Federated Learning Privacy Checker Segregating Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In federated learning systems, the transmission of model parameters from client devices to a centralized server can lead to data leakage and privacy violations, as local data used for training may be inferred or reconstructed at the server.
Innovation Solution
Implementing a privacy checker on each client device to segregate data into shareable and private portions, obscuring local model parameters before transmission, and using these obscured parameters for global model updates, thereby preventing the sharing of private data and maintaining local data privacy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If model parameters are transmitted from client devices to centralized server for aggregation, then global model training is enabled, but private data may be inferred or reconstructed leading to data leakage
Solution Approach 1:
The patent introduces an intermediary mechanism (privacy checker and obscuring process) between the client device and federated server. The privacy checker segregates data into shareable and private portions, while the obscuring process transforms local model parameters into obscured parameters before transmission. This intermediary layer enables global model training while preventing direct access to private data at the server side.
Solution Approach 2:
The patent segments data into two distinct categories: shareable data and private data. The privacy checker divides the local dataset into these portions, allowing only shareable data to be used for training that will be transmitted to the server. This segmentation ensures that private data remains isolated and cannot be reconstructed from the transmitted model parameters.
2Reliability
If local model parameters are sent to federated server, then model aggregation is achieved, but confidentiality of local data is compromised
Solution Approach 1:
The patent transforms the parameters being transmitted through an obscuring process. Local model parameters are converted into obscured local model parameters by applying transformations that preserve the training utility while removing sensitive information. This parameter change ensures that the server receives reliable model updates for aggregation while the original confidential data cannot be recovered from the transmitted parameters.
Solution Approach 2:
The privacy checker extracts and removes private data from the training process before model parameters are generated. By separating private data from shareable data at the source, the system ensures that only information suitable for sharing is used in training, and thus the transmitted parameters do not contain reconstructable private information.
3Object-affected harmful factors
If privacy checker segregates data into shareable and private portions, then data privacy is enhanced, but device complexity increases
Solution Approach 1:
The privacy checker is implemented as a self-contained module on the client device that autonomously performs data segregation and parameter obscuring without requiring external assistance or complex coordination with the server. The device handles its own privacy protection needs locally, which while adding some complexity, keeps the overall system architecture simpler by distributing the privacy function rather than requiring centralized privacy management.
Data Source
AI summary
Protecting data privacy in a federated learning computing environment includes receiving a model trained by a federated server with public data using global model parameters, getting local shareable data from a local shareable database, training the model with the local shareable data using the global model parameters to generate local model parameters, and obscuring the local model parameters. Protecting data privacy includes sending the local model parameters to the federated server, modifying the model to classify private data, and training the model with the private data using the local model parameters and updating the local model parameters.