Federated Data Aggregation via Encoded Node Intermediation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In federated computing, data privacy is compromised due to the risk of inference of private information from gradient values transmitted by client devices, and collusion among intermediate network nodes can breach privacy, especially when raw data is not encrypted and intermediate nodes aggregate individual data.
Innovation Solution
Client devices encode data using either information-theoretic (IT) privacy by adding random information or computational privacy through encryption, and transmit multiple instances to multiple network nodes to mitigate collusion risks, while network nodes aggregate encoded data to prevent inference, balancing privacy and communication resource usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If client devices transmit gradient values to network servers for federated computing, then computing tasks can be distributed and processed, but data privacy is compromised due to the risk of inference of private information from gradient values
Solution Approach 1:
The patent introduces intermediate network nodes as mediators between client devices and the network server. These nodes aggregate encoded gradient values before forwarding to the server, preventing direct exposure of private data while enabling federated computing. The intermediary layer adds a protective buffer that maintains privacy while preserving computational functionality.
Solution Approach 2:
The patent employs encoding schemes where client devices create multiple encoded copies of gradient values rather than transmitting original data. These encoded copies contain the necessary information for computing but obscure the underlying private data, allowing the server to perform computations on copies without accessing sensitive information.
2Loss of energy
If intermediate network nodes aggregate individual data from multiple client devices, then communication efficiency is improved, but privacy is breached when nodes collude
Solution Approach 1:
The patent transmits multiple encoded copies of data to different intermediate nodes rather than sending original data to a single aggregator. This approach allows efficient aggregation of encoded values while preventing any single node or colluding subset from reconstructing private information, as each node only possesses encoded copies.
Solution Approach 2:
The patent segments the data transmission process by distributing encoded gradient values across multiple intermediate nodes rather than consolidating all data at one aggregation point. This segmentation ensures that even if some nodes collude, they lack the complete information needed to breach privacy, while still enabling efficient parallel processing.
3Object-affected harmful factors
If client devices encode data using information-theoretic privacy by adding random information, then privacy protection is enhanced, but communication data volume increases
Solution Approach 1:
The patent creates multiple encoded copies of gradient values with different random information added to each copy. While this increases the total data volume, the encoding is designed so that the server can efficiently aggregate these copies to extract the necessary computational information while the random additions cancel out or become irrelevant in the aggregation process.
Solution Approach 2:
The patent designs the encoding scheme to serve multiple functions simultaneously: protecting privacy through random information addition, enabling efficient aggregation by maintaining mathematical properties for combination, and allowing the server to recover useful information without knowing the random components. This multi-functionality justifies the increased data volume.
Data Source
AI summary
A method performed by a client device is provided. The method includes obtaining, by a client device, data for transmission to a network server. The method includes determining, by the client device, a plurality of intermediate network nodes for transmission of the data, wherein the plurality of the intermediate network nodes are communicatively coupled to the network server. The method includes encoding, by the client device, the data using an encoding scheme. The method includes transmitting, by the client device and according to a data transmission scheme, a plurality of instances of the encoded data to the plurality of intermediate network nodes for aggregation by the intermediate network nodes with data received from one or more other client devices. Also provided are methods performed by an intermediate network node and a network server. The methods can be performed by one or more processors of an apparatus.