Federated Learning Control Device Asynchronous Synchronization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional federated learning systems face safety and efficiency issues due to the transmission of plaintext worker models, allowing the federated learning device to infer the learning data tendencies, and inefficient processing speed considerations.
Innovation Solution
A control device manages a federated learning system with secure federated learning devices, where model learning devices perform local processing and provide confidential information for secure aggregation, and the control device adjusts processing control to synchronize or asynchronous local processing based on processing times.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If plaintext worker models are transmitted in conventional federated learning, then the federated learning device can aggregate models efficiently, but data privacy is compromised as the device can infer learning data tendencies
Solution Approach 1:
The patent introduces an intermediary encryption mechanism where worker models are encrypted before transmission to the federated learning device. The encryption acts as a mediator that prevents the device from directly accessing or inferring information from the raw worker models, thereby protecting data privacy while still enabling model aggregation through decryption and processing
Solution Approach 2:
The patent transforms the worker models by applying encryption parameters before transmission. This parameter change converts the plaintext models into ciphertext forms, fundamentally altering their state to prevent inference of learning data tendencies while maintaining the ability to aggregate them through proper decryption and processing procedures
2Stability of the object's composition
If synchronous control is used in federated learning, then all model learning devices are coordinated, but processing efficiency decreases when devices have different speeds
Solution Approach 1:
The patent implements a dynamic control mechanism that adapts the synchronization strategy based on the processing speeds of different model learning devices. Instead of rigid synchronous control, the system dynamically adjusts timing and coordination parameters to accommodate varying device performance, thereby maintaining stability while improving overall processing efficiency
Solution Approach 2:
The patent allows faster devices to proceed with partial aggregation operations while slower devices are still processing, rather than waiting for all devices to complete. This partial action approach prevents faster devices from idle waiting, improving productivity while maintaining eventual synchronization through coordinated finalization of aggregation results
3Productivity
If asynchronous control is implemented to improve processing efficiency, then devices can operate at their own speeds, but coordination and synchronization become more complex
Solution Approach 1:
The patent incorporates feedback mechanisms where model learning devices report their processing status and timing information to the federated learning device. This feedback enables the system to track asynchronous operations, coordinate aggregation timing, and maintain synchronization without requiring complex centralized control, thereby managing coordination complexity through information feedback
Data Source
AI summary
A federated learning system including a model learning device and a secure federated learning device is controlled. The model learning device updates an aggregate model by machine learning using local learning data to obtain information for identifying a worker model and executes local processing for providing confidential information of the information for identifying the worker model to the secure federated learning device. The secure federated learning device obtains confidential information of information for identifying a new aggregate model obtained by aggregating a plurality of worker models through secure computation using the obtained confidential information of the information for identifying the worker model without obtaining the worker models, and executes secure aggregation processing for providing the information for identifying the new aggregate model or the confidential information of the information for identifying the new aggregate model to a plurality of model learning devices. In this case, asynchronous control for causing local processing of the plurality of model learning devices to be executed asynchronously is performed in a case in which a local processing time corresponding to the local processing is longer than an aggregation processing time corresponding to the secure aggregation processing, and synchronous control for synchronizing the local processing of the plurality of model learning devices is performed in a case in which the asynchronous control is not performed.