Federated Learning Transmission Modes for Reliable Differential Updates

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing federated learning systems face inefficiencies in communication between parameter servers and agent entities due to high resource consumption and the risk of agents missing differential updates, leading to incomplete model training.

Innovation Solution

Implementing alternating transmission modes for parameter vectors, where full models are sent in one mode and differential updates in another, with adaptive power and coding schemes based on channel conditions and agent status, enabling revert-back mechanisms.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of energy

If differential updates are sent to reduce communication overhead, then time/space/frequency resources are saved, but agents may miss the reception and fail to participate in subsequent iteration rounds

Engineering Contradiction:

Improvecommunication resource consumptionVSAvoidagent participation reliability

Core Design Contradiction:

Loss of energyVSReliability

Solution Approach 1:

The patent segments the parameter vector transmission into two distinct modes: full model transmission and differential update transmission. This segmentation allows the system to choose the appropriate transmission mode based on network conditions and agent status, thereby reducing communication overhead when differentials are used while maintaining reliability through full transmissions when needed.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent implements dynamic transmission mode selection where the system adapts between full model transmission and differential update transmission based on real-time conditions. The server entity dynamically determines which mode to use for each iteration round, optimizing the balance between communication efficiency and reliability.

Inventive Principle:

Principle #15Dynamics

2Reliability

If full parameter vectors are transmitted to ensure complete model updates, then all agents receive necessary updates reliably, but time/space/frequency resources are consumed excessively

Engineering Contradiction:

Improvemodel update completenessVSAvoidcommunication resource consumption

Core Design Contradiction:

ReliabilityVSLoss of energy

Solution Approach 1:

The patent divides the transmission process into two segmented modes: full model transmission for reliability-critical situations and differential update transmission for efficiency-critical situations. This segmentation enables the system to transmit only necessary information when conditions permit, reducing overall resource consumption while maintaining completeness when needed.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent changes the transmission parameter (complete model vs. differential update) based on system conditions. By adjusting this parameter dynamically, the system optimizes the trade-off between communication resource consumption and model update completeness, transmitting full models only when necessary and differentials when conditions are favorable.

Inventive Principle:

Principle #35Parameter changes

3Device complexity

If differential updates are used for parameter vectors, then communication overhead is reduced, but the complexity of tracking which agents received which updates increases

Engineering Contradiction:

Improvecommunication protocol simplicityVSAvoiditerative learning efficiency

Core Design Contradiction:

Device complexityVSProductivity

Solution Approach 1:

The patent incorporates feedback mechanisms where agents report their received model version to the server entity. This feedback loop enables the server to track which agents have received which updates, managing the complexity of differential transmission and ensuring all agents are synchronized without excessive overhead.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The patent performs preliminary actions by having agents store previous model versions before differential updates are applied. This preliminary storage enables agents to reconstruct complete models from differentials without requiring complex real-time tracking, simplifying the protocol while maintaining efficiency.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS20250363411A1Iterative Learning with Different Transmission Modes

Publication Date: 2025.11.27 TELEFONAKTIEBOLAGET LM ERICSSON (PUBL)

AI summary

There is provided techniques for an iterative learning process being performed between a server entity and agent entities. The iterative learning process pertains to a computational task to be performed by the agent entities. For each iteration round of the iterative learning process the server entity uses either a first transmission mode or a second transmission mode for sending a respective parameter vector of the computational task towards the agent entities. The agent entities are configured with the computational task, with which reception mode to use, and for the agent entities to, as part of performing one iteration round of the iterative learning process with the server entity, send local updates of computational results of the computational task to the server entity. The parameter vector of each iteration round is sent using either the first transmission mode or the second transmission mode.