Safety assessment framework building method for deep learning bypass analysis

A technology of security assessment and deep learning, which is applied in the field of security assessment framework construction of deep learning bypass analysis, which can solve problems such as cumbersome feature process, interference of feature extraction ability, inability to accurately extract data features, etc., and achieve low order of magnitude requirements and accurate measurement Effect

Inactive Publication Date: 2021-01-15
ARMY ENG UNIV OF PLA
View PDF4 Cites 1 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0003] Other statistics-based feature extraction methods such as DOM, SOSD, SOST, and T detection and measurement are proposed based on signal-to-noise ratio technology, and the feature extraction process needs to be classified according to the feature model, and the feature process is too cumbersome
The improved TVLA feature extraction technology is convenient and efficient, but it is the same as the signal-to-noise ratio, and the feature extraction capabilities of both will be disturbed for protected bypass data
In addition, variants of other dimensionality reduction techniques such as fractional analysis or kernel discriminant analysis methods are also applied to features, but cannot extract data-dependent features accurately

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Safety assessment framework building method for deep learning bypass analysis
  • Safety assessment framework building method for deep learning bypass analysis
  • Safety assessment framework building method for deep learning bypass analysis

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0025] Embodiment 1, key information amount and bypass security assessment

[0026] (1) Key information volume and bypass security assessment

[0027] Although DLSCA is a combination of deep learning and bypass analysis, it is still a bypass analysis problem. To solve the DLSCA evaluation problem, it is necessary to start from the perspective of bypass security analysis. In the whole process of SCA, the probability distribution of each moment is only related to the probability distribution of the previous moment, so the process can be regarded as a Markov process:

[0028] Lemma 3-1 The SCA process can be defined as a Markov process: Among them, D is the bypass distinguisher, and K^ is the key value corresponding to the DNN model prediction category.

[0029] According to the SCA Markov process, combined with the DLSCA implementation steps, the DLSCA security assessment should address the following issues:

[0030] Question 3-1 (Evaluation Question) Given a training datase...

Embodiment 2

[0047] Embodiment 2, key information amount and DNN performance evaluation

[0048] (1) Information bottleneck theory explains DNN

[0049] DNN has always been considered as a black box model due to the complexity of the learning process. The previous section pointed out that machine learning performance metrics are not available, so it is necessary to find additional metrics to correlate key information. According to information bottleneck theory

[0050] In theory, the association between DNN layers and layers can be regarded as a Markov process, and the mutual information transferred in this process can measure the degree of learning:

[0051] Lemma 3-2 (Information Bottleneck Theory) In the hypothesis space H, the DNN model The structure can be interpreted as a Bayesian hierarchical structure. Because the input of the hidden layer i is the output T of the previous layer i-1 ,but Can be equivalent to a Markov chain: The corresponding probability distribution is: ...

Embodiment 3

[0058] Example 3, DLSCA evaluation index - PI (Z; T, θ)

[0059] The essence of DNN training and learning is the optimization of the cross-entropy by the law of maximum likelihood. The more accurate the DNN model learning is, the cross-entropy CX, Z(θ) and the conditional guess entropy Gm tend to a fixed value H[Z|X] (Theorem 3 -1). Therefore, finding an evaluation index to establish a quantitative relationship with the cross entropy CX,Z(θ) can realize the DLSCA security evaluation framework with the key information as the core.

[0060] (1) Masure evaluation principle

[0061] Using the relationship between perceptual information and cross entropy function, DLSCA is evaluated by calculating the perceptual information of the X→Z process, but the error of this method is too large to accurately quantify and evaluate DLSCA. First introduce its evaluation principle:

[0062] ① Due to the random dimensionality reduction mechanism of the SGD algorithm, the actual sample size NSG...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

No PUM Login to view more

Abstract

The invention discloses a safety assessment framework building method for deep learning bypass analysis. The method comprises the steps that a secret key information quantum framework based on modeling output and a deep learning sub-framework composed of cross entropy and a loss function are included; and the bypass analysis sub-framework consists of a condition guessing entropy, a derivative guessing entropy and a success rate. According to information bottleneck theoretical analysis, perception information from the last hidden layer to the output layer of the DNN is calculated through probability distribution to serve as a quantitative evaluation index of the DLSCA, and experiments prove that the index is more accurate in measurement compared with an existing quantitative evaluation index and has lower requirements for the order of magnitude of data.

Description

technical field [0001] The invention relates to a method for building a security assessment framework, in particular to a method for building a security assessment framework for deep learning bypass analysis. Background technique [0002] DLSCA technology has the ability to automatically extract features, and there is no need for preprocessing of feature point selection in the modeling stage. But if you want to study the specific relationship between DNN structural parameters and side channel attack scenarios, feature extraction technology is indispensable. In addition to the ability to automatically extract features, due to the good robustness of neural networks, DLSCA also has good feature extraction capabilities for protected bypass power consumption data. Based on the above advantages of DNN feature extraction, if the process of learning features can be analyzed inside the DNN "black box", then for various bypass attack fields, the adversary can accurately extract the f...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Application Information

Patent Timeline
no application Login to view more
Patent Type & Authority Applications(China)
IPC IPC(8): G06F21/75G06N3/04G06N3/08G06N7/00
CPCG06F21/75G06N3/08G06N7/01G06N3/045
Inventor 陈开颜张阳李雄伟宋世杰王寅龙李玺谢志英李艳谢方方刘林云
Owner ARMY ENG UNIV OF PLA
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap