• 제목/요약/키워드: Inversion Attack

검색결과 16건 처리시간 0.01초

Textual Inversion을 활용한 Adversarial Prompt 생성 기반 Text-to-Image 모델에 대한 멤버십 추론 공격 (Membership Inference Attack against Text-to-Image Model Based on Generating Adversarial Prompt Using Textual Inversion)

  • 오윤주;박소희;최대선
    • 정보보호학회논문지
    • /
    • 제33권6호
    • /
    • pp.1111-1123
    • /
    • 2023
  • 최근 생성 모델이 발전함에 따라 생성 모델을 위협하는 연구도 활발히 진행되고 있다. 본 논문은 Text-to-Image 모델에 대한 멤버십 추론 공격을 위한 새로운 제안 방법을 소개한다. 기존의 Text-to-Image 모델에 대한 멤버십 추론 공격은 쿼리 이미지의 caption으로 단일 이미지를 생성하여 멤버십을 추론하였다. 반면, 본 논문은 Textual Inversion을 통해 쿼리 이미지에 personalization된 임베딩을 사용하고, Adversarial Prompt 생성 방법으로 여러 장의 이미지를 효과적으로 생성하는 멤버십 추론 공격을 제안한다. 또한, Text-to-Image 모델 중 주목받고 있는 Stable Diffusion 모델에 대한 멤버십 추론 공격을 최초로 진행하였으며, 최대 1.00의 Accuracy를 달성한다.

Autopilot Design for Agile Missile with Aerodynamic Fin and Thrust Vecotring Control

  • Lee, Ho-Chul;Choi, Yong-Seok;Choi, Jae-Weon
    • 제어로봇시스템학회:학술대회논문집
    • /
    • 제어로봇시스템학회 2003년도 ICCAS
    • /
    • pp.525-530
    • /
    • 2003
  • This paper is concerned with a control allocation strategy using the dynamic inversion which generates the nominal control input trajectories, and autopilot design using the time-varying control technique which is time-varying version of pole placement of linear time-invariant system for an agile missile with aerodynamic fin and thrust vectoring control. Dynamic inversion can decide the amount of the deflection of each control effector, aerodynamic fin and thrust vectoring control, to extract the maximum performance by combining the action of them. Time-varying control technique for autopilot design enhance the robustness of the tracking performance for a reference command. Nonlinear simulations demonstrates the dynamic inversion provides the effective nominal control input trajectories to achieve the angle of attack command, and time-varying control technique exhibits good robustness for a wide range of angle of attack.

  • PDF

Trajectory Guidance and Control for a Small UAV

  • Sato, Yoichi;Yamasaki, Takeshi;Takano, Hiroyuki;Baba, Yoriaki
    • International Journal of Aeronautical and Space Sciences
    • /
    • 제7권2호
    • /
    • pp.137-144
    • /
    • 2006
  • The objective of this paper is to present trajectory guidance and control system with a dynamic inversion for a small unmanned aerial vehicle (UAV). The UAV model is expressed by fixed-mass rigid-body six-degree-of-freedom equations of motion, which include the detailed aerodynamic coefficients, the engine model and the actuator models that have lags and limits. A trajectory is generated from the given waypoints using cubic spline functions of a flight distance. The commanded values of an angle of attack, a sideslip angle, a bank angle and a thrust, are calculated from guidance forces to trace the flight trajectory. To adapt various waypoint locations, a proportional navigation is combined with the guidance system. By the decision logic, appropriate guidance law is selected. The flight control system to achieve the commands is designed using a dynamic inversion approach. For a dynamic inversion controller we use the two-timescale assumption that separates the fast dynamics, involving the angular rates of the aircraft, from the slow dynamics, which include angle of attack, sideslip angle, and bank angle. Some numerical simulations are conducted to see the performance of the proposed guidance and control system.

