Privacy Auditing Network Function for Differential Privacy Verification

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing communication networks face challenges in ensuring that user equipment privacy is maintained and verified, particularly in scenarios where differential privacy mechanisms may not be adequately applied during machine learning model training, leading to potential privacy breaches.

Innovation Solution

A Privacy Auditing Network Function (PANF) is introduced to verify whether differential privacy mechanisms are applied correctly by comparing loss function values between training and test data, using shuffling techniques to obscure user identities and determining differences in loss values to set thresholds for privacy compliance.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If machine learning models are trained on user data without differential privacy mechanisms, then model training accuracy is improved, but user privacy protection deteriorates

Engineering Contradiction:
Improvemodel training accuracyVSAvoiduser privacy breach
Core Design Contradiction:
Measurement precisionVSObject-affected harmful factors

Solution Approach 1:

The patent applies differential privacy mechanisms to the training dataset before the machine learning model training process begins. This preliminary action ensures that privacy protection is embedded in the data itself, allowing the model to be trained on privacy-preserving data while maintaining acceptable training accuracy through proper noise calibration.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent introduces a privacy auditing network function as an intermediary that verifies whether differential privacy mechanisms are correctly applied during model training. This intermediary component monitors and validates the privacy protection process without directly interfering with the training accuracy, ensuring both privacy compliance and model performance.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Object-affected harmful factors

If differential privacy mechanisms are applied to training data, then user privacy protection is improved, but model training accuracy deteriorates

Engineering Contradiction:
Improveuser privacy protectionVSAvoidmodel training accuracy
Core Design Contradiction:
Object-affected harmful factorsVSMeasurement precision

Solution Approach 1:

The patent dynamically adjusts the noise level parameter in the differential privacy mechanism based on the sensitivity of the data and the required privacy budget. By optimizing this parameter, the system achieves the right balance between privacy protection strength and model training accuracy, preventing excessive degradation of performance.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent implements a feedback mechanism where the privacy auditing network function evaluates the impact of differential privacy on model training accuracy and adjusts the privacy parameters accordingly. This closed-loop approach ensures that privacy protection is maintained while minimizing negative effects on training performance.

Inventive Principle:
Principle #23Feedback

3Reliability

If privacy auditing mechanisms are implemented to verify differential privacy application, then privacy compliance is improved, but system complexity deteriorates

Engineering Contradiction:
Improveprivacy complianceVSAvoidsystem complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent implements self-service mechanisms where the system automatically verifies its own compliance with differential privacy requirements through the privacy auditing network function. This self-verification approach eliminates the need for external complex auditing infrastructure, reducing overall system complexity while maintaining high privacy compliance reliability.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The privacy auditing network function is designed to perform multiple functions: verifying differential privacy application, monitoring model training progress, and adjusting privacy parameters. This multi-functionality consolidates what could be separate complex systems into a single versatile component, reducing overall system complexity.

Inventive Principle:
Principle #6Universality (Multi-functionality)

4Measurement precision

If loss function comparison is used to detect differential privacy application, then detection accuracy is improved, but computational overhead deteriorates

Engineering Contradiction:
Improvedetection accuracyVSAvoidcomputational overhead
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent extracts only the essential loss function values at specific training stages for comparison, rather than analyzing the entire training process continuously. This extraction approach maintains high detection accuracy by focusing on critical moments while significantly reducing computational overhead by avoiding unnecessary calculations during intermediate steps.

Inventive Principle:
Principle #2Taking out (Extraction)

Data Source

PatentEP4557790B1Method, apparatus and computer program
Publication Date: 2026.04.08 NOKIA SOLUTIONS & NETWORKS OY
  • EP4557790B1 patent drawingFigure 1
  • EP4557790B1 patent drawingFigure 2
  • EP4557790B1 patent drawingFigure 3

AI summary

An apparatus comprising: means for storing (500) a dataset comprising first data and second data for at least one User Equipment, wherein the first data has been provided to a network function and the second data has not been provided to the network function, and wherein the network function trains a machine learning model based on the first data; means for storing (502) ground truth labels for the first data and the second data; means for sending (504) information based on the dataset to the network function; means for receiving (506), from the network function, information for determining a first value of a loss function, the first value of the loss function indicating a difference between the ground truth labels of the first data and at least one predicted ground truth label of the first data determined by the network function using the machine learning model; means for receiving (508), from the network function, information for determining a second value of the loss function, the second value of the loss function indicating a difference between the ground truth labels of the second data and at least one predicted ground truth label of the second data determined by the network function using the machine learning model; means for determining (510) a difference between the first value of the loss function and the second value of the loss function; means for comparing (512) the difference to a threshold to determine whether a Differential Privacy mechanism was applied to the first data before the network function trained the machine learning model based on the first data.