• Annual Conference of KIPS
• /
• 2024.05a
• /
• pp.207-209
• /
• 2024

딥러닝은 선형 연산과 비선형 연산을 조합하여 목표로 하는 시스템을 잘 표현할 수 있는 함수를 찾기 위해 사용하며, 이미지 분류 및 생성, 거대 언어 모델 및 객체 인식의 영역에서 활발하게 사용되고 있다. 그러나 딥러닝 연산을 위해서는 모델과, 연산을 수행하고자 하는 데이터가 하나의 공간에 저장되어야 한다. 모델과 데이터를 데이터 소유자가 관리할 경우, 데이터 소유자가 모델 데이터의 프라이버시를 침해할 수 있으며, 이는 모델을 적대적 예제 생성 공격에 취약하도록 만드는 원인이 된다. 한편 모델과 데이터를 모델 소유자가 관리할 경우, 모델 소유자는 데이터의 프라이버시를 침해하여 데이터 소유자의 정보를 악의적으로 이용할 수 있다. 본 논문에서는 딥러닝 모델과 데이터의 프라이버시를 모두 보호하기 위해 주어진 딥러닝 모델의 암호화와 복호화를 수행하는 EncNet 을 구현하였으며, MNIST 와 Cifat-10 데이터셋에 대하여 실효성을 테스트하였다.

• Annual Conference of KIPS
• /
• 2015.04a
• /
• pp.351-353
• /
• 2015

공공정보 공유 및 개방, 소셜네트워크서비스의 활성화 그리고 사용자 간의 공유 데이터 증가 등의 이유로 인터넷상에 노출되는 사용자의 개인 정보가 증가하고 있다. 인터넷상에 노출된 사용자들의 개인정보들은 연결공격(linkage attack), 배경지식 공격(background attack)으로 프라이버시를 침해할 수 있다. 이를 막기 위해 관계형 데이터베이스에서는 대표적으로 k-익명성(k-anonymity)을 시작으로 l-다양성(l-diversity), t-밀집성(t-closeness)이라는 익명화 모델이 제안되었으며 계속해서 익명화 알고리즘의 성능은 개선되고 있다. 하지만 k-익명성, l-다양성, t-밀집성 모델의 조건을 만족하기 위해서는 준식별자(quasi-identifier)를 일반화(generalization)처리 해주어야 하는데 이 과정에서 준식별자의 가치를 손실된다는 단점이 있다. 본 논문에서 준식별자의 정보 손실을 최소화하기 위해 k-익명성 모델을 만족시키는 과정에서 일반화와 데이터를 삽입을 사용하는 익명화 처리하는 방법을 제안한다.

• Convergence Security Journal
• /
• v.21 no.3
• /
• pp.49-56
• /
• 2021

Recently, artificial intelligence is regarded as an essential technology in our society. In particular, the invasion of privacy in artificial intelligence has become a serious problem in modern society. Split learning, proposed at MIT in 2019 for privacy protection, is a type of federated learning technique that does not share any raw data. In this study, we studied a safe and accurate segmentation learning model using known differential privacy to safely manage data. In addition, we trained SVHN and GTSRB on a split learning model to which 15 different types of differential privacy are applied, and checked whether the learning is stable. By conducting a learning data extraction attack, a differential privacy budget that prevents attacks is quantitatively derived through MSE.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.35 no.4B
• /
• pp.566-575
• /
• 2010

In recent, artificial immune system has become an important research direction in the anomaly detection of networks. The conventional artificial immune systems are usually based on the negative selection that is one of the computational models of self/nonself discrimination. A main problem with self and non-self discrimination is the determination of the frontier between self and non-self. It causes false positive and false negative which are wrong detections. Therefore, additional functions are needed in order to detect potential anomaly while identifying abnormal behavior from analogous symptoms. In this paper, we design novel network attack detection and response schemes based on artificial immune system, and evaluate the performance of the proposed schemes. We firstly generate detector set and design detection and response modules through adopting the interaction between dendritic cells and T-cells. With the sequence of buffer occupancy, a set of detectors is generated by negative selection. The detection module detects the network anomaly with a set of detectors and generates alarm signal to the response module. In order to reduce wrong detections, we also utilize the fuzzy number theory that infers the degree of threat. The degree of threat is calculated by monitoring the number of alarm signals and the intensity of alarm occurrence. The response module sends the control signal to attackers to limit the attack traffic.

