Model security detection method based on generative adversarial network
The invention discloses a model security detection method based on a generative adversarial network. The method comprises the following specific steps of: 1, setting a behavior similarity security threshold value delta; 2, constructing an initialization generator G and a substitution model D; 3, car...
Gespeichert in:
Hauptverfasser: | , , , |
---|---|
Format: | Patent |
Sprache: | chi ; eng |
Schlagworte: | |
Online-Zugang: | Volltext bestellen |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
Zusammenfassung: | The invention discloses a model security detection method based on a generative adversarial network. The method comprises the following specific steps of: 1, setting a behavior similarity security threshold value delta; 2, constructing an initialization generator G and a substitution model D; 3, carrying out iteration round by round, wherein the behavior similarity mu between the substitution model D and the tested model T is calculated, and the next step is carried out after the behavior similarity mu reaches a set value; and 4, evaluating the safety of the tested model T. The model stealing method based on the GAN is suitable for a black box attack scene without training data, and the behavior similarity of the substitution model and the tested model is rapidly improved by generating artificial data with relatively balanced categories. According to experimental results, the method has the characteristics of high adaptability, high efficiency and the like, and the risk of model stealing attack after the acce |
---|