Classifier Certainty Factor Correction Against Membership Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing class classification services lack sufficient resistance to membership inference attacks, which can lead to the leakage of personal information by determining if specific data is included in training data.
Innovation Solution
An information processing method that calculates certainty factors for classification data, corrects the maximum certainty factor to be within a certain value range, and maintains the order of certainty factors to prevent attackers from accurately determining the inclusion of data in training data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the maximum certainty factor is corrected to be within a certain value range, then resistance to membership inference attack is enhanced, but the original classification confidence is reduced
Solution Approach 1:
The patent applies parameter changes by modifying the certainty factor values within a specific range [ε, 1-ε] to prevent membership inference attacks. The correction process adjusts the maximum certainty factor and redistributes the difference among other classes, transforming the original confidence values into corrected values that maintain security while preserving classification utility.
Solution Approach 2:
The patent introduces an intermediary correction mechanism that acts as a mediator between the original model output and the final classification result. This intermediary process (certainty factor correction) transforms the raw model outputs into secured outputs without requiring changes to the underlying model or training data, thus protecting against attacks while maintaining service functionality.
2Object-affected harmful factors
If certainty factor correction is performed, then personal information protection is improved, but processing complexity increases
Solution Approach 1:
The patent extracts the security function from the core classification process by implementing a separate correction module that operates on the model outputs. This extraction allows the classification service to maintain its primary function while adding security capabilities through a distinct processing stage that handles certainty factor modification and redistribution.
Solution Approach 2:
The patent applies preliminary action by performing certainty factor correction before the classification results are used for decision-making. The correction process is executed in advance on the model outputs, ensuring that protected data is handled securely from the outset without requiring additional processing during the actual classification decision phase.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An information processing program that causes at least one computer to execute a process, the process includes acquiring each of a plurality of certainty factors representing a possibility that classification target data belongs to a class of a plurality of classes for each of the plurality of classes by using a trained model; determining whether a maximum certainty factor having a maximum value among the plurality of certainty factors of the plurality of classes is within a certain value range; correcting a value of the maximum certainty factor to a value within the certain value range when the maximum certainty factor is not within the certain value range; and outputting the plurality of certainty factors after the correcting as a result of class classification for the classification target data.