Federated Learning Node Grouping for Lower Network Overhead

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Federated Learning (FL) systems face challenges in managing network resources efficiently, particularly in distributed networks with heterogeneous nodes, leading to increased communication costs, longer convergence times, and network overhead due to unoptimized data exchange and lack of diversity in model training.

Innovation Solution

A system and method for training FL models that utilize network data by determining groups of network nodes based on topological position and statistical properties, appointing a group leader, and optimizing data exchange to reduce network overhead and communication costs.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If worker nodes are located far apart in a distributed network, then the system can handle larger scale and more diverse data, but network resource consumption increases due to multiple links being traversed

Engineering Contradiction:

Improvedistributed network capabilityVSAvoidnetwork resource consumption

Core Design Contradiction:

Adaptability or versatilityVSLoss of energy

Solution Approach 1:

The FL system is segmented into multiple hierarchical levels with local aggregators managing subsets of worker nodes. This segmentation reduces the communication distance and number of links traversed by model updates, thereby decreasing network resource consumption while maintaining the ability to handle large-scale distributed networks.

Inventive Principle:

Principle #1Segmentation

2Adaptability or versatility

If a large federation of worker nodes is used, then more diverse data can be utilized for training, but convergence time increases and network overhead from weight updates increases

Engineering Contradiction:

Improvedata diversityVSAvoidconvergence time

Core Design Contradiction:

Adaptability or versatilityVSLoss of time

Solution Approach 1:

The large federation is segmented into multiple smaller groups managed by local aggregators. Each aggregator trains models on its subset of worker nodes, enabling parallel training processes that reduce overall convergence time while maintaining data diversity across the entire federation.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The system introduces a hierarchical dimension to the FL architecture, with local aggregators operating at one level and a central server at another. This dimensional change allows simultaneous model training at multiple levels, reducing the time required for convergence across the entire federation.

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

3Quantity of substance

If a large federation of worker nodes is used, then more data can be processed, but establishing trust among a large group of nodes becomes more difficult

Engineering Contradiction:

Improvenumber of worker nodesVSAvoidtrust establishment

Core Design Contradiction:

Quantity of substanceVSReliability

Solution Approach 1:

The large federation is segmented into smaller groups, each managed by a local aggregator. Trust is established at the local level within each smaller group, making it more manageable and reliable. The hierarchical structure allows trust to be propagated from local to global level.

Inventive Principle:

Principle #1Segmentation

4Device complexity

If traditional FL architecture with single central server is used, then implementation is simple, but network overhead and communication costs increase for distributed nodes

Engineering Contradiction:

Improvesystem architectureVSAvoidcommunication cost

Core Design Contradiction:

Device complexityVSLoss of energy

Solution Approach 1:

The centralized FL architecture is segmented into multiple local aggregators distributed across the network. Each aggregator serves a local subset of worker nodes, reducing the communication distance and data transmission requirements. This segmentation maintains relatively simple implementation while significantly reducing communication costs for distributed nodes.

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS20250350534A1A System and Method for Training a Federated Learning Model Using Network Data

Publication Date: 2025.11.13 TELEFONAKTIEBOLAGET LM ERICSSON (PUBL)

AI summary

A system (200), a first network node (240), a method, a computer program and a computer program product for training of a Federated Learning. FL, model is disclosed. The system comprises network nodes. One of the network nodes is a first network node. Each network node has access to a part of the network data. The system obtains network information and determines groups of network nodes and assigns each network node to one of the determined groups based on the network information, each determined group of network nodes comprising at least two network nodes. For each of the groups, the system appoints a second network node from among the at least two network nodes, informs the at least two network nodes about the appointed second network node and trains an FL model using the parts of the network data accessible by the at least two network nodes.