AI 환경에서 모델 전도 공격에 안전한 차분 프라이버시 기술 (Differential Privacy Technology Resistant to the Model Inversion Attack in AI Environments)

  • 박철희;홍도원
    • 정보보호학회논문지
    • /
    • 제29권3호
    • /
    • pp.589-598
    • /
    • 2019
  • 온라인상에 축적되는 디지털 데이터의 양은 폭발적으로 증가하고 있으며 이러한 데이터들은 매우 큰 잠재적 가치를 갖고 있다. 국가 및 기업들은 방대한 양의 데이터로부터 다양한 부가가치를 창출하고 있으며 데이터 분석 기술에 많은 투자를 하고 있다. 그러나 데이터 분석에서 발생하는 프라이버시 문제는 데이터의 활용을 저해하는 큰 요인으로 작용하고 있다. 최근 신경망 모델 기반의 분석 기술에 대한 프라이버시 침해 공격들이 제안됨에 따라 프라이버시를 보존하는 인공 신경망 기술에 대한 연구가 요구되고 있다. 이에 따라 엄격한 프라이버시를 보장하는 차분 프라이버시 분야에서 다양한 프라이버시 보존형 인공 신경망 기술에 대한 연구가 수행되고 있지만, 신경망 모델의 정확도와 프라이버시 보존 강도 사이의 균형이 적절하지 않은 문제점이 있다. 본 논문에서는 프라이버시와 모델의 성능을 모두 보존하고 모델 전도 공격에 저항성을 갖는 차분 프라이버시 기술을 제안한다. 또한, 프라이버시 보존 강도에 따른 모델전도 공격의 저항성을 분석한다.

Model Inversion Attack: Analysis under Gray-box Scenario on Deep Learning based Face Recognition System

  • Khosravy, Mahdi;Nakamura, Kazuaki;Hirose, Yuki;Nitta, Naoko;Babaguchi, Noboru
    • KSII Transactions on Internet and Information Systems (TIIS)
    • /
    • 제15권3호
    • /
    • pp.1100-1118
    • /
    • 2021
  • In a wide range of ML applications, the training data contains privacy-sensitive information that should be kept secure. Training the ML systems by privacy-sensitive data makes the ML model inherent to the data. As the structure of the model has been fine-tuned by training data, the model can be abused for accessing the data by the estimation in a reverse process called model inversion attack (MIA). Although, MIA has been applied to shallow neural network models of recognizers in literature and its threat in privacy violation has been approved, in the case of a deep learning (DL) model, its efficiency was under question. It was due to the complexity of a DL model structure, big number of DL model parameters, the huge size of training data, big number of registered users to a DL model and thereof big number of class labels. This research work first analyses the possibility of MIA on a deep learning model of a recognition system, namely a face recognizer. Second, despite the conventional MIA under the white box scenario of having partial access to the users' non-sensitive information in addition to the model structure, the MIA is implemented on a deep face recognition system by just having the model structure and parameters but not any user information. In this aspect, it is under a semi-white box scenario or in other words a gray-box scenario. The experimental results in targeting five registered users of a CNN-based face recognition system approve the possibility of regeneration of users' face images even for a deep model by MIA under a gray box scenario. Although, for some images the evaluation recognition score is low and the generated images are not easily recognizable, but for some other images the score is high and facial features of the targeted identities are observable. The objective and subjective evaluations demonstrate that privacy cyber-attack by MIA on a deep recognition system not only is feasible but also is a serious threat with increasing alert state in the future as there is considerable potential for integration more advanced ML techniques to MIA.

Aircraft CAS Design with Input Saturation Using Dynamic Model Inversion

  • Sangsoo Lim;Kim, Byoung-Soo
    • International Journal of Control, Automation, and Systems
    • /
    • 제1권3호
    • /
    • pp.315-320
    • /
    • 2003
  • This paper presents a control augmentation system (CAS) based on the dynamic model inversion (DMI) architecture for a highly maneuverable aircraft. In the application of DMI not treating actuator dynamics, significant instabilities arise due to limitations on the aircraft inputs, such as actuator time delay based on dynamics and actuator displacement limit. Actuator input saturation usually occurs during high angles of attack maneuvering in low dynamic pressure conditions. The pseudo-control hedging (PCH) algorithm is applied to prevent or delay the instability of the CAS due to a slow actuator or occurrence of actuator saturation. The performance of the proposed CAS with PCH architecture is demonstrated through a nonlinear flight simulation.

