Secure-FL round: from local samples to a proven global model

Secure-FL round: from local samples to a proven global model A data-flow diagram generated by Archify. 01 / Local train 02 / Commit 03 / Prove 04 / Verify 05 / Aggregate Client shard · local samples · 01 / Local train · never leaves client Client shard local samples never leaves client Local Training · PyTorch SGD · 01 / Local train · Python Local Training PyTorch SGD Python State Hash · SHA-256 full state · 02 / Commit · initial + updated State Hash SHA-256 full state initial + updated Delta + Norm · learnable-only L2 · 02 / Commit · BN buffers excluded Delta + Norm learnable-only L2 BN buffers excluded Circuit Inputs · 200 sampled params · 03 / Prove · fixed point Circuit Inputs 200 sampled params fixed point Client Prover · Groth16 delta bound · 03 / Prove · transferable Client Prover Groth16 delta bound transferable Update Verifier · hash · norm · Groth16 · 04 / Verify · drop on mismatch Update Verifier hash · norm · Groth16 drop on mismatch FedJSCM · momentum + lr · 05 / Aggregate · Rust FedJSCM momentum + lr Rust Server Prover · aggregation circuit · 05 / Aggregate · self-checked Server Prover aggregation circuit self-checked Global Model · round r+1 params · 05 / Aggregate · distributed Global Model round r+1 params distributed mini-batches state dict Δ full state dict learnable delta witness inputs commitments proof JSON verified updates deltas, weights, momentum proof bundle new params next-round model Legend primary data policy / PII async batch data store data flow