• Journal of KIISE:Computing Practices and Letters
• /
• v.8 no.4
• /
• pp.457-467
• /
• 2002

In order to simulate cyber attacks and predict network behavior by attacks, we should represent attributes of network components in the simulation model, and should express characteristics of systems that carry out various cyber attacks and defend from these attacks. To simulate how network load may change under the cyber attacks, we extended SSF[9, 10] that is process-based event-oriented simulation system. We added a firewall class and a packet manipulator into the SSFNet that is a component of SSF. The firewall class, which is related to the security, is to simulate cyber attacks, and the packet manipulator is a set of functions to write attack programs for the simulation. The extended SSFNet enables to simulate a network with the security systems and provides advantages that make easy to port already exsiting attack programs and apply them to the simulation evironment. We made a vitual network model to verify operations of the added classes, and simulated a smurf attack that is a representative denial of sevive attack, and observed the network behavior under the smurf attack. The results showed that the firewall class and packet manipulator developed in this paper worked normaly.

• Convergence Security Journal
• /
• v.4 no.2
• /
• pp.17-25
• /
• 2004

As Internet and Internet users are rapidly increasing and getting popularized in the world the existing firewall has limitations to detect attacks which exploit vulnerability of web server. And these attacks are increasing. Most of all, intrusions using web application's programming error are occupying for the most part. In this paper, we introduced real-time web-server agent which analyze web-server based log and detect web-based attacks after the analysis of the web-application's vulnerability. We propose the method using real-time agent which remove Process ID(pid) and block out attacker's If if it detects the intrusion through the decision stage after judging attack types and patterns.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.38C no.11
• /
• pp.1074-1077
• /
• 2013

This paper analyzes the benefit of using directional antennas against eavesdropping attack in millimeter wave (mmWave)-based networks. All devices are equipped with a directional antenna or an omni-directional antenna in a single-hop communications. The probability of a device being detected by an eavesdropper is analyzed based on the exposure region of a device. The relative detection rate is introduced to represent the benefit of using directional antenna. Numerical results show that there exists an optimal number of devices that maximizes the detection probability and it varies according to the parameters such as antenna beamwidth. It shows that the use of directional antenna enables to protect the devices from the detection by an eavesdropper for almost the whole situation in mmWave band communication.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2012.10a
• /
• pp.753-756
• /
• 2012

The type of phishing changes rapidly and also threat model changes very fastly Accordingly, frauds develop new methods of attacks to avoid the counterparts. Recently, the type of phishing in internet banking is developing specifically. In this paper, to help encounter for it, we first review the meaning of phishing and the types of attacks in phishing in the second chapter,and in the third chapter, we will analyze phishing which is using malicious code in internet banking,and in the fourth chapter, we will describe the conclusion of this paper.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.15 no.8
• /
• pp.5256-5262
• /
• 2014

Recently, Distributed Denial of Service (DDoS) attacks, such as spreading malicious code, cyber-terrorism, have occurred in government agencies, the press and the financial sector. DDoS attacks are the simplest Internet-based infringement attacks techniques that have fatal consequences. DDoS attacks have caused bandwidth consumption at the network layer. These attacks are difficult to detect defend against because the attack packets are not significantly different from normal traffic. Abnormal traffic is threatening the stability of the network. Therefore, the abnormal traffic by generating indications will need to be detected in advance. This study examined the abnormal traffic detection technique using a forecasting model-based trend model.

• Journal of Digital Convergence
• /
• v.12 no.8
• /
• pp.229-234
• /
• 2014

Vehicular electronic communication system has continued to develop in favor of high performance and user convenience with the evolution of auto industry. Yet, due to the nature of communication system, concerns over intruder attacks in transmission sections have been raised with a need for safe and secure communication being valued. Any successful intruder attacks on vehicular operation and control systems as well as on visual equipment could result in serious safety and privacy problems. Thus, research has focused on hardware-based security and secure communication protocols. This paper proposed a safe and secure vehicular communication protocol, used the formal verification tool, Casper/FDR to test the security of the proposed protocol against different types of intruder attacks, and verified that the proposed protocol was secure and ended without problems.