider the intelligent attack types in real-time. Therefore, in this paper we experimentally evaluate the relationship between malware and sybils based on real cralwed dataset of goodlplay. Through the extensive evaluations, the correlation between malware and sybils is low for malware providers to hide themselves from Anti-Virus (AV).

• Journal of the Institute of Electronics and Information Engineers
• /
• v.50 no.3
• /
• pp.179-186
• /
• 2013

PACS is a medical system for digital medical images, and PACS expand to web-based service using public network, DICOM files should be protected from the man-in-the-middle attack because they have personal medical record. To solve the problem, we designed flexible secure transmission system using IPSec and adopted to a web-based three-dimensional medical image system. And next, we performed the performance evaluation changing integrity and encryption algorithm using DICOM volume dataset. At that time, combinations of the algorithm was 'DES-MD5', 'DES-SHA1', '3DES-MD5', and '3DES-SHA1, and the experiment was performed on our test-bed. In experimental result, the overall performance was affected by encryption algorithms than integrity algorithms, DES was approximately 50% of throughput degradation and 3DES was about to 65% of throughput degradation. Also when DICOM volume dataset was transmitted using secure transmission system, the network performance degradation had shown because of increased packet overhead. As a result, server and network performance degradation occurs for secure transmission system by ensuring the secure exchange of messages. Thus, if the secure transmission system adopted to the medical images that should be protected, it could solve server performance gradation and compose secure web PACS.

• Smart Media Journal
• /
• v.13 no.6
• /
• pp.9-15
• /
• 2024

In recent trends, there has been an increase in 'Qshing' attacks, a hybrid form of phishing that exploits fake QR (Quick Response) codes impersonating government agencies to steal personal and financial information. Particularly, this attack method is characterized by its stealthiness, as victims can be redirected to phishing pages or led to download malicious software simply by scanning a QR code, making it difficult for them to realize they have been targeted. In this paper, we have developed a classification technique utilizing machine learning algorithms to identify the maliciousness of URLs embedded in QR codes, and we have explored ways to integrate this with existing QR code readers. To this end, we constructed a dataset from 128,587 malicious URLs and 428,102 benign URLs, extracting 35 different features such as protocol and parameters, and used AutoML to identify the optimal algorithm and hyperparameters, achieving an accuracy of approximately 87.37%. Following this, we designed the integration of the trained classification model with existing QR code readers to implement a service capable of countering Qshing attacks. In conclusion, our findings confirm that deriving an optimized algorithm for classifying malicious URLs in QR codes and integrating it with existing QR code readers presents a viable solution to combat Qshing attacks.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.27 no.1
• /
• pp.39-47
• /
• 2017

Nowadays, there have been an increasing number of illegal use cases where people try to fabricate the working hours by using fake fingerprints. So, the fingerprint liveness detection techniques have been actively studied and widely demanded in various applications. This paper proposes a new method to detect fake fingerprints using CNN (Convolutional Neural Ntworks) based on the patches of fingerprint images. Fingerprint image is divided into small square sized patches and each patch is classified as live, fake, or background by the CNN. Finally, the fingerprint image is classified into either live or fake based on the voting result between the numbers of fake and live patches. The proposed method does not need preprocessing steps such as segmentation because it includes the background class in the patch classification. This method shows promising results of 3.06% average classification errors on LivDet2011, LivDet2013 and LivDet2015 dataset.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.17 no.3
• /
• pp.69-80
• /
• 2007

To protect personal information when releasing data, a general privacy-protecting technique is the removal of all the explicit identifiers, such as names and social security numbers. De-identifying data, however, provides no guarantee of anonymity because released information can be linked to publicly available information to identify them and to infer information that was not intended for release. In recent years, two emerging concepts in personal information protection are k-anonymity and $\ell$-diversity, which guarantees privacy against homogeneity and background knowledge attacks. While these solutions are signigicant in static data environment, they are insufficient in dynamic environments because of vulnerability to inference. Specially, the problem appeared in record deletion is to deconstruct the k-anonymity and $\ell$-diversity. In this paper, we present an approach to securely anonymizing a continuously changeable dataset in an efficient manner while assuring high data quality.