Secure aggregation scheme supports heterogeneous client quantization in federated learning
Original title: Information-Theoretic Secure Aggregation With Heterogeneous Client Quantization
It avoids leaking partial aggregates when clients quantize differently and up to T semi-honest clients collude with the server.
The paper proposes an information-theoretic secure aggregation protocol for federated learning when clients use different quantization levels. Existing heterogeneous schemes can reveal partial aggregates and, under some dropout and collusion patterns, segments of individual updates. The new protocol uses (T+1, N) threshold secret sharing over identical linear integer systems even when shares sit in different algebraic groups, and it tolerates client dropouts.
- The server should learn only the sum of N clients' model updates.
- Shares come from identical linear integer systems despite distinct groups.
- The threat model allows up to T semi-honest clients colluding with the server.
From the original article
1 min read at arxiv.orgIn federated learning (FL), secure aggregation enables a server to compute the sum of local model updates of $N$ clients, without revealing any information on the individual model updates to the server, beyond their sum. Different clients may be subject to heterogeneous bandwidth constraints, and consequently may choose different numbers of quantization levels to represent the information they send.
By Shreya Meel · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.