Hi FedML team, #2262
Unanswered
rwilliamspbg-ops
asked this question in
Q&A
Hi FedML team,
#2262
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
I built Sovereign Mohawk, a formally verified hierarchical FL runtime that scales to 10 M nodes with 55.5% Byzantine tolerance...
Hi Flower team,
I built Sovereign Mohawk, a formally verified hierarchical FL runtime that scales to 10 M nodes with 55.5% Byzantine tolerance (far beyond standard Krum or median-based defenses).
It uses a 4-tier tree (edge → regional → continental → global) + zk-SNARK verifiable aggregation + O(d log n) streaming gradients.
Formal proofs (six theorems):
[[ACADEMIC_PAPER.md] (https://github.com/rwilliamspbg-ops/Sovereign-Mohawk-Proto/blob/main/ACADEMIC_PAPER.md)
+
[proofs/ folder] (https://github.com/rwilliamspbg-ops/Sovereign-Mohawk-Proto/tree/main/proofs)
Key theorems excerpt:
Sovereign Mohawk is formally verified across six theorems (full details in ACADEMIC_PAPER.md + proofs/):
Theorem 1 (BFT): (∑ f_t)-Byzantine resilient if f_t < n_t/2 per tier → 55.5% global tolerance (5.5 M malicious nodes out of 10 M) via hierarchical Multi-Krum.
Theorem 2 (Privacy): Sequential RDP composition → strict ε=2.0 budget across entire population.
Theorem 3 (Comms): O(d log n) complexity (40 TB → 28 MB at 10 M nodes).
Theorem 4 (Stragglers): 99.99% success probability with 10× redundancy.
Theorem 5 (zk-SNARKs): 200-byte proofs, O(1) verification in ~10 ms.
Theorem 6 (Non-IID): Convergence rate O(1/√(KT)) + O(ζ²) under heterogeneity bound ζ².
Test suites & numbers:
Test suites (all public, Apache 2.0):
make test && make verifymake test-python-sdk(includes proof verification benchmarks)./genesis-launch.sh→ spins up 200-node local testnet + Grafana in <60 sKey results (March 2026):
Try it in one command (no account needed):
Beta Was this translation helpful? Give feedback.
All reactions