Federated Model Personalization for Low-IID Device Operation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for training machine learning models in distributed systems with low IIDness and limited bandwidth face challenges in maintaining model accuracy and efficiency, particularly when data distributions are not identical and independent.

Innovation Solution

A method involving client-server architecture that calculates similarity between client data sets, groups clients, determines a personalization factor based on average similarity, and updates a global model using dynamic personalization to adapt to varying data distributions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of energy

If sequential federated learning is used to train models in distributed systems with limited bandwidth, then network bandwidth consumption is reduced, but model training time and convergence speed increase significantly

Engineering Contradiction:
Improvenetwork bandwidth consumptionVSAvoidmodel training time
Core Design Contradiction:
Loss of energyVSLoss of time

Solution Approach 1:

The system dynamically adjusts the training sequence of clients based on their data characteristics and similarity to the global model. Instead of fixed sequential processing, clients are processed in optimized sequences that adapt to changing model states and data distributions, reducing both bandwidth usage and training time by avoiding redundant communications with already-well-represented clients.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The system changes the parameter of client processing from static sequential order to dynamic ordering based on data similarity metrics and convergence contributions. By calculating similarity between client data distributions and adjusting the processing sequence accordingly, the system optimizes the trade-off between bandwidth consumption and training efficiency.

Inventive Principle:
Principle #35Parameter changes

2Reliability

If distributed AI training is used to avoid centralizing data, then data privacy and security are improved, but model training efficiency and convergence decrease due to low IIDness

Engineering Contradiction:
Improvedata privacyVSAvoidmodel training efficiency
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The system applies local quality by customizing the training process for each client based on their specific data characteristics. Clients with data highly similar to the global model receive different processing treatment than those with divergent data distributions, allowing efficient convergence while maintaining data privacy through localized adaptation rather than uniform sequential processing.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The system changes parameters such as learning rates, batch sizes, and processing priorities based on each client's data similarity to the global model. This adaptive parameter adjustment maintains data privacy in distributed settings while significantly improving training efficiency by focusing computational resources on clients that contribute most to model convergence.

Inventive Principle:
Principle #35Parameter changes

3Measurement precision

If personalized model updates are applied to each client based on data similarity, then model accuracy for specific clients improves, but system complexity and computational overhead increase

Engineering Contradiction:
Improvemodel accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system implements local quality by providing personalized model updates tailored to each client's data characteristics. Clients receive customized update parameters based on their data similarity to the global model and other clients, improving accuracy for local data distributions while maintaining a manageable system architecture through systematic similarity-based grouping rather than fully custom processing for each client.

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP4700660A1Method and system for operating a technical device
Publication Date: 2026.02.25 SIEMENS AG
  • EP4700660A1 patent drawingFigure 1
  • EP4700660A1 patent drawingFigure 2
  • EP4700660A1 patent drawingFigure 3

AI summary

A computer-implemented method for operating a technical device through a client-server system with a server and clients, each connected to their respective technical devices, and performing the following steps: a. Providing operational data from the operation of the respective client's technical device, and generating operational meta-information about the operation of the technical device from the operational data using a provided, trained feature extraction model, by the server; b. Providing a trained base model to the server, and applying the operational data and meta-information of the respective client to the base model and obtaining the respective operational features, by the server; c. Calculating the similarity between the operational features of the respective clients in the form of a similarity value; d.Assigning clients to a respective client group from a set of client groups, where each client group is defined by a predefined range of similarity values ​​for that client group; e. Determining the largest client group from a set of client groups, which contains the most clients; f. Calculating the average similarity value from the respective similarity values ​​of the clients in the largest client group; g. Setting a personalization factor for each client based on the average similarity value, which expresses the similarity of the respective client to the average similarity value of the respective group; h. Generating global model parameters for each client based on the respective personalization factor and the difference between the respective operational data and the global model parameters; i.Update the base model using the global model parameters, j. Continue at step c) until a predetermined criterion for repeating steps c) - i) is met, otherwise continue with step k), k. Operate the technical device with the updated base model.