Testing method and system for attacking machine learning products

Abstract

The invention discloses a test method for attacking a machine learning product, comprising the following steps: using a malicious sample test set to periodically attack and access the machine learning product; the machine learning product obtains multiple pieces of malicious access data in the malicious sample test set characteristic factors, and analyze the characteristic factors, and classify and manage the characteristic factors according to the analysis results; once again, a piece of malicious access data in the malicious sample test set is used to attack and access the machine learning product, and the machine learning product maliciously accesses the piece of data The feature factors in the access data are analyzed, and compared with the feature factors in the classified normal access feature value library and abnormal access feature value library, it is judged whether it is a normal access or an abnormal access; the judgment result is output; the invention also discloses A test system for attacking machine learning products, the invention further improves the security of machine learning products.

Description

technical field [0001] The invention relates to the technical field of Internet security testing, in particular to a testing method and system for attacking machine learning products. Background technique [0002] With the rapid development of the Internet, the security of the Internet has been paid more and more attention by people, and the attack and protection methods of the Internet are also gradually upgraded. The rise and application of machine learning have brought a great challenge to security testing. After a long period of development, the application of machine learning has emerged in various fields. Since there is no effective test method for network application modules made by machine learning, a test method for network attackers to use machine learning products to attack has emerged as the times require. [0003] At present, when conducting security testing of machine learning products, there are limitations such as traditional, single testing methods, and lac...

AI Technical Summary

Problems solved by technology

[0003] At present, when conducting security testing of machine learning products, there are limitations such as traditional, single testing methods, and lack of significant test results.

Images