Secure Aggregation Integrity Verification via Super-Client Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing secure aggregation protocols in federated learning lack integrity protection, allowing a malicious server to tamper with the sum of clients' inputs without being detected.
Innovation Solution
Implementing a method that uses super-clients and backup clients to verify the integrity of the sum computed by the server through linearly-homomorphic commitments and random shares, ensuring that clients can check the validity of the published sum by comparing it with the results of a verification algorithm.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If single-server secure aggregation protocols are used to maintain privacy, then client input privacy is protected, but integrity protection against malicious server tampering is lost
Solution Approach 1:
The verification function is segmented and distributed to multiple super-clients. Each super-client independently verifies the aggregation result using secret-shared verification keys, eliminating the need for a trusted server while maintaining both privacy and integrity.
Solution Approach 2:
Secret-shared verification keys act as intermediaries between the server and clients. The server cannot access individual verification keys (maintaining privacy), yet clients can use these shared keys to verify aggregation integrity (providing reliability).
2Reliability
If multiple super-clients are introduced for verification, then integrity verification is enabled, but system complexity increases
Solution Approach 1:
Super-clients perform multiple functions: they act as regular clients in the aggregation protocol and simultaneously serve as verification nodes. This multi-functionality enables integrity verification without requiring separate dedicated verification infrastructure.
Solution Approach 2:
The system uses its own existing clients to perform verification tasks. Super-clients leverage their participation in the aggregation protocol to also conduct verification, eliminating the need for external or specialized verification components.
3Reliability
If commitment aggregation by super-clients is implemented, then verification capability is enhanced, but communication overhead increases
Solution Approach 1:
Super-clients aggregate commitment values from multiple clients and perform verification in a consolidated manner. This merging approach allows verification of multiple client inputs through a single aggregated verification process, reducing redundant communication.
Solution Approach 2:
Super-clients perform verification on aggregated commitments rather than individually verifying each client's commitment. This partial verification approach is sufficient to ensure overall integrity while significantly reducing communication overhead compared to full individual verification.
Data Source
AI summary
A method for secure aggregation, by a server, of client-provided inputs includes receiving, from each of a plurality of clients, a respective client input, for which a commitment is published. The commitments were computed using randomness and are aggregated by at least two super-clients and a sum of the aggregated commitments is published by each super-client. A sum of the received client inputs is published such that validity of the sum is checkable, by the clients, by comparing the sum of the received client inputs to a verification algorithm result that uses a sum of additive shares computed by the clients using the randomness, and by verifying that the published sum of the aggregated commitments is the same for each super-client. The method can be applied to use cases, for example, in digital medicine using medical data or smartcity applications to support decision-making.