Classification Model Replica Evaluation for Inversion Attack Risk

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

The susceptibility of machine-learned classification models to model inversion attacks is not well understood, posing risks to the security of data processing systems, as attackers can create replicas or infer decision boundaries, potentially leading to exploits.

Innovation Solution

A method to determine the susceptibility of classification models to model inversion attacks by training a replica using sample data, evaluating its properties, and providing a measure based on the number of samples and the replica's performance.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a complex evaluation framework is used to assess multiple aspects of model performance, then the comprehensiveness of evaluation is improved, but the complexity of the evaluation process increases

Engineering Contradiction:
Improveevaluation comprehensivenessVSAvoidevaluation process complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The evaluation framework is segmented into distinct evaluation modules, each assessing a specific aspect of model performance (e.g., accuracy, precision, recall, F1-score). This segmentation allows comprehensive evaluation while maintaining organizational clarity and reducing overall process complexity through modular design.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

A structured evaluation framework acts as an intermediary between the model output and the final assessment. This framework organizes multiple evaluation metrics and their relationships, mediating the complexity of assessing multiple performance aspects simultaneously.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Measurement precision

If multiple evaluation metrics are considered to provide a comprehensive assessment, then the accuracy of model evaluation is improved, but the time required for evaluation increases

Engineering Contradiction:
Improveevaluation accuracyVSAvoidevaluation time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The evaluation framework is designed to perform preliminary organization and structuring of evaluation metrics before actual evaluation execution. By pre-defining the evaluation architecture and metric relationships, the actual evaluation process becomes more efficient and less time-consuming while maintaining comprehensive accuracy.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The framework enables selective application of evaluation metrics based on specific needs. Rather than always evaluating all possible metrics, the system can apply partial evaluation to the most relevant metrics, reducing evaluation time while maintaining sufficient accuracy for the given context.

Inventive Principle:
Principle #16Partial or excessive action

3Loss of information

If detailed analysis of evaluation results is performed to understand model performance deeply, then the insight quality is improved, but the computational resources required increase

Engineering Contradiction:
Improveinsight qualityVSAvoidcomputational resources
Core Design Contradiction:
Loss of informationVSUse of energy by moving object

Solution Approach 1:

The evaluation framework extracts and highlights the most critical insights from comprehensive model performance analysis. By identifying and presenting only the essential evaluation results and key performance indicators, the system maintains high insight quality while reducing the computational resources required for detailed analysis.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The framework creates a structured representation or copy of the evaluation results that captures essential information in an optimized format. This structured copy enables detailed analysis and insight generation with reduced computational overhead compared to processing the full raw evaluation data.

Inventive Principle:
Principle #26Copying

Data Source

PatentEP4315134B1Classification model evaluation
Publication Date: 2026.05.20 BRITISH TELECOM PLC
  • EP4315134B1 patent drawingFigure 1
  • EP4315134B1 patent drawingFigure 2

AI summary

A computer implemented method, computer system and computer program are provided for determining a measure of susceptibility of a machine-learned classification model to model inversion attacks. The method obtains a set of sample data from the classification model, the set of sample data comprising a plurality of samples, each sample comprising a respective classification produced by the classification model for an associated set of input values. The method trains a replica of the classification model using the sample data. The method obtains a set of further sample data from the classification model, the set of further sample data comprising a plurality of further samples, each further sample comprising a respective classification produced by the classification model for an associated set of input values. The method evaluates a property of the replica of the classification model using the further sample data. The method provides a measure of the level of susceptibility of the classification model to model inversion attacks based on the number of samples in the sample data used to train the replica of the classification model and the property of the replica of the classification model.