Robustness Evaluation of Authentication Models via Local Lipschitz Constants

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing methods for evaluating the robustness of authentication models against adversarial examples are unable to calculate the robustness using a feature extractor, a template of authentication target data, and a threshold value.

Innovation Solution

A robustness evaluation device and method that calculates the similarity between an input and a template, estimates a local Lipschitz constant, and uses these values to determine the robustness evaluation of an authentication model, incorporating a similarity calculation unit, a local Lipschitz constant estimation unit, and an evaluation value estimation unit.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If a classifier outputs classification degrees for multiple classes, then classification capability is improved, but the ability to evaluate robustness using feature extractors and templates is lost

Engineering Contradiction:

Improveclassification capabilityVSAvoidrobustness evaluation capability

Core Design Contradiction:

Adaptability or versatilityVSReliability

Solution Approach 1:

The patent segments the authentication model into distinct functional components: a feature extractor that generates feature vectors from input data, and a template comparison unit that calculates similarity between features and templates. This segmentation enables separate optimization and evaluation of each component's contribution to robustness, allowing robustness evaluation even when the overall system uses complex multi-class classification.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent introduces feature vectors as an intermediary representation between raw input data and classification decisions. By evaluating robustness at the feature level rather than at the final classification output, the system can assess robustness properties independently of the specific classification architecture, enabling universal robustness evaluation across different model types.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Measurement precision

If deep learning is used to maximize classification degree of true class, then classification accuracy is improved, but vulnerability to adversarial examples increases

Engineering Contradiction:

Improveclassification accuracyVSAvoidadversarial example vulnerability

Core Design Contradiction:

Measurement precisionVSObject-affected harmful factors

Solution Approach 1:

The patent performs preliminary robustness evaluation by calculating the local Lipschitz constant of the feature extractor before deployment. This pre-assessment of feature sensitivity to input perturbations allows developers to identify and address adversarial vulnerabilities during the model development phase, rather than discovering them only after deployment when adversarial examples are encountered.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent replaces direct analysis of classification decisions with mathematical analysis of the feature extraction function's Lipschitz constant. This substitution transforms the problem from analyzing complex neural network classification behavior to analyzing the geometric properties of the feature mapping, providing a more tractable and interpretable approach to assessing adversarial vulnerability.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

3Measurement precision

If similarity calculation between feature and template is performed, then authentication accuracy is improved, but calculation of robustness evaluation value becomes complex

Engineering Contradiction:

Improveauthentication accuracyVSAvoidrobustness evaluation complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent changes the evaluation parameter from analyzing the entire authentication pipeline to focusing specifically on the local Lipschitz constant of the feature extractor. This parameter change simplifies the robustness evaluation by concentrating on the most critical component—the feature mapping—rather than attempting to analyze the complexity of similarity calculations, template matching, and classification decisions jointly.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS20240062109A1Robustness evaluation device, robustness evaluation method and recording medium

Publication Date: 2024.02.22 NEC CORP

AI summary

A robustness evaluation device includes a similarity calculation unit that calculates the similarity between a feature of an input to an authentication model and a feature of a template; a local Lipschitz constant estimation unit, that estimates a local Lipschitz constant of a function for calculating similarity between the feature of the input to the authentication model and the feature of the template, in a sphere centered on the input to the authentication model; and an evaluation value estimation unit that estimates the evaluation value of robustness of the authentication model based on the similarity, the determination threshold value for the similarity, and the local Lipschitz constant.