METHOD OF GENERATING ADVERSARIAL EXAMPLES FOR NETWORK INTRUSION DETECTION SYSTEM Russian patent published in 2024 - IPC G06N3/08 

FIELD: information technology.

SUBSTANCE: invention relates to information technology, in particular to information security, and can be used to train intrusion detection systems. In the method of generating adversarial examples for a network intrusion detection system, two checks are additionally carried out before testing in a real industrial control system, specifically estimating, for each found adversarial example, the perturbation tolerance applied to the original example, according to the specified allowable values in the selected feature space, and the result is used to check the dependence of the features or the found adversarial example is not accepted as an effective adversarial example. Further, a matrix of dependence of features in the formed feature space is formed, then for each feature of the found adversarial example, the value of which in the original example is changed, its dependent features defined in the feature dependency matrix are checked for change in the initial example and the result is used to proceed to testing or not to be accepted as an effective adversarial example.

EFFECT: technical result when generating adversarial examples for a network intrusion detection system is to increase the speed of searching for effective adversarial examples.

1 cl, 2 dwg