Federated Model Updates with Consistency-Driven Training Parameters

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

In federated learning, the fixed or ascending/descending training information settings can lead to suboptimal model updates, affecting the accuracy of model training as clients may miss the optimal upload occasions.

Innovation Solution

A method for implementing model update that involves calculating first training information based on difference consistency information from multiple clients, using formulas to determine optimal training batch quantities and learning rates, and applying an exponential moving average to improve accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If fixed or ascending/descending training information settings are used, then the service device can simplify the training process, but the client may miss optimal upload occasions and model training accuracy deteriorates

Engineering Contradiction:
Improvetraining process simplicityVSAvoidmodel training accuracy
Core Design Contradiction:
Ease of operationVSMeasurement precision

Solution Approach 1:

The patent transforms the static, fixed training information settings into dynamic settings by calculating training information (number of training batches and learning rate) based on real-time difference consistency information from multiple clients. This allows the training parameters to adapt automatically to the current model state and client performance, resolving the contradiction between operational simplicity and training accuracy.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent implements a feedback mechanism where the service device receives difference information from clients, calculates difference consistency information, and uses this feedback to dynamically adjust training information for the next round. This closed-loop feedback system ensures that training parameters are continuously optimized based on actual training outcomes, improving model training accuracy while maintaining automated operation.

Inventive Principle:
Principle #23Feedback

2Measurement precision

If the service device calculates training information based on difference consistency information from multiple clients, then model training accuracy is improved, but the calculation complexity and system overhead increase

Engineering Contradiction:
Improvemodel training accuracyVSAvoidservice device calculation complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent enables the system to self-regulate by automatically calculating training information based on difference consistency information without requiring manual intervention or complex external control systems. The service device autonomously processes difference information from clients, computes consistency metrics, and generates appropriate training parameters, reducing the need for complex external management while improving training accuracy.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The patent focuses calculation efforts on computing specific key parameters (difference consistency information, number of training batches, learning rate) rather than analyzing entire model structures. By changing and optimizing only the necessary parameters based on difference information, the system achieves improved training accuracy with controlled computational overhead.

Inventive Principle:
Principle #35Parameter changes

3Adaptability or versatility

If training information is set in ascending or descending manner only, then the service device can maintain simple control logic, but the client cannot adapt to optimal upload occasions and training effectiveness decreases

Engineering Contradiction:
Improvetraining adaptation capabilityVSAvoidtraining efficiency
Core Design Contradiction:
Adaptability or versatilityVSProductivity

Solution Approach 1:

The patent performs preliminary calculation of training information (number of training batches and learning rate) based on difference consistency information before the client executes the next training round. This preliminary action ensures that clients receive optimized training parameters in advance, enabling them to adapt to optimal upload occasions and improving overall training efficiency without requiring complex real-time adjustments during training execution.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4083832B1Method and device for implementing model update
Publication Date: 2025.11.05 HUAWEI TECH CO LTD
  • EP4083832B1 patent drawingFigure 1
  • EP4083832B1 patent drawingFigure 2
  • EP4083832B1 patent drawingFigure 3

AI summary

A method for implementing model update is provided and used in a federated update framework. The method includes: A service device receives difference information uploaded by at least two clients. The service device performs calculation based on the difference information uploaded by the at least two clients, to obtain first difference consistency information, where the first difference consistency information indicates a consistency degree of the difference information uploaded by the at least two clients. The service device performs calculation based on the first difference consistency information, to obtain first training information, where the first training information is used to train a third model, and the third model is obtained by updating the first model by the service device based on the difference information uploaded by the at least two clients. The service device sends the first training information to the at least two clients. This method can improve model accuracy.