Secure Global Model Computation Device for Federated Learning Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In federated learning, the high number of devices for training local models compared to the number of devices in a secure computation system leads to increased processing time for registering local models, thereby prolonging the model training process.
Innovation Solution
Implement a secure global model computation device that selects and configures a subset of secure global model computation devices with high availability of computation resources to form a secure global model computation system, enabling efficient training of local models within the federated learning system.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a large number of devices are used for training local models in federated learning, then model training efficiency and parallel learning speed are improved, but the processing time for registering local models in the secure computation system increases, leading to overall training time extension
Solution Approach 1:
The patent segments the K secure global model computation devices into N subsets, where each subset forms an independent secure computation system. This allows parallel processing of model training across multiple subsets, reducing the registration processing time while maintaining overall training efficiency. Each subset can operate independently without blocking others.
Solution Approach 2:
The patent uses N subsets where N is less than K (the total number of devices), meaning not all devices need to participate in every secure computation round. This partial action approach reduces the registration overhead while still providing sufficient computational power for efficient model training, resolving the contradiction between productivity and time loss.
2Reliability
If secure computation protocols are implemented to prevent data leakage, then data security is improved, but computational overhead and processing time increase
Solution Approach 1:
By dividing K devices into N subsets that form separate secure computation systems, the patent reduces the computational overhead per subset. Each subset performs secure computation with fewer devices, lowering the overhead while maintaining security. The independent subsets can process models in parallel, further reducing total computational time.
Solution Approach 2:
The patent changes the parameter of secure computation system size from K devices to N subsets of size N/K devices each. This parameter change reduces the computational overhead per subset while maintaining the security benefits of secure computation protocols, resolving the contradiction between reliability and time loss.
Data Source
AI summary
A technique for efficiently training a model by providing a function of training a local model by a device constituting a secure computation system for computing a global model in a federated learning. Arbitrary N secure global model computation devices among K secure global model computation devices can constitute a secure global model computation system for performing secure computation of a global model from N local models, wherein K is an integer of 3 or more and N is an integer satisfying 3≤N≤K, and a secure global model computation device includes a selection unit that selects N−1 secure global model computation devices having a large availability of computation resources from K−1secure global model computation devices excluding the secure global model computation device itself, and a system configuration unit that configures a secure global model computation system using N secure global model computation devices obtained by combining the selected N−1 secure global model computation devices and the secure global model computation device itself.