Machine learning model-oriented member reasoning privacy attack method and system

Abstract

The invention relates to a machine learning model-oriented member reasoning privacy attack method and system. The method comprises the steps of obtaining a target model and target data; generating an adversarial sample by adopting an adversarial sample generation algorithm according to the target data, the adversarial sample generation algorithm comprising an adaptive greedy algorithm and binary search algorithm combined method or an embedded mapping algorithm on a manifold interface by means of a principal component technology; determining an Euclidean distance between the target data and the corresponding adversarial sample; determining a judgment result according to the Euclidean distance, and realizing member reasoning; the judgment result comprises that the target data belongs to a training data set or a test data set. According to the method, the problems of high access cost, weak mobility and poor robustness of black box member reasoning attacks can be solved.

Description

Technical field [0001] The present invention relates to the field of machine learning, particularly to a member for privacy attacks reasoning methods and systems of machine learning models. Background technique [0002] Things, big data, cloud computing and other new technology enables the collection, storage and processing vast amounts of data possible, especially the rapid development of artificial intelligence machine learning theory and technology, has been widely used in various fields of security, transportation, health care and so on. At the same time, security and privacy issues of machine learning has become the focus of attention, some academics have suggested the attack against the sample, data poisoning attack, model inference and reasoning, and other members of the security model and privacy attacks. These effective method of attack has sparked fears of machine learning, but also become the driving force of endogenous development of machine learning, promote scientif...

AI Technical Summary

Problems solved by technology

[0007] The purpose of the present invention is to provide a machine learning model-oriented member reasoning privacy attack method and system, which can solve the problems of high access cost, weak transferability, and poor robustness in black-box member reasoning attacks

Images