Edge Node Differential Privacy Model Ensemble
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional federated learning methods in distributed systems are restricted by cryptography-based security model fusion policies, limiting the manner of model ensemble and affecting the model ensemble effect.
Innovation Solution
Implementing a data processing method where edge node devices train sub-models through differential privacy and transmit model training information in plaintext to a central node device, which performs model ensemble using a policy other than cryptography-based security, allowing for expanded ensemble methods while ensuring data security.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If cryptography-based security model fusion policy is used for model ensemble, then data security is ensured, but the manner of model ensemble is limited
Solution Approach 1:
The patent segments the model ensemble process into two distinct phases: a training phase where encrypted model gradients are aggregated using secure aggregation protocols, and an inference phase where the aggregated model performs predictions. This segmentation allows different security mechanisms to be applied at different stages, enabling both security and flexibility.
Solution Approach 2:
The patent introduces dynamic model ensemble policies that can switch between different aggregation strategies based on the operational context. During training, secure aggregation is used; during inference, more flexible ensemble methods can be applied. This dynamic approach resolves the contradiction by adapting the security level to the specific phase and requirements.
2Reliability
If federated average algorithm is used for model fusion, then data security is maintained, but model ensemble effect is affected
Solution Approach 1:
The patent changes the parameters and objectives of the aggregation function at different stages. During training, the aggregation function optimizes for security and privacy preservation. During inference, the same or a modified aggregation function can optimize for prediction accuracy and model performance. This parameter change allows both security and effectiveness to be achieved in their respective contexts.
Data Source
AI summary
A method, an apparatus, a computer device, a storage medium, and a program product for processing data are provided, which belong to the technical field of artificial intelligence. The method includes: acquiring model training information transmitted by each of at least two edge node devices, the model training information being transmitted in a form of plaintext, and being obtained by the edge node device by training sub-models through differential privacy; acquiring, based on the model training information transmitted by each of the at least two edge node devices, the sub-models trained by each of the at least two edge node devices; and performing, based on a target model ensemble policy, model ensemble on the sub-models trained by the at least two edge node devices, to obtain a global model. This solution expands the manner of model ensemble while ensuring the data security, thereby improving the model ensemble effect.