Secure ML Model Sharing via Granular Authorization in 5G Networks

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

In communication networks, particularly in 5G core networks, there is a need for enhanced security mechanisms to ensure the secure sharing of machine learning models, as existing methods lack authorization granularity and fail to prevent unauthorized access or revocation of permissions, leading to potential security breaches.

Innovation Solution

The implementation of a Model Training Logical Function (MTLF) that verifies requests from network function consumers to obtain machine learning models, informs an Analytical Data Repository Function (ADRF) about authorization status, and updates or revokes access rights, ensuring secure sharing by encrypting Uniform Resource Identifiers (URIs) and maintaining granular authorization records.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If existing sharing methods are used, then model sharing is simple, but security and authorization granularity are insufficient

Engineering Contradiction:
ImprovesecurityVSAvoidsystem complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent segments authorization management into granular permission records that track specific consumer-model relationships. Each authorization is broken down into individual permission entries stored in a data repository, allowing precise control over which consumers can access which models. This segmentation enables detailed security tracking without requiring complete system redesign.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces an intermediary authorization management system that sits between the model sharing mechanism and the consumers. This intermediary verifies authorization requests against stored permission records and mediates access control, providing enhanced security while keeping the underlying sharing infrastructure relatively simple.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Reliability

If authorization verification is implemented, then unauthorized access is prevented, but processing time increases

Engineering Contradiction:
Improveaccess controlVSAvoidverification time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent performs preliminary actions by pre-storing authorization permission records in the data repository before actual model access occurs. When a consumer requests a model, the system has already have the authorization criteria prepared and stored, enabling faster verification by simply checking against pre-established permissions rather than computing authorization from scratch each time.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent creates copies of authorization permission records and stores them in a centralized data repository. This allows multiple verification operations to reference the same copied permission data simultaneously, reducing the time burden on the original authorization source and enabling parallel verification processes.

Inventive Principle:
Principle #26Copying

3Measurement precision

If granular authorization records are maintained, then permission management is precise, but data storage requirements increase

Engineering Contradiction:
Improveauthorization granularityVSAvoiddata storage
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent creates a universal authorization data repository that serves multiple functions: storing permission records, verifying consumer access rights, tracking model usage, and supporting revocation operations. This multi-functional repository consolidates what could be scattered data structures into a single efficient storage system, reducing overall data redundancy while maintaining granular authorization tracking.

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

4Reliability

If encryption of URIs is implemented, then model confidentiality is enhanced, but processing overhead increases

Engineering Contradiction:
ImproveconfidentialityVSAvoidprocessing energy
Core Design Contradiction:
ReliabilityVSUse of energy by moving object

Solution Approach 1:

The patent applies encryption selectively rather than universally. URIs and sensitive model identifiers are encrypted only when transmitted to or stored for unauthorized consumers, while authorized consumers receive unencrypted or efficiently encrypted data. This local quality approach ensures confidentiality where needed while minimizing processing overhead for authorized access paths.

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP4391447A1Secure sharing of machine learning models in communication networks
Publication Date: 2024.06.26 NOKIA TECHNOLOGIES OY
  • EP4391447A1 patent drawingFigure 1
  • EP4391447A1 patent drawingFigure 2
  • EP4391447A1 patent drawingFigure 3

AI summary

According to an example aspect of the present invention, there is provided an apparatus comprising means for receiving, from a network function consumer, a request to obtain a machine learning model, verifying that the network function consumer is authorized to obtain the machine learning model and transmitting to an analytical data repository function, after said verification, a request to store information indicating that the network function consumer is authorized to obtain the machine learning model.