Adversarial sample detection method and device, computing equipment and computer storage medium

Abstract

The invention relates to the technical field of machine learning, and discloses an adversarial sample detection method and device, computing equipment and a computer storage medium. The method comprises the following steps: acquiring a training sample and a corresponding training sample label, wherein the training sample label comprises a normal sample and an adversarial sample; inputting the training sample into a target model to obtain a first prediction score vector of the training sample; adding N times of random disturbance to the training sample to obtain N groups of contrast training samples; respectively inputting the N groups of contrast training samples into a target model to obtain a second prediction score vector of each group of contrast training samples; constructing featuredata according to the first prediction score vector and the second prediction score vector of each group of contrast training samples; training a classification model according to the feature data andthe training sample label corresponding to the feature data to obtain a detector; and detecting the input test data according to the detector. According to the embodiment of the invention, reliable detection of the adversarial sample can be realized according to the detector.

Description

technical field [0001] The present application relates to the technical field of machine learning, and in particular to an adversarial sample detection method, device, computing device and computer storage medium. Background technique [0002] As an important data analysis tool, machine learning is widely used in many application fields such as biometric identification, automobile automatic driving, and machine vision. While machine learning brings great convenience to people, it also exposes some security issues. Adding small, imperceptible perturbations to the original samples generates adversarial samples, and machine learning models are vulnerable to adversarial samples. For example, according to the characteristics of the facial recognition model, the facial recognition model can make wrong classifications by adding small perturbations to the original facial pictures. There are also malicious controls for autonomous car driving, voice control systems, and more. Attac...

AI Technical Summary

Problems solved by technology

Currently, there is no adversarial example dete

Images