A Study on Longitudinal Phugoid Mode Affected by Application of Nonlinear Control Laws

  • Kim, Chong-Sup;Hur, Gi-Bong;Kim, Seung-Jun
    • International Journal of Aeronautical and Space Sciences
    • /
    • 제8권1호
    • /
    • pp.21-31
    • /
    • 2007
  • Relaxed Static Stability (RSS) concept has been applied to improve aerodynamic performance of modern version supersonic jet fighter aircraft. The T-50 advanced supersonic trainer employs the RSS concept in order to improve the aerodynamic performance. And the flight control system stabilizes the unstable aircraft and provides adequate handling qualities. The T-50 longitudinal control laws employ a proportional-plus-integral type controller based on a dynamic inversion method. The longitudinal dynamic modes consist of short period with high frequency and phugoid mode with low frequency. The design goal of longitudinal control law is optimization of short period damping ratio and frequency using Lower Order Equivalent System (LOES) complying the requirement of MIL-F-8785C. This paper addresses phugoid mode characteristics such as damping ratio and natural frequency that is affected by the nonlinear control laws such as angle of attack limiter, auto pitch attitude command system and autopilot of pitch attitude hold.

The Linearity of algebraic Inversion and a Modification of Knudsen-Nyberg Cipher

  • Lee, Chang-Hyi;Lim, Jong-In
    • 정보보호학회논문지
    • /
    • 제8권1호
    • /
    • pp.65-70
    • /
    • 1998
  • K. Nyberg and L.R. Knudsen showed a prototype of a DES-like cipher$^{[1]}$ which has a provable security against differential cryptanalysis. But in the last year, at FSE'97 T. Jakobsen ane L.R.Knudsen broked it by using higher order differential attack and interpolation attack$^{[2]}$ . Furthermore the cipher was just a theoretically proposed one to demonstrate how to construct a cipher which is procably secure against differential cryptanalysis$^{[3]}$ and it was suspected to have a large complexity for its implementation.Inthis paper the two improved results for the dfficidnt hardware and software implementation.

버퍼 오버플로우 공격 방어를 위한 스택 역위 탐지의 성능 평가 (Performance Evaluation of Detecting Stack Inversion for Defending from Buffer Overflow Attack)

  • 양한근;표창우;이경호
    • 한국정보과학회:학술대회논문집
    • /
    • 한국정보과학회 2002년도 가을 학술발표논문집 Vol.29 No.2 (1)
    • /
    • pp.586-588
    • /
    • 2002
  • 실행 시간 스택 프레임의 하단과 상단을 가리키는 프레임 포인터와 스택 포인터는 항상 일정한 대소 관계를 유지한다. 선형 스택 공격이 진행되면, 이관계가 반전된다. 이때 스택이 역위되었다고 한다. 본 논문은 x86프로세서 계열의 gcc 컴파일러에 스택 역위 탐지기능을 부여하여, 이 컴파일러를 사용하였을 때 실행 프로그램의 성능에 미치는 영향을 분석하였다.

  • PDF

프라이버시 침해에 대응하는 분할 학습 모델 연구 (A Study of Split Learning Model to Protect Privacy)

  • 유지현;원동호;이영숙
    • 융합보안논문지
    • /
    • 제21권3호
    • /
    • pp.49-56
    • /
    • 2021
  • 현대의 인공지능은 사회를 구성하는 필수적인 기술로 여겨지고 있다. 특히, 인공지능에서 프라이버시 침해 문제는 현대 사회에서 심각한 문제로 자리 잡고 있다. 개인정보보호를 위해 2019년 MIT에서 제안된 분할 학습은 연합 학습의 기술 중 하나로 개인정보보호 효과를 지닌다. 본 연구에서는 데이터를 안전하게 관리하기 위해 알려진 차분 프라이버시를 이용하여 안전하고 정확한 분할 학습 모델을 연구한다. 또한, SVHN과 GTSRB 데이터 세트를 15가지의 차등적인 차분 프라이버시를 적용한 분할 학습 모델에 학습시키고 학습이 안정적으로 되는지를 확인한다. 최종적으로, 학습 데이터 추출 공격을 진행하여, 공격을 예방하는 차분 프라이버시 예산을 MSE를 통해 정량적으로 도출한다.