Secure Aggregation via Super-Client Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing secure aggregation protocols in federated learning and smart metering applications face high communication overhead and require most clients to be online, making them unsuitable for applications with mobile or erratic client behavior.
Innovation Solution
The use of a random beacon service to select a subset of super-clients that aid the server in aggregating inputs, employing secret-sharing and homomorphic encryption to maintain privacy and reduce communication overhead.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If secure aggregation protocols are used to protect client input privacy, then privacy protection is improved, but communication overhead increases
Solution Approach 1:
The patent segments the client population into neighborhoods and designates specific super-clients within each neighborhood to handle aggregation tasks. This segmentation allows the system to distribute the communication burden across multiple super-clients rather than requiring all clients to participate in every aggregation operation, thereby reducing overall communication overhead while maintaining privacy protection through the structured use of secret-sharing and masking protocols.
Solution Approach 2:
The patent introduces super-clients as intermediary nodes that facilitate the aggregation process. These super-clients act as mediators between regular clients and the server, collecting masked inputs from clients in their neighborhood and coordinating the unmasking process. This intermediary structure reduces direct communication between all clients and the server, optimizing communication efficiency while preserving the security guarantees of the aggregation protocol.
2Measurement precision
If all clients must be online for secure aggregation, then aggregation accuracy is improved, but system adaptability to mobile/erratic client behavior deteriorates
Solution Approach 1:
The patent performs preliminary actions by pre-assigning super-clients to neighborhoods and pre-distributing necessary cryptographic materials (such as secret shares and masking keys) before the actual aggregation operation. This preliminary setup allows the aggregation process to proceed with only the necessary super-clients and their associated clients online, rather than requiring all clients to be present, thus accommodating mobile and erratic client behavior while maintaining aggregation functionality.
Solution Approach 2:
The patent implements partial action by allowing aggregation to proceed with only a subset of clients (specifically, those associated with active super-clients) rather than requiring full client participation. The super-clients can complete the aggregation process using the masked inputs received from their assigned clients, and the server can compute the final result without needing inputs from all clients in the system, making the protocol adaptable to scenarios where clients come and go.
3Reliability
If secret-sharing and masking protocols are implemented, then privacy protection is improved, but computational complexity increases
Solution Approach 1:
The patent segments the computational burden of secret-sharing and masking operations across multiple super-clients rather than concentrating it in a single node or requiring all clients to perform complex operations simultaneously. Each super-client handles the cryptographic operations for its assigned neighborhood, dividing the overall computational complexity into manageable pieces that can be executed in parallel, thereby reducing the peak computational load on any single component of the system.
Data Source
AI summary
A method for secure aggregation using public randomness is provided. The method includes determining, based on the public randomness from a random service beacon, a set of super-clients from a plurality of clients. The method further includes obtaining masked inputs from the clients and aggregated random values based on the set of super-clients. Each of the masked inputs is associated with a client from the plurality of clients and each of the aggregated random values is associated with a super-client from the set of super-clients. The method also includes aggregating the masked inputs from the plurality of clients and the aggregated random values from the set of super-clients.