Distributed ML Model Generation for Privacy and Low Network Load

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Centralized learning architectures in 5G networks face challenges such as data privacy breaches, latency issues, high network load, and high costs due to large data transfers, while decentralized architectures suffer from reduced accuracy and biased statistics, making existing machine learning model generation techniques inefficient and inaccurate.

Innovation Solution

A method for generating a machine learning model by training models locally at different network nodes within an NFV architecture, sharing model parameters rather than raw data, to enhance security, reduce latency, and improve accuracy by leveraging insights from multiple nodes.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If centralized learning architecture is used to train machine learning models, then model accuracy can be improved through comprehensive data analysis, but data privacy security deteriorates due to centralized data collection

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

Solution Approach 1:

The patent segments the centralized model training process into distributed training at multiple network nodes. Each node trains local models on its own data without sharing raw data, thus maintaining data privacy while still enabling distributed learning. The segmentation of training responsibilities across nodes resolves the contradiction between accuracy and privacy.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces model parameters as an intermediary between data sources and the central entity. Instead of transferring raw data, only aggregated model parameters are shared. This intermediary mechanism enables knowledge exchange while preserving data privacy, resolving the contradiction between comprehensive data analysis and privacy protection.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Measurement precision

If centralized data transfer is performed for model training, then model accuracy can be improved, but network load increases significantly

Engineering Contradiction:
Improvemodel accuracyVSAvoidnetwork load
Core Design Contradiction:
Measurement precisionVSLoss of energy

Solution Approach 1:

The patent extracts only the essential model parameters from the training process and transfers them, rather than transferring entire datasets. This extraction of necessary information reduces network load significantly while still enabling model improvement through parameter aggregation.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent segments the training workload across multiple network nodes, with each node performing local training independently. This segmentation eliminates the need for continuous large-scale data transfers between nodes and the central entity, reducing overall network load while maintaining model accuracy through parameter aggregation.

Inventive Principle:
Principle #1Segmentation

3Measurement precision

If centralized model training is performed, then comprehensive model accuracy can be achieved, but latency increases due to data transfer time

Engineering Contradiction:
Improvemodel accuracyVSAvoidlatency
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent performs preliminary model training actions at distributed network nodes before aggregation. Each node independently trains local models using its available data, so when parameters are aggregated, the central entity receives pre-trained models rather than raw data requiring processing. This preliminary action reduces the time required for centralized processing and reduces latency.

Inventive Principle:
Principle #10Preliminary action

4Object-affected harmful factors

If decentralized learning architecture is used, then data privacy can be maintained, but model accuracy reduces due to biased local statistics

Engineering Contradiction:
Improvedata privacyVSAvoidmodel accuracy
Core Design Contradiction:
Object-affected harmful factorsVSMeasurement precision

Solution Approach 1:

The patent merges the results from multiple distributed local models by aggregating their parameters. This combination of locally trained models compensates for the biased statistics of individual nodes, producing a more accurate global model that maintains data privacy while improving accuracy through the synthesis of multiple perspectives.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The patent implements a feedback mechanism where the central entity aggregates parameters from multiple nodes and can redistribute improved model parameters back to the nodes. This feedback loop allows the system to iteratively improve model accuracy while maintaining the privacy benefits of decentralized training, resolving the contradiction between privacy and accuracy.

Inventive Principle:
Principle #23Feedback

Data Source

PatentEP4200735B1Generating a machine learning model
Publication Date: 2026.03.11 TELEFONAKTIEBOLAGET LM ERICSSON (PUBL)
  • EP4200735B1 patent drawingFigure 1
  • EP4200735B1 patent drawingFigure 2
  • EP4200735B1 patent drawingFigure 2

AI summary

There is provided a method for generating a machine learning model. The method is performed by an entity. In response to receiving model parameters of a plurality of machine learning models, a machine learning model is generated (102) based on the model parameters of the plurality of machine learning models. Each of the plurality of machine learning models is trained by a different network node of a network functions virtualization (NFV) architecture based on data that is local to that network node.