heterogeneous quantization
In distributed settings such as federated learning, different clients compress model updates using different quantization levels (e.g., 8-bit, 4-bit) because of differences in bandwidth or compute capacity, resulting in inconsistent numerical ranges and algebraic structures.
Timeline (last 90 days)
Existing heterogeneous-quantization secure aggregation methods leak partial aggregations of model updates, and under certain dropout and client-server collusion patterns may leak fragments of individual model updates.
The paper proposes an information-theoretically secure aggregation scheme that tolerates client dropouts and collusion between up to T semi-honest clients and the server, while supporting heterogeneous client quantization.
Compared with existing heterogeneous-quantization secure aggregation methods, this scheme maintains information-theoretic security under all permitted dropout and collusion patterns.
The scheme does not require clients to use a uniform quantization level; clients can independently choose their quantization level according to their own bandwidth conditions without affecting the security and correctness of aggregation.
All Facts (4)
Existing heterogeneous-quantization secure aggregation methods leak partial aggregations of model updates, and under certain dropout and client-server collusion patterns may leak fragments of individual model updates.
50%UnverifiedThe paper proposes an information-theoretically secure aggregation scheme that tolerates client dropouts and collusion between up to T semi-honest clients and the server, while supporting heterogeneous client quantization.
50%UnverifiedCompared with existing heterogeneous-quantization secure aggregation methods, this scheme maintains information-theoretic security under all permitted dropout and collusion patterns.
50%UnverifiedThe scheme does not require clients to use a uniform quantization level; clients can independently choose their quantization level according to their own bandwidth conditions without affecting the security and correctness of aggregation.
50%