Server Apparatus for Federated Learning Model Parameter Similarity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In personalized federated learning, the high risk of information leakage occurs when clients share local models, making it difficult to obtain a model optimized for each client while minimizing data exposure.
Innovation Solution
A server apparatus and method that receive local model parameters from multiple client apparatuses performing neural network federated learning with multiplex branches, calculate the similarity between these parameters, and use the results to determine a global model parameter, which is then transmitted back to the clients, allowing each client to learn a local model appropriate for its data without sharing models directly.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If clients share local models in personalized federated learning, then each client can obtain a model optimized for its data, but the risk of information leakage increases
Solution Approach 1:
The patent introduces a server as an intermediary that receives local model parameters from multiple clients, calculates similarity between these parameters, and generates a global model parameter that is then distributed back to clients. This intermediary mechanism enables personalized learning without direct client-to-client model sharing, thus reducing information leakage risk while maintaining model optimization capability.
Solution Approach 2:
Instead of sharing actual local models between clients, the system uses local model parameters as copies or representations. The server calculates similarity based on these parameter copies and generates a global model parameter that captures essential information without exposing sensitive local model details, thereby enabling personalized learning while protecting information privacy.
2Adaptability or versatility
If local model parameters are shared among all clients, then personalized learning can be achieved, but the complexity of the system increases
Solution Approach 1:
The patent extracts only the necessary local model parameters from each client's local model, rather than sharing the entire local model structure and all parameters. The server then uses these extracted parameters to calculate similarity and generate a global model parameter. This extraction approach reduces system complexity by minimizing the amount of data that needs to be transmitted and processed while still enabling personalized learning.
3Stability of the object's composition
If a global model is calculated based on all local model parameters, then learning stability improves, but the computational cost increases
Solution Approach 1:
The patent changes the parameter representation from full local model parameters to similarity-based aggregated parameters. Instead of directly computing from all local model parameters, the system calculates similarity between parameters and uses this similarity metric to generate the global model parameter. This parameter transformation reduces computational energy consumption while maintaining learning stability by preserving the essential relationships between different clients' models.
Data Source
AI summary
A server apparatus has: a receiving unit that receives, from a plurality of client apparatuses that perform federated learning of a neural network model having multiplex branches capable of performing different operations on a common input and learn a local model parameter of each of the multiplex branches and a weight for each branch used in superposing outputs from the respective multiplex branches, the local model parameters corresponding to each of the branches; a similarity degree calculating unit that calculates a degree of similarity between the local model parameters corresponding to each of the branches, received from different client apparatuses; a parameter calculating unit that calculates a parameter of a global model based on the local model parameter selected based on a result of calculation by the similarity degree calculating unit; and a parameter transmitting unit that transmits the parameter calculated by the parameter calculating unit to the client apparatus.