Federated Learning Parameter Share Registration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing method of managing parameters of local models in a local model management table is inefficient, leading to significant time consumption when computing a global model in federated learning.
Innovation Solution
A secure global model computation device is introduced, which includes a transmission/reception unit and a parameter share registration unit. The parameter share registration unit registers shares of parameters of local models in a local model management table using K records, allowing for efficient registration without recreating the table.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If parameters of local models are registered in a local model management table using the existing method, then the table can store and manage model parameters, but the table must be recreated each time local models are registered, leading to significant time consumption
Solution Approach 1:
The patent segments the local model parameters into shares distributed across multiple secure computation devices. Instead of managing complete parameters in a single table that requires recreation, the system divides parameters into N shares and stores them in a distributed manner, allowing incremental updates without full table recreation.
Solution Approach 2:
The patent performs preliminary setup by establishing a local model management table structure that can accommodate parameter shares before actual model training begins. This preliminary structure includes predefined slots for storing shared parameters, enabling efficient registration during the federated learning process without requiring table recreation.
2Reliability
If secure computation is used to compute global models, then training data security is improved, but computational overhead increases due to the need for multiple secure computation devices and protocols
Solution Approach 1:
The patent makes each secure computation device multi-functional by enabling them to perform both local model training and global model computation tasks. The same devices that train local models also participate in secure aggregation and global model computation, eliminating the need for separate dedicated secure computation infrastructure.
Solution Approach 2:
The system enables self-service by allowing local model training devices to autonomously generate parameter shares and register them in the local model management table without requiring external secure computation devices. The devices also perform self-verification of cryptographic operations, reducing the burden on separate verification systems.
3Reliability
If a local model management table is recreated each time local models are registered, then data integrity is maintained, but computational efficiency deteriorates due to repeated table creation operations
Solution Approach 1:
The patent uses copying by creating and maintaining a local model management table structure that can be efficiently updated through incremental operations. Instead of recreating the entire table, the system copies and updates only the necessary parameter share entries, preserving data integrity while improving computational efficiency.
Solution Approach 2:
The patent changes the parameter representation from complete model parameters to shared parameters. This parameter transformation allows the system to work with smaller, modular parameter shares that can be efficiently registered and updated in the management table without requiring full table recreation, thus maintaining integrity while improving speed.
Data Source
AI summary
A technique for efficiently registering local models in a local model management table used when a global model is computed from local models in federated learning is provided. A secure global model computation device in a federated learning system including M local model training devices for training local models using training data and a secure global model computation system composed of N secure global model computation devices for secure computation of a global model from M local models includes a parameter share registration unit that receives shares of parameters of an m-th local model trained by an m-th local model training device (where m satisfies 1≤m≤M)) as an input and register the shares of the parameters of the m-th local model in a local model management table using K records having a set (m, k) of identifiers and shares of parameters of a k-th layer (1≤k≤K) of the m-th local model as one record.