Model stealing attack query detection method for network intrusion detection system based on machine learning

By combining decision boundary leakage assessment and reconstruction error assessment, this method can quickly identify model theft attacks in ML-NIDS, solving the problem of identification and defense delays in existing technologies and achieving efficient and accurate model theft attack detection.

CN122339792APending Publication Date: 2026-07-03HARBIN INST OF TECH
1 Cites 0 Cited by

Patent Information

Application Number
CN202610529645.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-04-21
Publication Date
2026-07-03

AI Technical Summary

Technical Problem

Existing machine learning-based network intrusion detection systems (ML-NIDS) struggle to quickly and accurately identify and defend against model theft attacks with a limited number of queries. Traditional methods require a large number of query samples and have significant response delays, failing to meet real-time defense requirements.

Method used

A method combining decision boundary leakage assessment and reconstruction error assessment is adopted. By generating query samples to minimize costs and maximize information extraction, reconstruction error loss is calculated using an autoencoder, and model security risk assessment is combined to quickly identify model theft attacks.

Benefits of technology

It enables rapid and accurate identification of model theft attacks with a small number of query samples, improves detection rate and usability, reduces system response latency, is suitable for real-time defense scenarios, and demonstrates high-efficiency defense performance.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention discloses a model stealing attack query detection method for a network intrusion detection system based on machine learning, and relates to the technical field of network attack detection. The technical problem to be solved by the invention is that comprehensive analysis is carried out by combining query samples and model response, and a mode of combining decision boundary leakage evaluation and reconstruction error evaluation is adopted, so that model stealing attacks are quickly identified by a small number of query samples. According to the technical key points, a cumulative distribution function is constructed by utilizing a prediction result of a query sample, the uncertainty of a model is measured, and potential boundary detection behaviors are detected; the reconstruction error loss of the query sample is calculated by using an automatic encoder, and potential abnormal queries which deviate from normal distribution but do not show obvious malicious features are identified; and calculating a model safety coefficient based on results of decision boundary leakage evaluation and reconstruction error evaluation, so as to comprehensively evaluate the possibility of whether the model suffers from model stealing attack within a set period of time. The method is suitable for scenes where the ML-NIDS needs rapid detection and real-time defense.
Need to check novelty before this filing date? Find Prior Art